Applied Non-Linear Dynamical Systems PDF

Springer Proceedings in Mathematics & Statistics
Jan Awrejcewicz Editor
Applied
Non-Linear
Dynamical
Systems
Volume 93
More information about this series at http://www.springer.com/series/10533

This book series features volumes composed of select contributions from workshops
and conferences in all areas of current research in mathematics and statistics,
including OR and optimization. In addition to an overall evaluation of the interest,
scientific quality, and timeliness of each proposal at the hands of the publisher,
individual contributions are all refereed to the high quality standards of leading
journals in the field. Thus, this series provides the research community with
well-edited, authoritative reports on developments in the most exciting areas of
mathematical and statistical research today.
Jan Awrejcewicz
Editor
Applied Non-Linear
Dynamical Systems
123
Editor
Jan Awrejcewicz
Department of Automation, Biomechanics
and Mechatronics
Łódź University of Technology
Łódź, Poland
ISSN 2194-1009 ISSN 2194-1017 (electronic)

ISBN 978-3-319-08265-3 ISBN 978-3-319-08266-0 (eBook)
DOI 10.1007/978-3-319-08266-0
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014951679
Mathematics Subject Classification (2010): 34-xx, 34A12, 34A26, 34A34, 34A35, 34A36, 34B15,
34C15, 34C23, 34C25, 34C29, 34Dxx, 34D06, 34D08, 34D10, 34D20, 24D23, 34E13, 34F10, 34F15,
34H10,37-xx, 37Bxx, 37B55, 37C27, 37C70, 37C75, 37D45, 37Gxx, 37G15, 37H20, 37Jxx, 37J20,
37J25, 37J55,70-xx, 70Exx, 70Hxx, 70Jxx, 70Kxx, 74-xx, 74Gxx, 74Jxx
© Springer International Publishing Switzerland 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
The 12th International Conference “Dynamical Systems-Theory and Applications”

(DSTA) was held on December 2–5, 2013, in Łódź, Poland. Organized by the
staff of the Department of Automation, Biomechanics and Mechatronics of the
Lodz University of Technology, the aim of the conference was to discuss and
illustrate the present state and perspective for modeling, simulation, and control
of nonlinear dynamical systems—a rapidly developing research front that includes
various disciplines in science, bioscience, and high technology.
The Scientific Committee of the conference include the following researchers:
Marcilio Alves (Brazil), Igor V. Andrianov (Ukraine), José M. Balthazar (Brazil),
Wojciech Blajer (Poland), Tadeusz Burczyński (Poland), Czesław Cempel (Poland),
Simona-Mariana Cretu (Romania), Virgil-Florin Duma (Romania), Horst Ecker
(Austria), Michal Fečkan (Slovakia), Barry Gallacher (UK), Józef Giergiel (Poland),
Peter Hagedorn (Germany), Katica Hedrih (Serbia), Ivana Kovacić (Serbia), Janusz
Kowal (Poland), Jan Kozanek (Czech Republic), Vadim A. Krysko (Russia),
Lidiya V. Kurpa (Ukraine), Claude-Henri Lamarque (France), Gennady A. Leonov
(Russia), Nuno M.M. Maia (Portugal), Leonid I. Manevitch (Russia), Yuriy Mikhlin
(Ukraine), Gerard Olivar (Colombia), Carla M.A. Pinto (Portugal), Bogdan Posi-
adała (Poland), Stanisław Radkowski (Poland), Bogusław Radziszewski (Poland),
Giuseppe Rega (Italy), Christos H. Skiadas (Greece), Alexander Seyranian (Russia),
Gábor Stépán (Hungary), Jerzy Świder (Poland), Andrzej Tylikowski (Poland),
Tadeusz Uhl (Poland), Ferdinand Verhulst (The Netherlands), Jerzy Warmiński
(Poland), Edmund Wittbrodt (Poland), Józef Wojnarowski (Poland), Ludmila V.
Yakushevich (Russia), Hamad M. Yehia (Egypt), Mikhail V. Zakrzhevsky, and
Klaus Zimmermann (Germany).
In the following 38 chapters we present only a small sample of different
approaches and understandings of dynamical systems and their applications in
physics, mechanics, automation, biomechanics, and applied mathematics.
In Chap. 1 Kovacic and Rand consider nonlinear oscillators with period indepen-
dent of amplitude and with Duffing-type restoring force. They present Lagrangians,
v
vi Preface
conservation laws, equations of motions, and solutions of the governing equations

for oscillators with hardening and softening cubic-type nonlinearities.
Pilipchuk in Chap. 2 presents physical insight and methodologies of asymptotics
of “rigid-body” motions for nonlinear dynamics. He shows that by tracking rigid
Euclidean transformation nonlinear models can be revealed.
In Chap. 3 Krysko et al. use the Bubnov–Galerkin method and the finite
difference method for studies of vibrations of flexible cylindrical and sector shells
subjected to the action of uniformly distributed loads. A few novel nonlinear
phenomena have been reported.
Bhem and Schwebke analyze in Chap. 4 the wormlike locomotion system. They
focused on gear shift patterns and presented procedure for their adjustments to
optimize speed and gait/crawling to predefined limits of actuator or spike force load.
Awrejcewicz et al. illustrate and discuss (in Chap. 5) periodic and chaotic
dynamics of plates and shells as well as a weak turbulent behavior exhibited by
these solid structures, while modeling them as 2D infinite objects. They present also
novel approaches to obtain reliable results of nonlinear differential equations, as
well as new methods of chaos monitoring.
In Chap. 6 Morcillo et al. propose a methodology of using bifurcation diagrams
for computation chaos controllers. They apply this to PWM-controlled power
converters method based on an adaptive control, where the sawtooth signal is
redefined as a function of the output and reference voltages.
In Chap. 7 Andrianov et al. present studies of anti-plane shear waves. They focus
on wave propagation through a cylindrical structured cancellous bone, applying
model of a two-dimensional mesh of elastic trabeculae filled by a viscous marrow.
Kizilova et al. focused in Chap. 8 on dynamics of postural sway in human.
Analyzing body sway patterns for the group of young healthy individuals and two
groups of patients with pathologies of spine and joints, they observed quasi-regular
and chaotic dynamics with certain asymmetry of the body acceleration, respectively.
In Chap. 9 Adamiec-Wójcik et al. present modeling of the slender system by
means of the rigid finite element method. They considered large deformations and
friction using rope as the model. Presented and analyzed are also dynamics of an
offshore crane lifting a load from a vessel.
Syta and Litak in Chap. 10 presented results of the investigations of Van der
Pol–Duffing system. They focused on the dynamical response of the system with
an external harmonic excitation and a memory of a fractional characteristic. The
obtained results indicate occurrence of bifurcations of quasiperiodic solutions with
short intervals of chaotic solutions.
Analysis of fluid flow around the cylinder is described by Akhmetov and Kutluev
in Chap. 11. They apply asymptotic methods for problem of the vortex structure
appearance in a stationary viscous incompressible fluid and investigate properties
of the flow function in boundary layer.
In Chap. 12 Awrejcewicz et al. present results of application of the asymptotical
approach in the form of limiting phase trajectories and multiple timescale methods
for analysis of dynamical problem of a two-degree-of-freedom mechanical sys-
tem. Application of those approaches in the case of spring pendulum allows for
Preface vii
determination of critical values of the parameters responsible for change of the

character of vibrations.
Kaliński et al. present (Chap. 13) an idea of the workpiece holder with adjustable
stiffness to be applied in milling of flexible details. This holder, together with
vibration surveillance system, allows for more efficient milling using slender ball-
end tools without interference of vibrations.
In Chap. 14 Blajer presents problem of servo-constraint in the inverse simulation
problem for underactuated mechanical system. Simulation results are discussed and
confronted with computational issues of the governing equations in the form of
ordinary differential equations and differential algebraic equations.
Global analysis of nonlinear dynamics of simple hybrid electronic systems
through application of the method of complete bifurcation groups is presented by
Pikulin in Chap. 15. He also proves that it is possible to design reliable switching
power convertor. The applied method of complete bifurcation groups allows to
predict and avoid occurrence of undesirable regimes in operation.
Udwadia and Mylapilli in Chap. 16 discuss interrelations and connections
between tracking control of nonlinear systems and constrained motion of mechan-
ical systems. By providing diverse examples, they illustrate how ease, simplicity,
and efficiency in control can be achieved. They also show that closed-form forces
obtained are optimal and minimize the control cost at each instant of time still
providing exact trajectory tracking.
In Chap. 17 Ruchkin presents new method of the intellectual investigations of the
nonlinear dynamical system. By application of this method the system with a special
Hamiltonian structure is studied and its regular and chaotic behavior is analyzed.
Lateral dynamics behavior of the two-axle freight wagon with the UIC double-
link suspension in dependence on chosen parameters is shown in Chap. 18 by
Matej et al. Based on the Coulomb law regarding friction, applied are the non-
smooth mechanics, which allows for derivation of the mathematical models with
and without lateral bump-stop.
Zimmerman et al. focus on kinematics and dynamics of a mechanical system
with mecanum wheels (Chap. 19). They proceed with comparison of nonholonomic
model and approximate model used in robotics, obtaining similar results, which
resulted in the production of prototype of a mobile robot with four mecanum wheels.
Chapter 20 by Ritto and Sampaio is devoted to reliability analysis of horizontal
drill-string dynamics. It focuses on reliability of the operation, defined as the prob-
ability of not achieving a target efficiency and measured by the mean input/output
ratio.
Hedrih presents in Chap. 21 changes of elasticity 3D matrix surrounding
mammalian oocyte. She considers changes occurring in zona pellucida during
maturation and fertilization processes. Applying an oscillatory spherical net model
of mouse, eigen circular frequencies of mouse oocyte and mouse embryo have been
estimated.
Problem of mass points interacting gravitationally is presented by Szumiński
and Przybylska in Chap. 22. They illustrate complicated behavior of trajectories
of system with applied certain holonomic constraints using Poincarè cross sections.
viii Preface
Verhulst applies in Chap. 23 slow–fast timescales in the framework of Fenichel

geometric singular perturbation theory for analysis of equations with periodic
coefficients for singularly perturbed growth.
In Chap. 24 basing on fractional calculus and the concept of fractals Abramov
proposes a model of nonlinear fractal oscillator. Investigated are stress field for
different structural states of structural dislocation in a nanosystem, as well as its
deformation and behavior of energy spectrum.
Dynamics in diagnostic expert systems are addressed by Cholewa in Chap. 25.
He introduces dynamic statement network, where statements consist of contents and
values. After detailed comparison of networks, he proposes manner of transforma-
tion from a static network into dynamic one.
In Chap. 26 Palej proposes method for solving the nonstandard two-point
boundary value problem. He considers the case, where the number of boundary
conditions is higher than the number of first-order ordinary differential equations
containing certain number of unknown parameters and where difference between
standard and nonstandard boundary value problem consists in the size of the initial
value problem that needs to be solved.
Lainscsek et al. in Chap. 27 propose a differential equation with time delay for
automatic sleep scoring from single electrode. Assuming that there is a parameter,
where at least one of coefficients varies depending on sleep stages, it is possible to
construct hypnogram. In this case brain activity is considered as resulting from a
dynamical system.
Chapter 28 by Gyebrószki and Csernák is focused on numerical methods for
quick analysis of micro-chaos. They describe and compare different methods used
for characterizing chaotic behavior and compare them with simple cell mapping
for investigations of chaotic behavior for the case of digitally controlled inverted
pendulum.
Studies of a bouncing ball impacting with a periodically moving limiter are
presented in Chap. 29 by Okniński and Radziszewski. They analyze this problem
within two defined frameworks of the table motion for four cubic polynomial and
sinusoidal motion models.
In Chap. 30 Pascal and Stepanov present results of investigations of the behavior
of the strongly nonlinear vibrating system excited by dry friction and harmonic
force. They consider as the model a system composed of two masses connected by
linear springs, where one of the masses is in contact with a driving belt moving at a
constant velocity and friction force with Coulomb’s characteristics acting between
mass and belt.
Selyutskiy in Chap. 31 focuses on studies of aerodynamic pendulum dynamics
in low-speed airflow. He applies as an investigation model a phenomenological
mode with the internal dynamics of the flow simulated by an oscillator attached
to the pendulum and obtained results of simulations are in good correlation with
experimental results.
Studies of nonlinear interactions of two coupled oscillators at different timescales
are presented by Ture Savadkoohi and Lamarque in Chap. 32. Analytical develop-
ments of those investigations are compared with numerical results and discussed
Preface ix
is the possibility of the passive control of the main system by means of the time-
dependent nonlinear energy sink.
In Chap. 33 Szklarz and Jarz˛ebowska discuss a systematic coordinate-free
approach for the formulation of the no-slip condition for wheeled robot models that
can be used for derivation of the nonholonomic constraint equations. This approach
yields models that are unified, verifiable, comparable, and repeatable.
Original unitary theory of the optical choppers with rotating wheels working
with top-hat laser beams is discussed in Chap. 34 by Circa and Duma. They present
program developed for designing of the optical choppers with rotating wheels and
verified it by application to classical and eclipse choppers.
In Chap. 35 modification of the mathematical model of the HIV dynamics in
HIV-specific helper cells is presented by Pinto and Carvalho. Considered are two
types of the delay—a latent period for the interval for cells, first one with contact
with the virus, to be infected by the virons and released by them and the second one
is a viron production period for the virons to be infected by virons and released to
the blood stream from the infected cells.
Gallacher et al. discuss in Chap. 36 application of parametric excitation and
amplification for MEMS ring gyroscope. Theoretically proved is that the parametric
excitation offers suitable excitation method for the rate integrating gyroscope.
In Chap. 37 Piccirillo et al. present analysis of the influence of external
parameters on the oscillator dynamics. They investigate two smart materials, i.e.,
shape memory alloy and magneto rheological damper as well as effect of system
dissipation energy related to their hysteretic behavior.
Generalized method of studying plane topographical Poincarè systems to higher
dimensions is presented by Shamolin in Chap. 38. He shows also elaborated meth-
ods for qualitative study of dissipative systems and systems with anti-dissipation
yielding a possibility of obtaining conditions for bifurcation of birth of stable and
unstable auto-oscillations.
As can be noticed by the variability of topics of the chapters, dynamical systems
analysis can be found in very wide range of scientific disciplines and their constant
development is unavoidable and necessary from both theoretical and practical point
of view.
I do hope that the readers of this book will be attracted by the chosen topics. I
greatly appreciate help of Springer Editor Dr. Eve Mayer in publishing the presented
chapters recommended by the Scientific Committee of the DSTA 2013 after the
standard review procedure. Finally, I would like to thank all referees for their help
in reviewing the manuscripts.
Łódź and Warsaw, Poland Jan Awrejcewicz

Contents
Duffing-Type Oscillators with Amplitude-Independent Period . . . . . . . . . . . . . 1

Ivana N. Kovacic and Richard H. Rand
Asymptotic of “Rigid-Body” Motions for Nonlinear Dynamics:
Physical Insight and Methodologies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Valery Pilipchuk
Non-linear Phenomena Exhibited by Flexible Cylindrical
and Sector Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
V.A. Krysko, J. Awrejcewicz, I.V. Papkova, V.B. Baiburin,
and T.V. Yakovleva
Gear Shift Patterns in Uncertain Terrestrial Locomotion Systems. . . . . . . . . 37
Carsten Behn and Silvan Schwebke
Turbulent Phenomena in Flexible Plates and Shells . . . . . . . . . . . . . . . . . . . . . . . . . 49
J. Awrejcewicz, A.V. Krysko, V.A. Krysko, E.Yu. Krylova,
S.A. Mitskievich, I.V. Papkova, T.V. Yakovleva, V.M. Zakharov,
and V. Dobriyan
Using Bifurcation Diagrams for Controlling Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Daniel Morcillo, Daniel Burbano, Fabiola Angulo,
and Gerard Olivar
Shear Waves Dispersion in Cylindrically Structured
Cancellous Viscoelastic Bones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
I.V. Andrianov, V.V. Danishevs’kyy, and J. Awrejcewicz
Quasi-regular and Chaotic Dynamics of Postural
Sway in Human. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Natalya Kizilova, Elena Karpinska, and Michael Karpinsky
xi
xii Contents
Modelling of Ropes with Consideration of Large Deformations

and Friction by Means of the Rigid Finite Element Method . . . . . . . . . . . . . . . . 115
Iwona Adamiec–Wójcik, Jan Awrejcewicz, Lucyna Brzozowska,
and Łukasz Drag
˛
Dynamical Response of a Van der Pol–Duffing System with an
External Harmonic Excitation and Fractional Derivative . . . . . . . . . . . . . . . . . . . 139
Arkadiusz Syta and Grzegorz Litak
Vortex Structure Around the Cylinder at a Flow of Viscous Fluid . . . . . . . . . 151
Rustyam G. Akhmetov and Ruslan R. Kutluev
Asymptotic Analysis and Limiting Phase Trajectories
in the Dynamics of Spring Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Jan Awrejcewicz, Roman Starosta,
and Grażyna Sypniewska-Kamińska
Vibration Surveillance System with Variable Stiffness Holder
for Milling Flexible Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Krzysztof J. Kaliński, Marek Chodnicki, Michał R. Mazur,
and Marek A. Galewski
Diversities in the Inverse Dynamics Problem for Underactuated
Mechanical Systems Subject to Servo-constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Wojciech Blajer
Rare Phenomena and Chaos in Switching Power Converters. . . . . . . . . . . . . . . 203
Dmitry Pikulin
Constrained Motion of Mechanical Systems and Tracking
Control of Nonlinear Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Firdaus E. Udwadia and Harshavardhan Mylapilli
The General Conception of the Intellectual Investigation
of the Regular and Chaotic Behavior of the Dynamical System
Hamiltonian Structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Constantin Ruchkin
Dynamic Properties of Two-Axle Freight Wagon with UIC
Double-Link Suspension as a Non-smooth System with Dry Friction . . . . . 255
Jan Matej, Jarosław Seńko, and Jan Awrejcewicz
Dynamics of Mechanical Systems with Mecanum Wheels . . . . . . . . . . . . . . . . . . 269
Klaus Zimmermann, Igor Zeidis, and Mohamed Abdelrahman
Reliability Analysis of the Dynamics of a Horizontal Drill-String . . . . . . . . . . 281
T.G. Ritto and Rubens Sampaio
Transition in Oscillatory Behavior in Mouse Oocyte and Mouse
Embryo Through Oscillatory Spherical Net Model of Mouse
Zona Pellucida . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Andjelka N. Hedrih
Contents xiii
Constrained n-Body Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

Wojciech Szumiński and Maria Przybylska
Hunting French Ducks in Population Dynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Ferdinand Verhulst
Model of Nonlinear Fractal Oscillator in Nanosystem . . . . . . . . . . . . . . . . . . . . . . . 337
Valeriy S. Abramov
Dynamical Statement Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Wojciech Cholewa
The Shooting Method for Non-standard Boundary Value Problem . . . . . . . . 363
Rafał Palej
Automatic Sleep Scoring from a Single Electrode Using Delay
Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Claudia Lainscsek, Valérie Messager, Adriana Portman,
Jean-François Muir, Terrence J. Sejnowski, and Christophe
Letellier
Methods for the Quick Analysis of Micro-chaos. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Gergely Gyebrószki and Gábor Csernák
Bouncing Ball Dynamics: Simple Motion of the Table
Approximating the Sinusoidal Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Andrzej Okniński and Bogusław Radziszewski
Periodic Motions of Coupled Oscillators Excited by Dry
Friction and Harmonic Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
Madeleine Pascal and Sergey Stepanov
Limit Cycle Oscillations of an Aerodynamic Pendulum . . . . . . . . . . . . . . . . . . . . . 415
Yury D. Selyutskiy
Vibratory Energy Localization by Non-smooth Energy Sink
with Time-Varying Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Alireza Ture Savadkoohi and Claude-Henri Lamarque
Coordinate-Free Formulation of Nonholonomic Constraints
for Wheeled Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
Sanjuan Szklarz Paweł Cesar and Elżbieta Jarzȩbowska
Optomechatronic Choppers with Rotating Elements:
Design Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
O. Cira and V.-F. Duma
A Delay Mathematical Model for HIV Dynamics
in HIV-Specific Helper Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
Carla M.A. Pinto and Ana R.M. Carvalho
xiv Contents
The Application of Parametric Excitation in Resonant MEMS

Gyroscopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Barry J. Gallacher, Zhongxu Hu, Kiran M. Harish,
Stephen Bowles, and Harry Grigg
Influence of Smart Material on the Dynamical Response
of Mechanical Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
Vinícius Piccirillo, Ângelo Marcelo Tusset,
José Manoel Balthazar, Davide Bernardini, and Guiseppe Rega
Dynamical Pendulum-Like Nonconservative Systems . . . . . . . . . . . . . . . . . . . . . . . 503
Maxim V. Shamolin
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
Duffing-Type Oscillators
with Amplitude-Independent Period
Ivana N. Kovacic and Richard H. Rand
Abstract Nonlinear oscillators with hardening and softening cubic Duffing

nonlinearity are considered. Such classical conservative oscillators are known
to have an amplitude-dependent period. In this work, we design oscillators with
the Duffing-type restoring force but an amplitude-independent period. We present
their Lagrangians, equations of motion, conservation laws, as well as solutions for
motion.
1 Introduction
Classical Duffing oscillators are governed by
xR C x ˙ x 3 D 0: (1)
Their restoring force F D x ˙ x 3 includes a linear geometric term as well as a

cubic geometric term: a positive sign in front of the cubic term corresponds to a
hardening Duffing oscillator (HDO) and the negative one to a softening Duffing
oscillator (SDO) [4, 6]. Unlike the majority of nonlinear oscillators, both of them
have a closed-form exact solution, which is expressed in terms of Jacobi cn or sn
elliptic functions. These solutions corresponding to the following initial conditions
x.0/ D A; xP .0/ D 0; (2)
I.N. Kovacic ()

Department of Mechanics, Faculty of Technical Sciences, University of Novi Sad,
21125 Novi Sad, Serbia
e-mail: ivanakov@uns.ac.rs
R.H. Rand
Department of Mathematics, Cornell University, Ithaca, NY 14853, USA
Department of Mechanical and Aerospace Engineering, Cornell University, Ithaca,
NY 14853, USA
e-mail: rhr2@cornell.edu
© Springer International Publishing Switzerland 2014 1

J. Awrejcewicz (ed.), Applied Non-Linear Dynamical Systems, Springer Proceedings
in Mathematics & Statistics 93, DOI 10.1007/978-3-319-08266-0__1
2 I.N. Kovacic and R.H. Rand
Table 1 Solutions for motion of the HDO and SDO ( in case of the SDO one requires jAj < 1
for the closed orbits around the stable origin)
Solution for motion Frequency Elliptic modulus
2 A2
HDO: xHDO D Acn .!HDO t; kHDO / 2
!HDO D 1 C A2 2
kHDO D 2!A2 D 2.1CA2 /
HDO
A2 A2 2
SDO : xSDO D Asn .!SDO t; kSDO / 2
!SDO D1 2
2
kSDO D 2
2!SDO
D A
2

2 1 A2
T
10
8 SDO
2p
HDO
A
2 1 0 1 2
Fig. 1 Period of the HDO (blue solid line) and SDO (red dashed line) as a function of amplitude A
are given in Table 1 together with their frequencies ! and elliptic moduli k (see, for
example, [4, 6] for more detail).
Given the fact that the period of cn and sn is T D 4K .k/ =!, where K .k/
stands for the complete elliptic integral of the first kind, the following expressions
are obtained for the period of the HDO and SDO:
!
r r
2
A2 4K A
4K 2.1CA2 / 2 1 A2
2
THDO D p ; TSDO D q : (3)

1 C A2 1 A2
2
As seen from these expressions and Fig. 1, the period of the HDO and SDO
depends on the amplitude A. This leads us to the question of designing an oscillator
having the Duffing-type restoring force F D x ˙ x 3 , but a constant, amplitude-
independent period, corresponding to the so-called isochronous oscillators [1, 2].
In what follows we present such Duffing-type oscillators, both hardening and
softening, modelled by
xR C G .x; x/
P C x ˙ x 3 D 0; (4)
finding also the corresponding Lagrangians, conservation laws, as well as solutions

for motion.
Duffing-Type Oscillators with Amplitude-Independent Period 3
2 Derivation
The derivation of the mathematical model (4) is based on the transformation

approach [5], in which the kinetic energy Ek and potential energy Ep of nonlinear
oscillators are made equal to the one of the simple harmonic oscillator (SHO)
1 P2 1 2
EkSHO D X ; EpSHO D X ; (5)
2 2
which is known to have a constant, amplitude-independent period (note that its
generalized coordinate is labelled here by X ).
2.1 Case I
To find the first mathematical model of the form (4), we assume that Ep D Ep .x/
and let Ep EpSHO D X 2 =2, obtaining
p
XD 2Ep : (6)
Then, we also make the kinetic energy Ek of nonlinear oscillators equal to the one
of the SHO and use Eq. (6) to derive
2
XP 2 Ep0
Ek D D xP 2 ; (7)
2 4Ep
where Ep0 D dEp =dx. The differential equation of motion stemming from the
Lagrangian L D Ek Ep has a general form
!
Ep00 Ep0 2Ep
xR C xP 2 C D 0: (8)
Ep0 2Ep Ep0
The last term on the left-hand side 2Ep =Ep0 is required to correspond to the Duffing
restoring force F =x ˙ x 3 , which gives the potential energy
x2
Ep D : (9)
2 .1 ˙ x 2 /
Equation (7) now yields the kinetic energy
1
Ek D xP 2 ; (10)
2 .1 ˙ x 2 /3
so that Eq. (8) becomes
3x
xR xP 2 C x ˙ x 3 D 0: (11)
1 ˙ x2
Based on X 2 =2 C XP 2 =2 D const:, the system exhibits the first integral
1 x2 A2
xP 2 C D : (12)
.1 ˙ x 2 / 3 1 ˙ x2 1 ˙ A2
Taking the solution for motion of the SHO in the form X D a cos .t C ˛/, where
a and ˛ are constants, and using Eqs. (6), (9) and (2), the solution for motion of
Eq. (11) is obtained
A x
p cos t D p : (13)
1˙A2 1 ˙ x2
Numerical verifications of the analytical results for the motion (13) and phase
trajectories (12) are shown in Figs. 2 and 3 for the HDO and SDO, respectively.
These figures confirm that the analytical results coincide with the solutions obtained
by solving the equation of motion (11) numerically. In addition, they illustrate the
fact that the period stays constant despite the fact that the amplitude A changes, i.e.,
that the systems perform isochronous oscillations.
a b .
1.0 x
1.0
0.5
0.5
x 0.0
x
-1.0 -0.5 0.5 1.0
-0.5
-0.5
-1.0 2p 4p 6p
0 5 10 15 -1.0
t
Fig. 2 Isochronous oscillations of the HDO, Eq. (11) for A D 0:25; 0.5; 0.75: (a) time histories
obtained numerically from Eq. (11) (black dots) and from Eq. (13) (blue solid line); (b) phase
trajectories obtained numerically from Eq. (11) (black dots) and from Eq. (12) (blue solid line)
(upper signs are used in all these equations)
.
x
a b
1.0 1.0
0.5
0.5
x 0.0 x
-1.0 -0.5 0.5 1.0
-0.5 -0.5
-1.0 2p 4p 6p
-1.0
0 5 10 15
t
Fig. 3 Isochronous oscillations of the SDO, Eq. (11) for A D 0:25; 0.5; 0.75: (a) time histories
obtained numerically from Eq. (11) (black dots) and from Eq. (13) (red dashed line); (b) phase
trajectories obtained numerically from Eq. (11) (black dots) and from Eq. (12) (red dashed line)
(lower signs are used in all these equations)
2.2 Case II
In this case we consider the system whose potential and kinetic energies are
1 2 1 1 1 2
X D .x f /2 ; Ek D XP 2 D
Ep D xP f C x 2 f 0 ; (14)
2 2 2 2
Rt
where f f .I /, I D 0 x .t / dt , and f 0 D df =dI .
The corresponding Lagrange’s equation is
f 0 f 00 3
xR C 3x xP CxC x D 0: (15)
f f
This system has two independent first integrals. The first one is the energy
conservation law stemming from XP 2 =2 C X 2 =2 D const:
2
xP f C x 2 f 0 C .x f /2 D h1 ; h1 D const: (16)
The other firstRintegral is related to the principle of conservation of momentum for

the SHO XP C 0 Xdt D XP .0/ D const. By using X and XP from (14) and knowing
t
that dI =dt D x, we obtain

Z
0
xP f C x f C 2
f .I / dI D h2 ; h2 D const: (17)
In addition, as the solution for motion for the SHO can be written down as X D
a sin .t C ˛/, the following should be satisfied:
a sin .t C ˛/ D x f; a cos .t C ˛/ D xP f C x 2 f 0 : (18)
00
For the hardening-type nonlinearity in Eq. (15), one requires f =f D 1, which
leads to
Z t
fHDO D exp x .t / dt ; (19)
0
and the equation of motion takes the form
xR C 3x xP C x C x 3 D 0: (20)
Two first integrals (16) and (17) are

h 2 i
exp .2x1 / x3 C x22 C x22 D h1 ; (21)
and

exp .x1 / x3 C x22 C 1 D h2 ; (22)
where
Z t
x1 D x .t / dt; x2 D xP 1 D x; x3 D xP 2 D x;
P (23)
0
with initial conditions being [see Eq. (2)]
x1 .0/ D 0; x2 .0/ D xP 1 .0/ D A; x3 .0/ D xR 1 .0/ D 0: (24)
Equation (21) is plotted in Fig. 4a for h1 D 1. To analyze phase trajectories in more

detail, Eq. (22) is squared and divided by Eq. (21) to obtain
2
x3 C x22 C 1
2 D B; B D const: (25)
x3 C x22 C x22
This expression agrees with the first integral obtained and studied in [3] and is
plotted in Fig. 4b, where periodic solutions correspond to the case B > 1. Note
that for the initial conditions (24), one has B D 1 C 1=A2 , which implies that B is
always higher than unity.
x3
x1
x2
b
10
B=1.1
5 B=1.5
B=2
B=5
x3 0
-5 B=1
B=0.1
-10
-4 -2 0 2 4
x2
Fig. 4 (a) 3D plot of Eq. (21); (b) phase trajectories obtained from Eq. (25) for different
values of B
By using (18) and (19) one can derive
xP sin .t C ˛/ C x 2 sin .t C ˛/ x cos .t C ˛/ D 0: (26)
Its solution satisfying Eq. (2) is

sin t C arc tan A1
xDq : (27)
1 C A12 cos t C arc tan A1
1.0
0.5
x 0.0
-0.5
-1.0 2p 4p 6p
5 10 15
t
Fig. 5 Time response of the HDO, Eq. (20) for A D 0:25; 0.5; 0.75: numerically obtained solution
from Eq. (28) (black dots) and from Eq. (27) (blue solid line)
To compare the analytical solution for motion (27) with a numerically obtained
solution of the equation of motion (20), the latter is written down in the form
«1 C 3xP 1 xR 1 C xP 1 C xP 13 D 0;
x (28)
and numerically integrated by using the initial conditions (24). This comparison,
plotted in Fig. 5, shows that these two types of solution are in full agreement as well
as that the period is amplitude-independent.
The equation of motion (15) corresponds to the SDO if f 00 =f D 1, which is
satisfied for
Z t
fSDO D cos x .t / dt : (29)
0
This equation of motion is now given by

Z t
xR 3x xP tan x .t / dt C x x 3 D 0: (30)
0
By using the notation given in Eq. (23), the equation of motion (30) transforms to
«1 3xP 1 xR 1 tan x1 C xP 1 xP 13 D 0;
x (31)
with the initial conditions given in Eq. (24). Two first integrals (16) and (17) become
2
x3 cos x1 x22 sin x1 C x22 cos2 x1 D h1 ; (32)
and
x3 cos x1 x22 sin x1 C sin x1 D h2 : (33)
These two integrals can be manipulated to exclude x1 and to derive
.x22 1/2 .h1 C .1 h1 C h22 /x22 C x24 /2 C 2.1 h1 h22

C.1 C h1 h22 /x22 /.h1 .x22 1/ x22 .h22 C x22 1//x32
C..1 C h1 /2 2.1 C h1 /h22 C h42 /x34 D 0: (34)
For the initial conditions (24) one has h1 D A2 and h2 D 0. Introducing these
values into Eq. (34) and solving it with respect to x3 , the following explicit solution
for phase trajectories is obtained:
s
A2 x22
x3 D ˙.x22 1/ : (35)
1 A2
Combining equations in (18) and using a D A and ˛ D =2, we derive

p
xP A cos t x 2 x 2 A2 cos2 t C xA sin t D 0: (36)
Its solution satisfying Eq. (2) is

A cos t
xDp : (37)
1 A2 sin2 t
This solution is plotted in Fig. 6 together with the numerical solution of Eq. (31)
with Eq. (24) for different values of A. These solutions coincide and confirm
isochronicity.
1.0
0.5
x 0.0
-0.5
2p 4p 6p
-1.0
5 10 15
t
Fig. 6 Time response of the SDO, Eq. (30) with A D 0:25; 0.5; 0.75: numerically obtained
solution from Eq. (31) (black dots) and from Eq. (37) (red dashed line)
3 Conclusions
In this work we have considered nonlinear oscillators with a hardening and softening
Duffing restoring force. Unlike classical conservative Duffing oscillators, which
have an amplitude-dependent period, the designed Duffing-type oscillators have the
period that does not change with their amplitude and are, thus, isochronous. Two
separate cases are considered with respect to the form of their potential and kinetic
energy, which are made equal to the corresponding energies of the SHO, which is
known to be isochronous. Corresponding equations of motions are derived, as well
as their solutions for motion. Numerical verifications of these isochronous solutions
are provided. In addition, two independent first integrals are presented: the energy-
conservation law and the principle of conservation of momentum.
Acknowledgements Ivana Kovacic acknowledges support received from the Provincial Secre-
tariat for Science and Technological Development, Autonomous Province of Vojvodina (Project
No. 114-451-2094).
References
1. Calogero, F.: Isochronous Systems. Oxford University Press, Oxford (2008)

2. Calogero, F.: Isochronous dynamical systems. Phil. Trans. R. Soc. A 369, 1118–1136 (2011)
3. Iacono,R., Russo, F.: Class of solvable nonlinear oscillators with isochronous orbits. Phys. Rev.
E 83, 027601 (4 pages) (2011)
4. Kovacic, I., Brennan, M.J.: The Duffing Equation: Nonlinear Oscillators and their Behaviour.
Wiley, Chichester (2011)
5. Kovacic, I., Rand, R.: About a class of nonlinear oscillators with amplitude-independent
frequency. Nonlinear Dyn. 74, 455–465 (2013)
6. Rand, R.H.: Lecture notes on nonlinear vibrations (version 53). http://dspace.library.cornell.edu/
handle/1813/28989
Asymptotic of “Rigid-Body” Motions
for Nonlinear Dynamics: Physical Insight
and Methodologies
Valery Pilipchuk
Abstract The purpose of the present work is to show that an adequate basis
for understanding the essentially nonlinear phenomena must also be essentially
nonlinear but still simple enough to play the role of a basis. It is shown that such
types of “elementary” nonlinear models can be revealed by tracking the hidden links
between analytical tools of analyses and subgroups of the rigid-body motions or,
in other terms, rigid Euclidean transformation. While the subgroup of rotations is
linked with linear and weakly nonlinear vibrations, the translations with reflections
can be viewed as a geometrical core of the strongly nonlinear dynamics associated
with the so-called vibro-impact behaviors. It is shown that the corresponding
analytical approach develops through non-smooth temporal substitutions generated
by the impact models.
1 Introduction
This work is motivated by the intent to introduce a unified physical basis for analyz-
ing vibrations of essentially unharmonic, non-smooth, or maybe discontinuous time
shapes. Transitions to non-smooth limits can make investigations especially difficult
due to the fact that the dynamic methods were originally developed within the
paradigm of smooth motions based on the classical theory of differential equations.
From the physical standpoint, such way is natural for modeling the low-energy
motions. Although the impact dynamics has also quite a long prehistory, non-
smooth behaviors are often viewed as an exemption rather than a rule. Notice that
the classical theory of differential equations usually avoids non-differentiable and
discontinuous functions. Presently, however, many theoretical and applied areas
are dealing with the high-energy phenomena accompanied by strongly nonlinear
spatiotemporal behaviors making the classical smooth methods difficult to apply.
For instance, such phenomena are considered in engineering analyses of dynamical
systems under constraint conditions, friction-induced vibrations, structural damages
V. Pilipchuk ()
Wayne State, Detroit, MI, USA
e-mail: pilipchuk@wayne.edu

12 V. Pilipchuk
due to cracks, liquid sloshing impacts, etc. Similarly to the well-known analogy
between mechanical and electrical harmonic oscillators, the so-called Schmitt
trigger circuits generate non-smooth signals whose temporal shapes resemble the
mechanical vibro-impact processes. In many such cases, it is still possible to adapt
different smooth methods of the dynamic analyses through strongly nonlinear
algebraic manipulations with state vectors or by splitting the phase space into
multiple “smooth domains.” As a result, formulations are often reduced to the
discrete mappings in a wide range of the dynamics from periodic to stochastic.
It will be shown that a complementary analytical tool can be built on generating
models developing essentially unharmonic behaviors as their inherent properties.
For instance, the methodology presented in this work employs elementary impact
systems as a physical basis for describing different types of unharmonic processes.
This is implemented through the non-smooth time substitutions introduced origi-
nally for strongly nonlinear but smooth models [5]. Besides, as was shown in [4],
such methodology reveals explicit links between the impact dynamics and the
algebra of hyperbolic numbers analogously to the link between harmonic vibrations
and conventional (elliptic) complex analyses.
2 Physical and Mathematical Principles
2.1 Linear and Elementary Nonlinear Phenomena
Although the notion of linearity is quite clear in terms of mathematical formulations,

the attempt to directly associate the mathematical definition with physical phenom-
ena faces ambiguities due to the fact that the differential equations of motion for
the same model may appear to be either linear or nonlinear if switching between
different types of coordinates. For instance, a mechanical system, which is linear
in Cartesian coordinates, becomes nonlinear in polar coordinates. Nevertheless,
recognizing the unique role of Cartesian coordinates, it was suggested to define
a mechanical system as linear if the corresponding differential equations of motion
are linear in Cartesian coordinates.1 The purpose of this subsection is to determine
the most elementary dynamic phenomena that can already be qualified as essentially
nonlinear. For that reason, it is convenient to consider the linear situation first, for
instance, on a typical mass-spring model; see Fig. 1a, where all the springs are
assumed to be linearly elastic. The corresponding position vector is represented in
the following complex form:
q D A1 exp .i !1 t C '1 / C A2 exp .i !2 t C '2 / (1)
where Ak are constant complex vectors and ! k and ®k (k D 1, 2) are the modal
frequencies and initial phases.
1
This definition was suggested by V. Zhuravlev (private communication, Moscow, 1989).
Asymptotic of “Rigid-Body” Motions for Nonlinear Dynamics: Physical. . . 13
a b
q = {u1,u2} w2
w1
u1 u2 A1 A2
c d
reflection translation
t=a t
perfectly stiff plane (mirror)
Fig. 1 Vibration of the linearly elastic system (a) is represented by the rigid-body rotations (b);
the impact oscillator; (c) is associated with the rigid-body translation and reflection (d)
The fragment (b) in Fig. 1 shows another mechanical system whose dynamics is
described by the same expression (1), if the system’s position is identified by the
two points fixed at the edges of the discs. Therefore, the one-dimensional dynamics
generated by linearly elastic restoring forces can be represented by the rigid-body
rotations. In other words, the linearly elastic forces are effectively eliminated by
increasing the dimension of space. To some extent, such simple observation provides
an explanation why the sine and cosine waves possess so convenient mathematical
properties. Namely these harmonic functions represent the subgroup of elementary
rigid-body motions, such as rotation. Furthermore, the link between the two models,
(a) and (b), in Fig. 1, enables one to associate the linearity with the subgroup of
rotations.
Apparently, the rotations do not cover all the rigid-body motions – rigid
Euclidean transformations. These include also translations and reflections. The
translation itself seems to have a little physical content. However, combinations
of translation and reflection appear to be more interesting as shown in Fig. 1c, d
through the corresponding mechanical representations. In particular, the fragment
(c) shows a typical impact oscillator, whereas the fragment (d) illustrates a single
reflection case, which can be viewed as the most elementary nonlinear phenomenon.
2.2 Impact Models and Non-smooth Temporal Substitutions
Based on the class of linear generating models, whose typical representative is

shown in Fig. 1a, the quasi-harmonic methods for nonlinear vibration theory are
14 V. Pilipchuk
Mechanical model Basic function Generated

Temporal
substitution
1 t=a v =1
s =| t - a | t = a + s s
s = 1
2
s(t)
2 1 1
v =1 2 æ pt ö t = 1 + (t - 1)t
t = arcsin sin ç ÷
τ(t) p è 2 ø t 2 = 1
3 v =1 s ( t; d ) ¥
x
1
t= å (t i
+ s i ) si
= ( d + | t | - | t - d |) i=0
x =0 s(t;d ) x =1 2 si2 = 1
Fig. 2 Different versions of non-smooth temporal substitutions generated by the impact models
well developed and widely used. In contrast, using the models associated with
the translation-reflection, as those shown in Fig. 1c, d, is less common. Despite
of being geometrically simple, such models are not described within the classical
theory of differential equations due to the presence of discontinuities in the dynamic
states. This fact essentially complicates any direct use of the impact models as
a basis for building asymptotic or iterative procedures. It is shown, however, that
appropriate preliminary adaptations of the differential equations of motion can be
conducted through the non-smooth time substitutions listed in the third column
of Fig. 2. In particular, the first row explains how such type of substitutions
is introduced. Namely, the basic function s D jt aj is the model’s coordinate,
which can be interpreted as an eigentime of the system provided that nothing
else happens except the translation and reflection under consideration. The goal
is to introduce a new temporal argument, say s, in order to obtain the differential
equation of motion for the model, which is shown on the left of first row in Fig. 2.
Apparently, the substitution t ! s cannot work directly since no inversion t D t(s)
does exist on the entire time domain. Nevertheless, the following generalization
holds t D t .s; sP /; see the first row on its right in Fig. 2 for details. In particular, this
inversion appears to have the specific algebraic structure of hyperbolic numbers
with the basis f1; sP g. In contrast to the conventional elliptic complex numbers,
the unipotent sP , if squared, gives the positive sign, namely, sP 2 D 1. Interestingly
enough, the hyperbolic numbers have been known for about one and a half century
as abstract algebraic elements with no relation to the nonlinear dynamics or non-
Fig. 3 The “oscillating time” x

£ changes its direction
whenever a system makes a A
U-turn, while the original
physical time t runs to infinity
x
-a a t
a t t
-A
smooth functions (http://en.wikipedia.org/wiki/Split-complex_number). Moreover,

such numbers have been reintroduced multiple times under different names; see [1]
for an explanation of this fact and further details. In the present case, the hyperbolic
structures are generated by the non-smooth temporal arguments of impact systems,
while the unipotent sP possesses a clear physical interpretation as the normalized
velocity of impact system. Different applications to the vibration problems are based
on the periodic version illustrated by the second row of Fig. 2. In particular, the
equation on the right of the row means that during one period, the time argument is
expressed through the dynamic states of impact oscillator in the form of hyperbolic
number with the unipotent P [4, 5]. This statement therefore applies to any periodic
process. Figure 3 provides a geometrical interpretation of the “oscillating time”
for the particular case of periodic process described by the even function x(t) with
respect to the quarter of period, t D a.
Finally, in the third row of Fig. 2, the original time argument is structurised to
match the one-dimensional dynamics of rigid-body chain of identical particles. The
continuous “global” time is associated with the propagation of linear momentum,
whereas a sequence of non-smooth “local” time arguments describes the behavior
of individual physical particles. Such an idea helps to incorporate the temporal
symmetries of the dynamics into the differential equations of motion in many other
cases of regular or irregular sequences of internal impacts or external pulses. Since
the local times are bounded, a wider range of analytical tools becomes applicable.
2.3 Different Asymptotic Approaches to the Vibration Theory
Figure 4 provides further illustration for logical links between two alternative
approaches to the vibration problems. Interestingly enough, the illustration is possi-
ble within the same one-degree-of-freedom model, which is shown at the first row
of Fig. 4. Note that the oscillators with power-form characteristics were considered
for quite a long time. For instance, Lyapunov obtained such oscillators while
investigating degenerated cases of the dynamic stability problems [2]. Besides,
he introduced a couple of special functions, cs and sn, in order to invert the
16 V. Pilipchuk
Harmonic limit Impact limit

1

x + x 2 n−1 = 0, n= 1 2 n−1
x + x = 0, n→∞
Harmonic oscillator Impact oscillator
2
Sine & cosine waves Triangular sine wave

Subgroup of rotations Translation-reflection
3 z = x + ix x = X (τ ) + Y( τ) τ
i2 = −1 τ 2 = 1
Elliptic complex numbers Hyperbolic complex numbers
4 yi Yτ
i
z
τ
ϕ
1 x 1 X
i− | x |H = 1
| z |= 1 | x |H = 0
| z |2 = x 2 + y 2 | x | 2H = X 2 − Y 2
Elliptic complex plane Hyperbolic complex plane

∞ ∞
5 1 1 ( k)
∑ A cos kt + B
k k sin kt ∑ k !X
k=0
(k)
(0)τ k + Y (0)τ kτ
k!
k=0
Fourier series Saw-tooth power series
Fig. 4 Two alternative approaches to the vibration theory based on the harmonic and impact limits
corresponding quadratures. This relatively simple model nevertheless depicts the

gradual transition from linear to strongly nonlinear dynamics as the exponent n
runs from unity to infinity. Notably, all the temporal mode shapes of the oscillator
are described by the special functions cs and sn, except the two boundaries of the
interval 1 n < 1. Both boundaries represent simple asymptotic limits described
within the class of elementary functions.
Consider first the limit of harmonic oscillator (n D 1), generating the sine and
cosine waves; see the left of the second row. The widely known convenience of
using this couple of functions can be explained by their links to the elementary
rigid-body motions, namely the subgroup of rotations. The algebra of conventional
(elliptic) complex numbers with the corresponding complex plane representation
can be viewed as a next hierarchic step here due to the well-known Euler formula.
Finally, taking the linear combination of harmonic waves with different frequencies
and keeping in mind the idea of parameter variations lead the area of harmonic and
quasi-harmonic analyses of vibrating systems. Such tools therefore represent the
dynamic processes as a combination of the elementary rigid-body rotations with
different angular speeds.
Let us consider now the limit n ! 1, when the restoring force vanishes inside
the interval 1 < x < 1 but becomes infinitely large as the system reaches the
potential barriers at x D ˙ 1. The physical meaning of this limit is introduced at
the top of the right column in Fig. 4. Despite of the strong (impact) nonlinearity, the
limiting oscillator is also described by quite simple elementary functions such as the
P These two non-smooth functions
triangular sine and rectangular cosine, say and .
are associated with another subgroup of the rigid-body motions, namely translation
and reflection. Therefore, analogously to the case n D 1, the upper limit n D 1 can
play the same fundamental role by generating a hierarchy of tools as listed in the
right column of Fig. 4.
3 Further Mathematical Properties and Examples
This section describes the basic mathematical properties of the non-smooth temporal
substitutions introduced in Fig. 2. These properties are used then for derivations in
different illustrating examples.
3.1 Mathematical Properties
Consider first the single reflection case; see the first row of Fig. 2. Algebraic,
differential, and integral properties are as follows:
• Isomorphism with 2 2 symmetric matrixes:
t 2 D .a C s sP /2 D a2 C s 2 C 2as sP
2 2
2 a s a C s2 2as
b
t D D
sa 2as a2 C s 2
• Functional linearity holds for any function x(t):
x.t / D x .a C s sP / D X.s/ C Y.s/Ps (2)

18 V. Pilipchuk
X.s/ D 12 Œx. a C s /Cx. a s /

Y.s/ D 12 Œx. a C s /x. a s /
For instance, exp.t / D exp.a/.cosh s C sP sinh s/

• Division is conditioned to exclude the possibility of zero denominators:
.a s sP / a s sP a s
t 1 D D 2 D 2 sP
.a C s sP / .a s sP / a s2 a s 2 a2 s 2
.s ¤ jaj/
• Sequential differentiation remains in the algebra of hyperbolic numbers under

the smoothness conditions at s D 0:
d
ŒX.s/ C Y . s /Ps D Y 0 .s/ C X 0 .s/Ps
dt (3)
if Y.0/ D 0
d2
ŒX.s/ C Y . s /Ps D X 00 .s/ C Y 00 .s/Ps
dt 2
if Y.0/ D 0; X 0 .0/ D 0
• Integration remains in the algebra of hyperbolic numbers:
Z
ŒX .s.t // C Y .s.t // sP .t /dt
2s 3 2s 3
Z Z
D 4 Y.z/d z C C 5 C 4 X.z/d z5sP
0 0
3.2 Sample Solution Procedure for the s-Case
Let us consider the following initial value problem:
xP C x D 2pı .t a/ p sR
x.0/ D 0 (4)
s D jt aj
Introducing the new temporal argument, t ! s, generates the following substitu-

tion for the unknown function: x.t / D X.s/ C Y.s/Ps . As a result, the differential
equation takes the form
0
0
Y
„ ƒ‚C X
… C „X C Y sP C .Y p/ sR D 0 (5)
ƒ‚ … „ ƒ‚ …
regular stepwise discont: singular term
Equating separately the terms of different levels of singularity in (5) to zero leads
to the following boundary value problem with no singular terms:
8 0
< Y C X D 0
X 0 C Y D 0 (6)
:
p Y.0/ D 0
The initial condition in (4) yields
X.a/ Y.a/ D 0 (7)
The boundary value problem (6) and (7) is easy to solve in few steps. Then the
corresponding solution of the original initial value problem (4) is obtained in the
closed-form x.t / D p exp .s/ .1 C sP /; see Fig. 5 for illustration.
3.3 Properties and Sample Solution for the -Case
Algebraic and differential properties of the periodic -version, which is shown in the
second row of Fig. 2, are similar to those listed in Sect. 3.1. Note that, although the
analytical definitions for the basic functions and P look more complicated in this
case, there is no need for memorizing them. What is necessary for solving problems
is the following properties:
0
Fig. 5 Solution of problem
(4) under the following 1
parameters: P D 1.0; a D 1.0; 0 1 2 3 4 5
D 1.0 t
20 V. Pilipchuk
P D e; e 2 D 1; eP ¤ 0 () D ˙1 (8)
The third relationship in (8) means that whenever £ reaches its amplitude values,
a ı-spike occurs from the following series:
1
X
P /D2
e.t Œı .t C 1 4k/ ı .t 1 4k/
kD1
In other words, the present situation is quite similar to analytical manipulations

with the conventional trigonometric functions using only the function properties
with no involvement of their definitions. In order to illustrate the manipulations, let
us consider the overdamped oscillator under the rectangular cosine loading:
xP C x D pe.t / (9)
The unknown periodic solution is represented in the form [4]
x D X ./ C Y ./P (10)
Substituting (10) in (9) and taking into account properties (8) gives
0
0
Y C X C X C Y p e C Y eP D 0 (11)
„ ƒ‚ … „ ƒ‚ … „ƒ‚…
regular stepwise discont: singular
Equating separately to zero the terms of different levels of singularity gives the
autonomous boundary value problem with no discontinuities:
8 0
< X C Y D p
Y 0 C X D 0 (12)
:
Y .˙1/ D 0
Substituting solution of the boundary value problem (12) in (10) gives finally the
periodic closed-form solution of the original equation (9); see Fig. 6 for illustration:

p sinh. / cosh ./
xD C 1 e (13)
cosh cosh
Obviously, (9) can be also solved by means of either Fourier series or Laplace
transforms or directly by matching different pieces of solution under periodicity
conditions. However, using the Fourier series, for instance, requires a very large
number of terms near the non-smoothness points as shown in Fig. 6.
Fig. 6 Solution (13) for two 1.0

different magnitudes of the
parameter: D 1.0 – solid
line, larger amplitude; 0.5
D 5.0 – thin line, smaller
amplitude 0.0
0.5
1.0
0 2 4 6 8
t
4 Conclusions
This paper outlines the basic principles of non-smooth temporal substitutions and
gives exactly solvable illustrating examples. Note that the transition to non-smooth
temporal argument should be viewed as a preliminary stage of analyses. Such stage
of transformation incorporates the specifics of external loading or/and inherent
properties of a physical system into the new equations. As a result, a much wider
range of analytical and numerical methods becomes possible to apply. This is due
to the fact that the new temporal arguments vary within narrower domains and
automatically capture the major temporal symmetries of the dynamics, such as
periodicity and reflections. The corresponding analytical algorithms and solutions
for different strongly nonlinear oscillators can be found in the reference [4]. The
typical form of such solutions is power series with respect to the triangular sine
; see row 5 in Fig. 4 for explanation. Note that direct power series expansions
with respect to the original time t usually make little sense for vibration problems
due to the loss of periodicity. The amplitude and frequency modulated motions can
be analyzed by adding a slow time argument to representation (10) and then using
the idea of two variables or multiple-scale expansions. Let us mention also a new
area of applications, which is being developed due to an interesting observation
that the temporal mode shapes of the phase variable, describing the 1:1 resonance
energy exchange between weakly nonlinear oscillators, resemble the dynamic states
of impact oscillator [3]. In particular, it was found that such “impacts” take place
when the entire energy is involved into the exchange process.
Finally, let us summarize the revealed links between specific cases of the
Euclidean rigid transformations, the induced algebraic structures, and linear and
nonlinear dynamics. The class of rigid transformations T of an arbitrary vector r is
described by the expression T(r) D Ar C b, where A and b are the orthogonal matrix
and a constant translation vector, respectively. Then the abovementioned logical
links are illustrated by the table, which can be viewed as an extension of Fig. 4:
22 V. Pilipchuk
8
< C1 () Rotation () i 2 D 1 Elliptic numbers () Linear dynamics
det .A/ D 0 () Singular case () i 2 D 0 Parabolic numbers () ‹
:
1 () Reflection () i 2 D C1 Hyperbolic numbers () Nonlinear dynamics
where the singular case generates an open question.
References
1. Kisil, V.V.: Induced representations and hypercomplex numbers. Adv. Appl. Clifford Algebras
23(2), 417–440 (2013)
2. Lyapunov, A.M.: Investigation of a singular case of the problem of stability of motion. Math.
Sbornik 17, 252–333 (1893)
3. Manevitch, L.I.: New approach to beating phenomenon in coupled nonlinear oscillatory chains.
Arch. Appl. Mech. 77, 301–312 (2007)
4. Pilipchuk, V.N.: Nonlinear Dynamics: Between Linear and Impact Limits. Springer, Berlin
(2010)
5. Pilipchuk, V.N.: Transformation of oscillating systems by means of a pair of nonsmooth periodic
functions. Dokl. Akad. Nauk Ukrain. SSR, Ser. A (4), 37–40 (1988)
Non-linear Phenomena Exhibited by Flexible
Cylindrical and Sector Shells
V.A. Krysko, J. Awrejcewicz, I.V. Papkova, V.B. Baiburin, and T.V. Yakovleva
Abstract Vibrations of flexible cylindrical and sector shells subjected to the

action of uniformly distributed static loads are studied. The analyzed problems
are solved using two methods: the Bubnov–Galerkin method (BGM) and the finite
difference method (FDM). Validity and reliability of the results is verified through a
comparison to the results obtained by Andreev et al. (Stability of Shells Under Non-
Symmetric Deformation. Nauka, Moscow, 1988) in the case of a nonlinear static
problem.
1 Introduction
The variety of loading applied plays a crucial role while estimating the strength of
materials in numerous constructions working in high-temperature fields. Proper
estimation of the construction strength requires the detailed analysis of the
elastic-plasticmaterial behavior, initial deflections, interaction of the construction
elements, or interaction of those elements with the surrounding medium. The
proper estimation of the construction stability requires development of suitable
computational algorithms [12].
V.A. Krysko () • I.V. Papkova • T.V. Yakovleva

Department of Mathematics and Modeling, Saratov State Technical University,
Politehnicheskaya 77, 410054 Saratov, Russian Federation
e-mail: tak@san.ru; ikravzova@mail.ru; Yan-tan1987@mail.ru
J. Awrejcewicz
Department of Automation, Biomechanics and Mechatronics,
Lodz University of Technology, 1/15 Stefanowski Str., 90-924 Lodz, Poland
Department of Vehicles, Warsaw University of Technology,
84 Narbutta Str., 02-524 Warsaw, Poland
e-mail: jan.awrejcewicz@p.lodz.pl
V.B. Baiburin
Department of Information Security of Automated Systems, Saratov State
Technical University, Politehnicheskaya 77, 410054 Saratov, Russian Federation
e-mail: tak@san.ru

24 V.A. Krysko et al.
Many researchers have applied the Bubnov–Galerkin methods (BGMs) in the

Vlasov form as well as Ritz and FDM methods to solve problems of the stability
of beams, plates, and shells subjected to the action of a constant transversal load
and taking into account the geometric nonlinearities. The mentioned numerical
approaches yield reliable and validated results regarding a wide class of both
stationary and nonstationary problems of mathematical physics. In the case of
periodic loading, chaotic vibrations of the mentioned structural members may
appear [3, 13, 14]. The so far mentioned computational approaches reduce the
continuous problems to those of finite degrees of freedom [9, 18].
In order to investigate stability loss one may apply a few different criteria.
Since stability loss of an arbitrary deformed object takes place in time, therefore
it should be studied using various approaches to dynamics. However, a majority of
the stability problems of construction can be studied within static approaches, where
the equilibrium states are formulated without the inclusion of inertial forces.
Investigation of the stability loss is carried out using a dynamic criterion.
Namely, we define it through a buckling of an equilibrium form. Those loads being
responsible for the buckling occurrence are further named the critical loads.
We omit here an overview of the fundamental works dealing with the mentioned
problems, but we mention the method developed by Feodosev [10] regarding
nonlinear problems of shells, which is rather omitted in English-language literature.
In the latter one being originally named by Feodosev as the variational-step
method, the system deformation is considered as a process independent of either
fast or slow changes of the external load. For this purpose, time is introduced
artificially and equations of motion are derived. Nowadays this method refers to the
iteration process of finding solutions to nonlinear algebraic equations, where results
obtained in each computational step are improved, finally approaching the desired
exact solution of the problem. In this method of relaxation a solution to PDEs is
reduced to the Cauchy problem of ODEs.
The proposed algorithm is used to solve a wide class of static and dynamic
problems. We show a few possibilities of this approach to solve geometrically
nonlinear static and dynamic problems. We consider a mechanical system subjected
to the action of the transversal uniformly distributed constant load over the shell
surface and we consider the load in the form of the impulse with infinite action.
Since the problem of a critical static load plays here a crucial role, we briefly
describe the known criteria of stability proposed by numerous researchers.
Already Volmir [19] proposed the following criterion: either a fast increase of
deflection corresponding
2 toa small decrease of load appears or an inflexion point of
@ q
the relation q.w/ @w 2 D 0 occurs. On the other hand a load, where the increased
process of time is responsible for the achievement of the first maximum in the
load–time characteristic, is treated as the critical one. Kantor, who solved numerous
problems of axially symmetric spherical shells using the Ritz method, proposed the
following dynamic criterion responsible for the beam buckling [11]. The buckling
occurs if in the shell center its deflection achieves K Š 2f , where f D f = h and
f denotes deflection, whereas h is the shell thickness.
Non-linear Phenomena Exhibited by Flexible Cylindrical and Sector Shells 25
In references [7, 16] different criteria are proposed. Namely, the system transits
into a new dynamic state with the corresponding zero velocity. It can be explained
in the following manner. In the beginning the inertial forces act against the external
load, and after transition through zero they change sign and support the action of
the external load. It means that in a certain time instant the beam center velocity
achieves zero and then a sudden change of deflection occurs. In reference [8] the
time instant is taken as the stability loss criterion, where the displacement of an
elastic body changes without a change of the associated accelerations and velocities.
In some works the problem of dynamical stability loss is reduced to a quasi-
dynamical problem. Owing to this approach, the precritical stress of the middle shell
process is analyzed via static approaches. There are also works where a dynamic
criterion of the stability loss is matched with the occurrence of plastic deformations
of shell structures.
In reference [17], arcs were investigated and their buckling process was charac-
terized by two different mechanisms. In the case of the direct buckling mechanism
an unstable construction state was realized via symmetric forms. In the case of
indirect buckling, the system lost its stability via nonsymmetric forms. Since
the system stability loss via symmetric and nonsymmetric forms is qualitatively
different, one may expect two different dynamic criteria of the stability loss.
In this work by a critical load we mean limiting load values or the point of
inflexion of the relation wmax .q/. Further on, we will investigate critical loads acting
on axially symmetric spherical and conical shells, on a closed cylindrical shell as
well as on a spherical sector shell.
2 Shallow Closed Cylindrical Shells
We study shallow shells, i.e., objects in R3 with the associated curvilinear coor-
dinates x; y; z, introduced in the following manner. In the shell body the middle
surface z D 0 is fixed; axes ox and oy overlap the main shell curvatures, whereas
axis oz shows curvature surface
n origin (Fig. 1). In the given coordinates
h the
io shell is
defined as follows: ˝ D x; y; z=.x; y; z/ 2 Œ0; a Œ0; b h=2; h=2 , where
dimensional quantities are denoted by bars.
The governing nonlinear dynamics of the shell shown in Fig. 1 is obtained
assuming that the shell material is isotropic, homogeneous, and elastic and it
satisfies the Kirchhoff–Love hypotheses. Furthermore, we assume that the length
of the shell fiber along shell thickness remains unchanged [19].
Therefore, in the nondimensional form, the equation of motion of the shell
element as well as the deformation compatibility equations have the following
nondimensional forms:
Fig. 1 Computational scheme of a cylindrical shell
2 2
1 @2 w @2 ./ 2 2
2 @ w @ ./ @2 w @2 ./ @ w @ ./
2 2 2
C 2 2
C 2.1 / C C
@x @x @y @y @x@y @x@y @x 2 @y 2

@2 w @2 ./ @2 w @w
C rk2 F L.w; F / C M q.t / C" D 0;
@y 2 @x 2 @t 2 @t

@2 F @2 F @2 ./ 1 @2 F @2 F @2 ./
2 2
2 C 2 C (1)
@y @x @y 2 2
@x 2 @y @x 2

@2 F @2 ./ 1
C2.1 C / C rk2 w C L.w; w/ D 0:
@x@y @x@y 2
The following relations hold between dimensional and nondimensional quantities:
w D hw; F D Eh2 F ; t D t0 t ; " D "=; x D Lx; y D Ry;
r
h Eh4 LR
ky D k y .kx D 0/; qDq ; D ; (2)
R2 L2 R 2 h Eg
L
M D ky2 ; D ;
R
where L and R D Ry correspond to the shell length and radius, respectively.

In addition, we have t , time; ", damping coefficient; D 0:3; and q.x; y; t /,
transversal load. One of the following boundary conditions is taken:
1. Moving clamping
@w @F
w D 0I D 0I F D 0I D0 for x D 0I 1; (3)
@x @x
@w @F
w D g.x; y; t /I D p.x; y; t /I F D u.x; y; t /I D v.x; y; t / for y D 0I :
@y @y
2. Pinned support
@w @2 F
w D 0I D 0I F D 0I D 0 for x D 0I 1; (4)
@x @x 2
@w @F
@y @y
3. Moving clamping with ribs
@2 w @F
w D 0I D 0I F D 0I D 0 for x D 0I 1; (5)
@x 2 @x
@w @F
@y @y
4. Pinned support with flexible ribs
@2 w @2 F
w D 0I D 0I F D 0I D0 for x D 0I 1; (6)
@x 2 @x 2
@2 w @2 F
w D g.x; y; t /I D r.x; y; t /I F D u.x; y; t /I D z.x; y; t / for y D 0I :
@y 2 @y 2
Here we take D 2 for a closed cylindrical shell. In addition, the following

initial conditions are applied:
wjtD0 D w0 ; P tD0 D w0 :
wj (7)
2.1 The Bubnov–Galerkin Method (BGM)
After application of the BGM the following system of algebraic-differential equa-

tions is obtained:
G.AR C "A/
P C HA C C1 B C D1 AB D Qq.t /;
(8)
C2 A C PB C D2 AA D 0;
where H D kHijrs k, G D kGijrs k, C1 D kC1ijrs k, C2 D kC2ijrs k, D1 D

kD1ij klrs k, D2 D kD2ij klrs k, P D kPijrs k - square matrices of dimensions
2 N1 N2 2 N1 N2 , A D kAij k, B D kBij k, Q D kQij k matrices of dimension
2 N1 N2 1.
The second equation of system (8) is solved regarding B on each of the
computational steps:
B D ŒP1 D2 A P1 C2 A: (9)
P D R, the following Cauchy problem is formulated

Multiplying by G1 and taking A
for nonlinear ODEs:
P D "R ŒG1 C1 C G1 D1 A B G1 HA C G1 Qq.t /;

R
P D R: (10)
A
It is solved via the fourth-order Runge–Kutta method, and the computational step
in time is chosen using the Runge rule. We apply the method of relaxation for the
closed cylindrical shells with D 2 and we compare our results with the solution
obtained by Andreev et al. [1] for the corresponding static problem. We consider the
case of transversal external load whose location is defined by the central angle '0 .
In order to get qcr .'0 / we need to construct a set fqi ; wi g for 8'0 2 Œ0I 2, which
yields the critical load qcr . As it has been shown in [15], an increase in the number
of approximations yields a remarkable improvement of the obtained results.
The following conclusion can be formulated: in the case of nonhomogeneous
load, the use of a small number of the series terms yields large computational
errors and the obtained results depend essentially on the number of introduced
approximations. However, the situation changes qualitatively beginning from N D
13. Namely, the dynamical properties of the cylindrical shell are stabilized and a
further increase of N does not improve the obtained results either qualitatively
or quantitatively. Therefore, beginning from N D 13, a convergent series is
obtained and all further computations are carried out for N D 13. Consequently,
we constructed the relation of the critical loads versus width of the pressure zone
q cr .'0 / for N D 13 reported in [15].
Dependencies q cr .'0 / reported by Andreev et al. [1] for the closed cylindrical
shell for D 2 are in agreement with the results obtained by our method. Hence,
the obtained results indicate high efficiency of the proposed method for solving
static problems.
3 Shallow Sector Shells
Consider now a non-axially symmetric spherical shell in R2 in the polar coordinates

bounded by a contour , introduced in the following way: ˝ D ˝ C D
f.r;
; z/jr 2 Œ0; b;
2 Œ0;
k ; z 2 Œh=2; h=2. Equations governing the dynamics
of shallow shells are obtained from a system of equations of the rectangular
spherical shell via transition to the polar coordinates:
w00 C "w0 D r 2 r 2 w C N.w; F / C r 2 F C 4q;

(11)
r 2 r 2 F D r 2 w N.w; w/;
where
@2 ./ 1 @./ 1 @2 ./

r 2 ./ D 2
C C 2 ;
@r r @r r @
2
@4 ./ 2 @3 ./ 1 @./ 1 @./

r 2 r 2 ./ D 4
C 3
2 2 C 3 C
@r r @r r @r r @r
2 @4 ./ 2 @3 ./ 4 @2 ./ 1 @4 ./

C C C ;
r 2 @
2 @r 2 r 3 @
2 @r r 4 @
2 r 4 @
4

@2 w 1 @F 1 @2 F @2 F 1 @w 1 @2 w
N.w; F / D C 2 2 C C 2 2 (12)
@r 2 r @r r @
@r 2 r @r r @

@ 1 @w @ 1 @F
2 ;
@r r @
@r r @

2
@2 w 1 @w 1 @2 w @ 1 @w
N.w; w/ D 2 C 2 2 2 :
@r 2 r @r r @
@r r @
Boundary conditions follow:

1. Pinned support of arc slices
@2 w v @w @F
w D 0; C D 0; F D 0; D 0: (13)
@r 2 r @r @r
2. Pinned support of radial slices
@2 w @2 F
w D 0; D 0; F D 0; D 0: (14)
@
2 @
2
Fig. 2 Mesh of a sector shell
3. Sliding clamping of arc slices
@w @F
w D 0; D 0; F D 0; D 0: (15)
@r @r
4. Sliding clamping of radial slices
@w @2 F
w D 0; D 0; F D 0; D 0: (16)
@
@
2
Initial conditions are as follows:
w D f1 .r;
/ D 0; w0 D f2 .r;
/ D 0 for t D 0: (17)
3.1 Finite Difference Method
In order to reduce the continuous system governed by (12)–(17) to a lumped system

by the FDM with the approximation O.2 / versus spatial coordinates r and
(Fig. 2), the following difference operators are applied:
.w/ C rr w.F C rr F / C rr F .w C rr w/
2 r
wr
F C F C 4qi D .wt t C "wt /i;j ; (18)
.F / D rr w.w C rr w/ C .r
w/2 w;
where
1 1 1
./ D rr ./Cr ./; r ./ D ./r ; rr ./ D ./rr ; r
./ D 2 ./
C ./r
;
ri2 ri ri
1 1
rr ./ D Œ./iC1 2./i C ./i1 ; r ./ D Œ./iC1 ./i1 :
2 2 ri2
Boundary conditions:
1. Pinned support of arc slices
v
wN;j D 0; rr w r w D 0; FN;j D 0; r w D 0; j D 1; : : : ; M 1:
b
(19)
2. Pinned support of radial slices
wi;j D 0;
w D 0; Fi;j D 0;
F D 0; j D 0; M; i D 0; : : : ; N:
(20)
3. Sliding clamping of arc slices
wN;j D 0; r w D 0; FN;j D 0; r F D 0; j D 1; : : : ; M 1: (21)
4. Sliding clamping of radial slices
wi;j D 0;
w D 0; Fi;j D 0;
F D 0; j D 0; M; i D 0; : : : ; N:
(22)
The system of (18)–(22) should be supplemented by conditions to be satisfied in
the shell cusp and the matching conditions. In the majority of cases it is assumed
that a shell has a circular hole of small dimension in its cusp, and this assumption
does not influence computational results essentially. In this work, while solving
nonsymmetric problems for
D 2 , the approximating functions in the point
r D 0 are interpolated by the Lagrange formula of the second order. We have
f0;j D 3 f1;j 3 f2;j C f3;j ; (23)
where fi;j D f .ri /j , ri D i h .i D 0; 1; 2; 3/, 0 j M 1, and h is the

distance between the nodes of interpolation. In the case of a point lying out of the
contour the following symmetry condition holds:
f1;j D f1;j for 0 j M 1: (24)
Matching conditions for non-axially symmetric problems

D 2 follow:
wi;j D wi;M Cj ; Fi;j D Fi;M Cj for j D 0I 1; 0 i N 1: (25)

a b
q0 q0 b = 5
b=5 b=6 b=6
1.2 1.2
b=7
1 1
b=7 b=8
0.8 0.8
b=9
b=9
0.6 0.6
b=8 b = 10
0.4 0.4
b = 10
0.2 0.2
0 0
0 5 10 15 20 Wmax 0 5 10 15 20 Wmax
c
q0
b=8 b = 10
1.2
b=7 b=9
1
0.8
b = 11
0.6
0.4
b = 12
0.2
0
0 5 10 15 20 Wmax
Fig. 3 Computational results of sector shells analysis
The Cauchy problem (18)–(25) is solved via the fourth-order Runge–Kutta method,
where the computational step has been chosen due to the Runge rule [2, 4–6].
Figure 3 shows the dependencies q.w/ for sector angles
k D 32 (a), for
k D (b), and for

k D 2 (c) and for various shallow parameters for
k D 32
and
k D .b D 5; 6; 7; 8; 9; 10/, and for
k D 2 for the shallow parameter
b D 7; 8; 9; 10; 11; 12 (lower values of b are not used, since the sector shell behaves
like a plate).
The obtained graphs imply the following conclusions. Beginning from a certain
value of the parameter b the critical points appear. For a shell with
k D 32 I I 2
and b D 8; 9 and 11, respectively, the jump-type buckling occurs. Figure 4 illustrates
the curves of equal deflections (isoclines) for all studied angles
D 2 I I 32
. In
what follows we compare the behavior of curves for the critical and post-critical
load q0 for different
and b. For
D 2 and
D and for an arbitrary shallow
parameter the obtained results coincide and the maximum deflection is achieved on
the intersection of the bisectrix of angle
and the central shell radius. For
D 32
and for b D 7 the jump-type buckling is not observed, and the system dynamics
a b
0.8 0.8
0 0
10 0
0.6 0.6
2
3 5
0.4 1 0.4 5 10
4 10 0
2 10
5 15
0.2 3 4 3 0.2 5 10
2 5
1 2
0 1 0
0 0 0 0 0 0
c d
1 0 1 0 0
0 0 0
0 2 0
1 0 0
5
1 43 10
0.5 0.5
−1 35 65 2 0 0 10 15 0
0 5
2 6 7 34 1 5
0 2 0 0 10
0
12 4 6 1 5 15 5
3 5
−1 0 1 4 −0.5
0 10
−0.5 0 0
0 0 0
0 0 0 0
0
−1 0 0 0 −1 0
Fig. 4 Computational results of sector shells (isoclines) for

D 2 ; b D 12; q0 D 0:245 (a),
D

2
; b D 12; q0 D 0:246 (b),
D 3
2
; b D 10; q0 D 0:27 (c) and
D 3 2
; b D 10; q0 D 0:28 (d)
is not changed qualitatively. In Fig. 3a the same sector angle and the first-order
discontinuity imply qualitative changes in the location of isoclines for both critical
and post-critical load (see Fig. 4). For
D 2 , b D 12 a fast shell deflection is
observed but without a change of the shell form. For
D 3 2
, b D 10 for the
critical load, the maximum shell deflection occurs in the middle of the bisectrix,
whereas in the case of post-critical load two zones of maximum deflection appear,
being symmetric with respect to the angle bisectrix.
4 Concluding Remarks
In this work the relaxation method has been applied to study flexible cylindrical and
sector shells. The Bubnov–Galerkin and the finite difference approaches allowed
us to reduce the problems to the algebraic-ordinary differential equations. A few
interesting nonlinear phenomena exhibited by the analyzed continuous systems have
been reported.
Acknowledgements This work has been supported by the grant RFBR No. 12-01-31204 and the
National Science Centre of Poland under the grant MAESTRO 2, No. 2012/04/A/ST8/00738, for
the years 2013–2016.
References
1. Andreev, L.V., Obodan, N.I., Lebedev, A.G.: Stability of Shells Under Non-Symmetric
Deformation. Nauka, Moscow (1988) (in Russian)
2. Awrejcewicz, J., Krysko, V.A., Krysko, A.V.: Thermodynamics of Plates and Shells. Springer,
Berlin (2007)
3. Awrejcewicz, J., Krylova, E.Yu., Papkova, I.V., Yakovleva, T.V., Krysko, V.A.: On the memory
of nonlinear differential equations in theory of plates. Vestnik Niznegorodskogo Universiteta
im. Lobachevskogo 4(2), 21–23 (2011) (in Russian)
4. Awrejcewicz, J., Krysko, A.V., Papkova, I.V., Krysko, V.A.: Routes to chaos in continuous
mechanical systems. Part 1: mathematical models and solution methods. Chaos Solitons
Fractals 45, 687–708 (2012)
mechanical systems: Part 2. Modelling transitions from regular to chaotic dynamics. Chaos
Solitons Fractals 45, 709–720 (2012)
mechanical systems. Part 3: the Lyapunov exponents, hyper, hyper-hyper and spatial-temporal
chaos. Chaos, Solitons Fractals 45, 721–736 (2012)
7. Bakhtieva, L.U.: Investigation of stability of thin shells and plates subjected to dynamic loads.
Ph.D. Thesis, Kazan (1981) (in Russian)
8. Darevskiy, V.M.: Nonlinear equations in theory of shells and their linearizations in problems of
stability. In: Proceedings of VI Russian Federation Conference on Theory of Plates and Shells,
pp. 355–368. Nauka, Moscow (1969) (in Russian)
9. Deresiewicz, H.: Solution of the equations of thermoelasticity. In: Proceedings of 3rd National
Congress of Applied Mechanics ASME, pp. 685–690, 1958
10. Feodosev, V.I.: On the solution of non-linear stability problems of deformable systems. PMM
27(2), 265–274 (1963) (in Russian)
11. Kantor, B.Ya.: On the Non-linear Theory of Shells. Dynamics and Strength of Machines.
Kharkiv National University, Kharkiv (1967)
12. Krysko, V.A., Awrejcewicz, J., Kuznetsova, E.S.: Chaotic vibrations of shells in a temperature
field. In: Proceedings of the International Conference on Engineering Dynamics, Carvoeiro,
Algarve, Portugal, , pp. 21–28, 2007
13. Krysko, V.A., Saltykova, O.A., Yakovleva, T.V.: Nonlinear dynamics of antennas in cosmo-
nautic coupling tools. Izvestia VUZ. Aviation Tech. 2, 60–62 (2011) (in Russian)
14. Krysko, A.V., Koch, M.I., Yakovleva, T.V., Nackenhorst, U., Krysko, V.A.: Chaotic nonlinear
dynamics of cantilever beams under the action of signs-variables loads. In: PAMM, Special
Issue: 82nd Annual Meeting of the International Association of Applied Mathematics and
Mechanics (GAMM), vol. 11(1), pp. 327–328, Graz, 2011
15. Krysko, V.A., Awrejcewicz, J., Papkova, I.V., Baiburin, V.B., Yakovleva, T.V.: Method of
relaxation in theory of flexible shells. In: Dynamical Systems Theory. pp. 477–490. TU of
Lodz Press, Lodz (2013)
16. Kuznetsov, E.B.: On action of dynamic load on systems with buckling. Probl. Appl. Mech.
12–35 (1974) (in Russian)
17. Lock, M.H.: Snapping of a shallow sinusoidal arch under a step pressure load. AIAA J. 4(7),
31–41 (1966)
18. Takezono, S., Tao, K., Inamura, E., Inoue, M.: Thermal stress and deformation in functionally
graded material shells of revolution under thermal loading due to fluid. JSME Int. J. Ser. A
Solid Mech. Mater. Eng. 39(4), 573–581 (1996)
19. Volmir, A.S.: Stability of Elastic Systems. Fizmatgiz, Moscow (1963) (in Russian)
Gear Shift Patterns in Uncertain Terrestrial
Locomotion Systems
Carsten Behn and Silvan Schwebke
Abstract The paper focusses on various gear shift patterns in an uncertain

terrestrial locomotion system, i.e., in a wormlike locomotion system (WLLS).
A WLLS in this theory is understood as a system living in a straight line
(one dominant linear dimension) with no active (i.e., driving) legs nor wheels.
A mechanical model comprises a chain of discrete mass points (1) connected by
viscoelastic force actuators and (2) having ground interaction via spikes which
make the velocities unidirectional. A spike means any device which realizes this
restriction. The distances between two consecutive mass points are changed by an
adaptive controller. In combination with the ground contact spikes, this results in an
undulatory locomotion of the system. Optimal gaits which achieve a defined number
and succession of resting mass points as well as the resulting velocity are developed
in numerical investigations. We present a gait shifting procedure incorporating a
combination of speed adjustment and gait change that enables optimal crawling for
predefined limits of actuator or spike force load.
1 Introduction: Model and Control Approach
A mechanical model of a WLLS comprises of a chain of interconnected, discrete

mass points. Akin to the earthworm, the worm is equipped with spikes that inhibit
backward motion. Actuators are assumed to be positioned between two adjacent
mass points to change the distance between them (change of shape of the worm).
This results in a global movement (undulatory locomotion [5]).
Regarding the movement pattern of WLLS, current literature (see [3, 7, 8])
often takes the basic idea from the movement of biological worms. However, the
mathematical description of these patterns is based on a purely kinematic view,
at times just using two states for the system’s actuating elements—elongated and
contracted. The generation of specific gaits is discussed, but they are not analyzed
and compared in terms of dynamics. Such analyses are the focus of this work.
C. Behn () • S. Schwebke

Department of Technical Mechanics, Ilmenau University of Technology,
Max-Planck-Ring 12, 98693 Ilmenau, Germany
e-mail: carsten.behn@tu-ilmenau.de; silvan.schwebke@tu-ilmenau.de

38 C. Behn and S. Schwebke
For that, the existing theory for gait generation as described in [9] is used. Since
all system parameters are assumed to be unknown or, more precisely, uncertain,
an adaptive controller is used to achieve the tracking of certain gaits. After finding
optimal gaits, a procedure for automatic gait shift based on the measurement of
certain restricted system variables is implemented.
1.1 Modeling
The starting point is the chain of mass points in a common straight line as shown
in Fig. 1. The variables xi .t / (i D 0; : : : ; n) are the current coordinates of the
segments. The distance between mi and mi1 are labeled as lj .t / WD xi1 .t /xi .t /.
Each mass point is equipped with spikes, which are assumed to be ideal, i.e., they
restrict velocities from being negative [10].
To achieve the desired movement due to the paradigm earthworm, the distances
lj .t / have to be changed in an appropriate way. In this model, viscoelastic actuators
are used to apply forces between mass points. The acting forces on mass points are
(j D 1; : : : ; n) (see Fig. 2):
• spring forces Fc;j D cj .xj 1 xj l0;j /, where l0;j is the initial length of the
spring;
• damping forces Fd;j D dj .xP j 1 xP j /;
Fig. 1 Chain of mass points with spikes
Fig. 2 Mass point with forces

Gear Shift Patterns in Uncertain Terrestrial Locomotion Systems 39
• actuator forces uj and u0 D unC1 D 0;

• spike forces FZ;i , used to model the ideal spikes, therefore fulfilling three
conditions according to [9]:
xP i 0; FZ;i 0; xP i FZ;i D 0; (1)
with i D 0; : : : ; n. These conditions are fulfilled by the following explicit

equation given in [9]:
1
FZ;i D 1 sign.xP i / 1 sign.Fi / Fi ; (2)
2
with i D 0; : : : ; n, where Fi is the sum of all other forces acting upon mi ;
• the weight of a mass point is considered in the x direction only and thus depends
on the ground inclination ˛: FGx;i .˛/ D mi g sin.˛/.
The equations of motion are (using Newton’s second law):
mi xR i D ciC1 .xi xiC1 l0;iC1 / C ci .xi1 xi l0;i /

(3)
diC1 .xP i xP iC1 / uiC1 C ui C di .xP i1 xP i / C FZ;i C FGx;i
with i D 0; : : : ; n and c0 D cnC1 D d0 D dnC1 D u0 D unC1 D 0. This system can

only be influenced using the actuator forces u1 .t / through un .t /. They shall serve as
control inputs to achieve global movement: the distances between mass points shall
follow prescribed time-variant functions.
1.2 Control
To achieve the mentioned objective, the following variables are considered:

• lj .t / D xj 1 .t / xj .t /, the distances between adjacent mass points, which
represent the system outputs;
• lref;j .t /, the reference functions of these distances;
• ej .t / WD lj .t / lref;j .t /, the error of the output variables.
The task of the controller is to find appropriate actuator forces, i.e., find a mapping
e 7! u. The controller shall stabilize the WLLS with unknown or uncertain (known
within an interval) parameters. Therefore, adaptive control as described in [1] is
used. The controller consists of a PD-structure and a gain adaption law to achieve
stability. The control goal is to track a reference function of the output variables
within a given tolerance (accuracy of tracking), which is called -tracking [4].
The following controller was proposed in [2]:
9
e.t / WD l.t / lref .t / >
>
>
>
>
u.t / D k.t / e.t / C k.t / eP .t / >
>
>
>
>
>
D k.t / .e.t / C eP .t // >
>
=
8 (4)
ˆ
ˆ .ke.t /k /2 ; ke.t /k C 1 >
ˆ
ˆ >
>
< .ke.t /k /0:5 ; >
>
P /D ke.t /k 2 ŒI C1/ >
>
k.t >
>
ˆ
ˆ >
.ke.t /k < / ^ .t tE < td / >
ˆ0; >
>
:̂ ;
k.t /; .ke.t /k < / ^ .t tE td /
with > 1; > 0; > 0; td 0; > 0; k.0/ > 0.

The adaption law results in an increase of the gain k./ if the error norm
exceeds . The quadratic increase in case of ke.t /k C 1 forces the error signal
to the tube in generating a quick rise of the gain for these very large error values.
To make sure the gain increase is not too small for error norms that are smaller, but
P
exceed , the square-root function is used to determine k./—now the error signal
is forced inside the tube. The variable tE denotes the time of the last entrance of
the error norm into the tube. When ke.t /k is smaller than , the gain remains at a
constant value for a predefined time of stay td and then decreases exponentially.
2 Gaits: Construction and Analysis
The construction of reference trajectories lref;j .t / (i.e., gaits) that enable the desired
global movement of the WLLS consists of two steps, as described in [9]:
First, it is chosen which mass points should have active spikes (i.e., are resting)
at which periods of time. The number of active spikes shall be identical for every
time and is denoted by a 2 f1; : : : ; ng. All sequences A.t / (e.g., A.t / D f0g !
f1g ! f2g ! f3g), which describe the set of active spikes as a function of time,
shall be periodic: mode-functions. Once the active spikes for specific time intervals,
i.e., equal parts of one period, are determined, it can be deduced which lj .t / have to
contract, which have to elongate, and which have to keep their length in each time
interval.
A short notation of the gait shall consist of the set of active spikes at t D 0,
A0 , and the direction in which the traveling wave of active spikes moves, d i r. The
direction from head to rear through the WLLS is called ` (left).
In addition to the previous definitions, the detailed lref;j .t / functions are yet to
be chosen. They should be of class C 2 (continuous up to the second derivative)
to ensure there are no discontinuous steps in the acceleration terms, as this is
undesirable in applications.
The reference functions are built as described in [9] on time intervals

T T
t 2 p ; .p C 1/ ; p 2 N0 :
N N
Table 1 Optimal gaits

for WLLS with N D 4
aD1
d i r D `, A0 D f1g
d i r D r, A0 D f2g
aD2
d i r D `, A0 D f1; 2g
d i r D r, A0 D f1; 2g
aD3
d i r D `, A0 D f2; 3; 0g
d i r D r, A0 D f3; 0; 1g
using a sin2 ./-approximation. Using D t p NT we exemplarily get as a part of

the reference gait
9
P
l./ D "l0 2Nf sin2 .f N / >
=
1 (5)
l. / D l0 C "l0 Nf "l0 sin .2f N / ;>
;
2
- j"j 2 .0I 1/ is the relative factor of the maximum contraction/elongation;

- f is the frequency of the A.t / sequence with periodic time T D f1 ;
- l0 > 0 is the initial (non-contracted, non-elongated) length;
- l0 is the length at the beginning of a time interval ( D 0), either l0 or l0 .1 C "/,
depending on the previous movement.
Numerical simulations are carried out to compare all gaits for the WLLS with
N D 4 that have a set of consecutive active spikes that moves in a traveling wave
through the system, like the paradigm earthworm does it. With spike forces, actuator
forces, and actuator powers—all of which can be relevant for applications—as
primary criteria, a set of six gaits is found, two for each a 2 f1I 2I 3g [6]. This
selection of optimal gaits is shown in Table 1.
3 Shift of Gaits
The change of gaits can adjust the whole system to a changing environment, e.g., a
varying slope. It is desired to implement an automatic, internal gait shifting relying
solely on measured system variables. The options for varying the movement pattern
are twofold:
1. changing the number of active spikes (value a) by switching the A.t / sequence;
and/or
2. changing the gait frequency (value f ) while keeping the general A.t / pattern.
This is possible at any time.
Isolated implementations of either of these principles limit the gait adjustment—
especially the mere change of the number of active spikes while keeping a constant
frequency, which only allows for n different reference speeds. The reference speed
can be obtained from kinematical considerations of the reference functions; it is
vN ref .a; f / D .N a/ " l0 f: (6)
This reference speed can be altered by changing a and f . To be able to adjust the
speed of the WLLS in an optimal way, a combined switching of the two variables is
desired.
Possible criteria for gait shift can include certain limits of actuator or spike
forces, to avoid reaching prohibited values; actuator power, as a maximum or
(moving) average value; and gain, which is influenced by external conditions such
as friction or slope of the terrain.
In simulations with a number of slopes, gaits, and speeds, shown in [6], it was
found that—if the maximum allowed spike forces or actuator forces are chosen as
criteria—the choice of the optimal gait depends on the current slope and speed,
i.e., there is a certain gait that allows for the largest speed for given force limits.
The implementation of an automatic gait shift requires the determination of n 1
speed thresholds, at which the number of active spikes should be switched. Between
the thresholds, the gait frequency should be adjusted. The exact values of these
thresholds can vary during locomotion, as they are not only dependent on internal
system parameters but also on external variables such as the slope.
Investigating the spike forces will enable the choice of speed thresholds. In case
of xP i D 0, the spike force FZ;i equals the sum of all other forces acting upon mass
point mi , with opposite sign:
FZ;i .0/ D uiC1 ui FGx;i C Fc;iC1 Fc;i C Fd;i1 Fd;i FR;i (7)
For the following estimation, all spring, damping, and friction forces are neglected.
As shown in parameter studies in [6] (p. 29 ff.), larger damping has little effect on
the spike forces, since it is compensated by higher actuator forces. The estimation
of the expected maximum value of uiC1 ui DW umax can be found with knowledge
of the reference functions lref ./ and the masses. For equal masses mi D m, the
actuator force maximum according to the reference functions (5) is

umax D m.N a/ 2N 2 f 2 "l0 C g sin.˛/ (8)
80
a=1
a=2
60 a=3
max(FZ) estimate
40
20
0
0 0.1 0.2 0.3 0.4
vmean,ref
Fig. 3 System with N D 4; estimate of the maximum spike force according to (9) vs. vN ref
according to (6), shown for different numbers of active spikes; slope ˛ D 25ı
The spike force estimate is completed by adding the x-component of the weight of
mass mi :
FZ;i;max .f; a/ D umax C mi g sin.˛/ ; (9)
which makes FZ;i;max a function of frequency (or reference speed) and number of
active spikes a. The estimated spike force maxima are shown in Fig. 3. For the
WLLS with N D 4 and given parameters, the estimates for different numbers of
active spikes a, a D 1; 2; 3, are plotted. As observed in previous simulations, the
actual speed resulting from a certain gait frequency can slightly vary, due to the
allowed error of .
The combination of the plotted functions exhibits certain speed intervals:
• for speeds of up to 0:18 ms1 , the gait with a D 3 requires the lowest spike force
maxima;
• between 0.18 and 0:31 ms1 , the gait with a D 2 is advantageous;
• for speeds higher than 0:31 ms1 , the gait with a D 1 is superior.
Remark 1. Consider driving a car as an analogy. A large value of a is equivalent
to a low gear of the car. To perform a gear shift in order to increase the speed, the
driver has to accelerate (open the throttle) first, i.e., increase f of the gait for the
worm system.
This corresponds qualitatively to the results of the simulations of the worm with
N D 4. The exact values of the speed thresholds differ slightly, possibly due to
neglecting spring and damping forces.
The relevant parts of Fig. 3 are the intersections of the curves. They represent the
speed thresholds at which the number of active spikes should be changed to achieve
the lowest possible spike forces. The speed thresholds are noted as vN ref;q$qC1 (with
q D 1; : : : ; n 1), meaning the gait reference speed at which the number of active
spikes should be switched from q to q C 1 or q C 1 to q.
The switching rule is the following:
• for speeds smaller than vN ref;q$qC1 , switch to a D q C 1 (larger number of active
spikes);
• for speeds larger than vN ref;q$qC1 , switch to a D q (smaller number of active
spikes).
Using (9) and (6), the general form of these speed thresholds is, for arbitrary N
(number of mass points),
s
q 1 1
vN ref;q!qC1 D 1 1 "l0 g sin.˛/; (10)
N 2 N q
whereby these speed thresholds only depend on N and ˛.

Remark 2. Note that we only assumed that all masses are equal, i.e., mi D m; their
actual value is not required to find these speed thresholds for switching the number
of active spikes.
Now, if the slope is known (measured), it is then possible to choose adequate
gaits, with respect to low spike forces.
Remark 3. The estimate of the actuator forces (see (8)) gives identical speed
thresholds, as it differs from the spike force estimate only by the addition of a weight
term of one mass; this term does not influence the intersection of curves in Fig. 3 for
different a.
It should therefore be possible to use the speed thresholds above to achieve
optimal locomotion, i.e., to use the gait that yields the lowest actuator/spike forces
(see Remark 3).
To adjust the reference speed in variable steps, it appears useful to apply
closed-loop control between the input variable reference frequency and an output
variable, which may be the measured (maximum) value of actuator forces, spike
forces, or actuator powers. This nice idea is based on [6]. The implementation is
based on simple P-feedback. This forms a proportional frequency control. First, a
reference value Xref for the output criterion is chosen. The actual output is built by
measurements of the chosen criterion over given intervals of time, in this case NT .
Measurement and frequency adjustment according to the control are evaluated after
each of these NT -time intervals, so that the result is a time-discrete controller which

provides discrete frequency values fk D f k NT .
The deviation between the measurement variable and its reference is evaluated
and fed back into the loop to form the new frequency. The measurement shall be
performed continuously over given time intervals, in order to find the maximum
value, i.e.,

Xerr;k D Xref Xmeas;k ; with Xmeas;k D max jX0 . /j; : : : ; jXn . /j :
2Œ T4 .k1/; T4 k
(11)
This error shall contribute proportionally to the factor of frequency adjustment. For
the frequency adjustment, a factor is used instead of absolute values, because a
feasible range of frequencies might not be known in advance and a relative factor
should enable proper adjustment for any range. The new frequency is calculated
according to
fkC1 D fk C fk Xerr;k kP D fk .1 C Xerr;k kP / ; (12)
with a gain kP 2 RC ; f0 2 RC . As described before, this implementation entails

relative frequency changes, where a certain deviation of the measured variable from
its reference always results in the same percental frequency change.
For practical implementations, the variable X may represent actuator forces,
actuator powers, or spike forces.
4 Simulation
In the following simulation of an automatic gait shift for the WLLS with N D 4,
the parameter set shown in Table 2 is used.
The slope is defined as a function of time:
8
ˆ
ˆ15ı ; t 10 s
<
˛.t / D 60ı ; 10 s < t < 20 s (13)
ˆ
:̂25ı ; t 20 s:
Three optimal gaits are chosen from Table 1:

• A0 D f1g; d i r D `;
• A0 D f1; 2g; d i r D `;
Table 2 Parameter set

t D Œ0; 20 s l0 D 1 m j"j D 0:4
ı
˛ D 25 f D 0:5 Hz D 0:2s1
mi D 1 kg D 0:05 m td D 2 s
cj D 10 N=m D 500 k.0/ D 10N=m
1
dj D 5 kg s D 1s
• A0 D f2; 3; 0g; d i r D `.
As an initial value, the gait with a D 1 and a frequency of f jtD0 D 0:5 Hz are
selected. The frequency is adjusted by the frequency P-feedback (12) presented
before.
The frequency thresholds for the shift of the number of active spikes are derived
according to (10); they are
s
g sin .˛/ p
f1 .˛/ D 2
0:4032 sin .˛/ Hz and
3N "l0
s
g sin .˛/ p
f2 .˛/ D 2
0:3492 sin .˛/ Hz:
4N "l0
With these thresholds, the rule for switching the number of active spikes is
8 9
ˆ
ˆ ifa D 2 ^ f 32 f1 W switch to a D 1 >
>
< =
if.a D 1 ^ f < f1 / _ .a D 3 ^ f > 2f2 / W switch to a D 2
ˆ ifa D 2 ^ f < f2 W switch to a D 3 >
>
:̂ ;
else W no switch
When switching a, a frequency change is applied to keep the same reference

speed, according to (6), as before. The reference speed shall only be changed by
the frequency P-feedback. For this feedback, the following parameters are chosen
according to (11) and (12): FZ;ref D 20 N, kP D 0:03 N1 . In the following plots,
the points of slope change (at 10 s and at 20 s) are marked vertical lines; additional
vertical lines denote points of gait shift. In the plot of the worm motion [Fig. 4(left)],
the different movement patterns of the gaits are visible.
For the larger slope, the gait is almost instantly shifted to a D 2 [see
Fig. 4(right)]. This happens even before the reference speed is reduced, because the
Fig. 4 Left, worm movement; right, reference speed vN ref

Fig. 5 Left, spike forces; right, gain k.t /
(measured) larger slope influences vN 1!2;N D4 , as seen in (10), so that the gait with
a D 2 is considered optimal. The speed is then reduced to a value that requires the
gait with a D 3 to be used. Later, when the slope is smaller, the gait is shifted back
to a D 2.
Within t D Œ0; 10 s the maxima of the spike forces [see Fig. 5(left)] adhere to
the reference value very closely. Later, when different gaits are used, the maxima of
some of the spike forces clearly exceed the reference value once per period by up
to 30%. The gait and slope switches require different gain levels [Fig. 5(right)]. The
necessary adjustments are performed quickly, within fractions of a period.
5 Conclusion
Overall, the adaptive controller is capable of quickly adjusting the gain when
frequency and gait pattern are changed, so that the control goal can be accomplished.
The combined switching of frequency and number of active spikes enables quick
speed adjustments in order to adhere to given limits of a system variable such as
spike or actuator force. The gait shift, i.e., the change of the number of active spikes,
helps to choose the gait that is optimal for a certain speed and slope. With this,
predefined restrictions of forces or powers of the system are utilized in such a way
that the maximum speed is reached.
Future work shall be directed to improvements of the concepts for automatic
gait shift (in order to achieve a faster and more robust shifting), analysis of shifting
concepts for WLLS with N > 4, and experimental verification.
References
1. Behn, C., Zimmermann, K.: Robotics Auton. Syst. 54, 529–545 (2006)
2. Behn, C., Steigenberger, J.: Int. Rev. Mech. Eng. (IREME) 4(7), 886–898 (2010)
3. Chen, I.M., Yeo, S.H., Gao, Y.: Robotica 19, 535–542 (2001)
4. Ilchmann, A.: Non-Identifier-Based High-Gain Adaptive Control. Springer, London (1993)
5. Ostrowski, J.P., Burdick, J.W., Lewis, A.D., Murray, R.M.: The mechanics of undulatory
locomotion: the mixed kinematic and dynamic case. In: Proceedings IEEE International
Conference on Robotics and Automation, Nagoya, Japan, 1995
6. Schwebke, S.: Contributions to adaptive control strategies of biomimetic, worm-like locomo-
tion systems and their use for gait shifting. Bachelor Thesis, TU Ilmenau (2012)
7. Seok, S., Onal, C.D., Wood, R., Rus, D., Kim, S.: Peristaltic locomotion with antagonistic
actuators in soft robotics. In: Proceedings 2010 IEEE International Conference on Robotics
and Automation, 2010
8. Slatkin, A.B., Burdick, J., Grundfest, W.: The development of a robotic endoscope. In:
Proceedings 1995 IEEE/RSJ International Conference on Intelligent Robots and Systems:
Human Robot Interaction and Cooperative Robots, vol. 2, pp.162–171, 1995
9. Steigenberger, J., Behn, C.: Robotics Auton. Syst. 5, 555–562 (2011)
10. Steigenberger, J., Behn, C.: Worm-like locomotion systems: an intermediate theoretical
approach. Oldenbourg, Munich (2012)
Turbulent Phenomena in Flexible Plates
and Shells
J. Awrejcewicz, A.V. Krysko, V.A. Krysko, E.Yu. Krylova, S.A. Mitskievich,

I.V. Papkova, T.V. Yakovleva, V.M. Zakharov, and V. Dobriyan
Abstract The aim of this chapter is to present a study of periodic and chaotic
dynamics of plates and shells and weak turbulent behavior exhibited by these solid
structural members modeled as 2D infinite objects. Besides the new results obtained
with respect to the transition from a regular to weak turbulent and weak hyper
turbulent behavior, we also present novel methods and approaches to get reliable and
validated results of numerical analysis of nonlinear partial differential equations.
In particular, besides the standard numerical techniques for chaos monitoring, new
effective approaches are presented and applied including the wavelet-based analysis,
charts of vibration regimes, computation of the spectra of Lyapunov exponents via
generalization of the classical Benettin’s approach, and application of the neural
network technique. This common strategy aimed at numerical computations through
various types of robust discretization allowed us to obtain novel scenarios of
transition from regular/laminar (periodic, quasiperiodic) to spatiotemporal chaotic
(weak turbulent) dynamics of flexible shells either parametrically excited (Sect. 2)
or through the periodic shear load action in the shell volume unit (Sect. 3), flexible
multilayer rectangular (Sect. 4) and cylindrical shells with gaps (Sect. 5), as well as
a flexible plate of infinite length (Sect. 6).
J. Awrejcewicz ()
Department of Automation, Biomechanics and Mechatronics,
Lodz University of Technology, 1/15 Stefanowski Str., 90-924 Lodz, Poland
Department of Vehicles, Warsaw University of Technology, 84 Narbutta Str.,
02-524 Warsaw, Poland
A.V. Krysko • V.A. Krysko • E.Yu. Krylova • S.A. Mitskievich • I.V. Papkova
T.V. Yakovleva • V.M. Zakharov • V. Dobriyan
Saratov State Technical University, Applied Mathematics and Systems Analysis,
Politekhnicheskaya 77, 410054 Saratov, Russian Federation
e-mail: anton.krysko@gmail.com; tak@san.ru

50 J. Awrejcewicz et al.
1 Introduction
This chapter addresses an important problem of turbulent wave dynamics exhibited

by a solid which has not been satisfactorily explained and clarified yet through the
investigations of vibrations of flexible plates and shells. It extends our earlier studies
on beams/plates/shells spatiotemporal chaotic dynamics. In particular, the wavelet
approach is applied, which contrary to the Fourier analysis (FFT) makes it possible
to follow the frequency spectrum time evolution and hence reveal the so far hidden
nonlinear phenomena that have not been described in the existing literature.
Recently, the attention of numerous researchers has been focused on the appli-
cation of the wavelet analysis to study nonlinear dynamics of continuous structural
members (beams, plates and shells) governed by partial difference equations (PDEs)
using the finite elements method (FEM), finite difference method (FDM), higher-
order Bubnov–Galerkin approach, or semi-discrete techniques.
It has been shown that wavelets allow for a solution decomposition into a set
depending on the scale and location, and the wavelet bases in multi-scales combined
either with FEM or FDM yield sometimes novel results, which have been omitted
using the standard techniques. The applied scaling functions provide a multi-
resolution analysis and a robust strategy for mesh refinement. For example, the
feasibility of a hybrid scheme of the combination of Daubechies wavelet functions
and FEM is outlined in [1], and a few results regarding the analysis of a Euler–
Bernoulli beam element and a Mindlin–Reissner plate element are given. The theory
of wavelets is widely used in identification problems, where a studied system evo-
lution is deduced from the wavelet transform response. Wavelets of the Daubechies,
Morlet, and Gauss families are used to detect faults in gears and gearboxes as
well as to estimate the location of cracks [21, 25, 30]. The wavelet-type analysis is
applied to lumped mechanical systems governed by ordinary differential equations
[12, 17]. The proposed procedure can be viewed as a logarithmic transform applied
to a fitted component of the original response. Damping ratios of a multi-degrees-
of-freedom systems are estimated via the wavelet-based formula allowing for the
separation of purely random modal contributions from the fundamental mode.
Linear time-varying dynamic systems governed by ODEs were studied in [2, 10]
using the wavelet-based Galerkin procedure. In particular, in the latter paper [10],
the damping and stiffness parameters were identified using the robust wavelet-based
algorithm. On the other hand, the Daubechies wavelets of order 1 (Haar wavelets)
are simple in applications due to their orthonormality and compact support, and they
are used while studying the dynamics of elastic–plastic beams [13, 18–20].
So far, we have briefly overviewed the application of wavelet concepts to studies
on nonlinear dynamics of the systems governed by ODEs and PDEs. Below,
we describe the state of part of the research on turbulent behavior exhibited by
continuous solid dynamic systems.
Mordant [24] applied an experimental method to monitor both temporal and
spatial evolutions of wave turbulence in a thin elastic plate. Various Fourier spectra
of the wave deformations have been analyzed. His research follows a series of
Turbulent Phenomena in Flexible Plates and Shells 51
earlier works devoted to a study of the Föppl–von Kármán model [8, 9, 23].
Despite qualitative good agreement with the kinetic weak turbulence theory, the
obtained energy spectrum has not been confirmed by a theoretical prediction. It has
been shown that the wave turbulence can also be exhibited by vibrating structural
members including beams, plates, and shells, where in spite of the standard Fourier
components various wavelet-type methods have been applied allowing for a deeper
insight into the wave turbulence in a solid [3–6]. Recently, the transition scenario
from periodic to wave turbulence regime in the forced vibrations of thin free-edge
circular and simply supported rectangular plates has been reported in [26, 27]. It
has been shown that only two bifurcations are required to achieve a spatiotemporal
chaotic dynamics. The first bifurcation is associated with a stability loss of the
directly excited plate mode in favor of the quasiperiodic dynamics. The second
bifurcation implies the stability loss and the wave turbulence appearance. It has
been confirmed numerically that the obtained power spectra are in good agreement
with the theoretical prediction given by the wave turbulence theory with respect to
the von Kármán equations for perfect undamped plates.
The paper is organized in the following manner. Section 2 is dedicated to the
analysis of chaotic/turbulent dynamics of rectangular periodically forced flexible
shells. Section 3 illustrates and discusses the application of various types of wavelets
to study nonlinear dynamics of flexible rectangular isotropic shells subjected to
the periodic shear load action. Section 4 presents turbulent dynamics of multilayer
flexible spherical panels putting emphasis on the simultaneous occurrence of both
temporal and spatial chaos. In Sect. 5 a turbulent behavior of two-layer cylindrical
shells coupled via boundary conditions and subjected to local periodic transversal
load action is reported. In Sect. 6 a weak hyper turbulent plate behavior is observed
while monitoring numerically the spectrum of Lyapunov exponents.
2 Chaotic Parametric Vibrations of Flexible Plates
2.1 Mathematical Model
A mathematical model of the flexible rectangular plate with constant stiffness

and density under the action of a periodic load (Fig. 1) is constructed using
the Kirchhoff–Love hypotheses and taking into account the nonlinear relations
between deformations and displacements in the von Kármán form. In the rectangular
coordinates the 3D plate space is defined as ˝ D fx1 ; x2 ; x3 j.x1 ; x2 / 2 Œ0I aŒ0I b,
x3 2 ŒhI hg, 0 t < 1. In initial time interval t 2 Œ0I 1 we introduce a small
static load (its lack defines the governing differential equations as homogenous
ones).
We study the following non-dimensional PDEs governing dynamics of the
shallow shells [29]:
Fig. 1 Plate computational scheme
1 4 @2 w @w
r w rk2 F L.w; F / 2 " q.x1 ; x2 ; t / D 0;
12.1 /
2 @t @t (1)
1
r4 F C rk2 w C L.w; w/ D 0;
2
where r4 , L.w; F / and rk2 are the known nonlinear operators, whereas w and F are
the functions of deflection and stress, respectively.
The following non-dimensional parameters are introduced: D a=b, x1 D ax 1 ,
x2 D bx 2 , kx1 D a2 =Rx1 .2h/, kx2 D b 2 =Rx2 .2h/ are the non-dimensional shell
curvature parameters regarding x1 and x2 , respectively; w D 2hw is the deflection;
4
F D E.2h/3 F is the stress function; t D t0 t is the time; q D E.2h/
a2 b 2
q is the external
load; " D .2h/" is the damping coefficient; and P D E.2h/ P is the external
3
longitudinal load. Bars over non-dimensional quantities in the governing equations

have already been omitted. The following notation is introduced: a; b are the plate
plane dimensions in x1 and x2 directions, respectively; is the Poisson’s coefficient.
Equation (1) is supplemented by the following boundary conditions [16]:
@2 w @2 F
w D 0I D 0I F D 0I D px 2 for x1 D 0I 1;
@x12 @x12
@2 w @2 F
w D 0I D 0I F D 0I D px 1 for x2 D 0I 1; (2)
@x22 @x22
and the following initial conditions:
@w
w.x1 ; x2 /jtD0 D 0; D 0: (3)
@t
2.2 Method of Solution and Results
Our mechanical object is studied keeping fixed the following parameters: D

1, Poisson’s coefficient D 0:3. We apply the following longitudinal load
px1 D px2 D p0 sin.!p t /. After the application of FDM with approximation
.h2 / regarding spatial coordinates the differential problem (1)–(3) is solved by the
Runge–Kutta method of the fourth order. In addition, on each time step we need
to solve a large system of linear algebraic equations (AEs) regarding the stress
function. Time step is yielded through the Runge principle. The number of partitions
of spatial coordinates is n D 14 (the number of partitions choice and convergence
of the obtained numerical results were discussed by Awrejcewicz et al. [6]).
In this work we take the amplitude and frequency of exciting longitudinal load
acting on the shell perimeter. Our aim was to construct charts displaying the
system vibration regimes in the plane of parameters Œ!p ; p0 with the resolution
of 300 300. In order to construct each of the charts 90,000 differential problems
were solved. Each of the mentioned problems required the analysis of signals (time
histories), phase and modal portraits, Poincaré cross-sections and maps, Fourier
and wavelet frequency power spectra, autocorrelation functions, and signs of the
Lyapunov exponents. Figure 2 gives charts of the shell vibration regimes depending
on the geometric shell parameters. In the first subharmonic regime zones of Hopf
bifurcations are wide. It is seen that an increase of the geometric parameters implies
the increase of bifurcation zones and chaotic zones and a decrease of the zones of
periodic vibrations. Besides, one may observe that in the case of shell curvatures
kx1 D kx2 D 24 a transition from periodic into chaotic/turbulent vibrations appears
suddenly without any other transitional zones. The following general remarks are
based on the computational results. Small exciting load amplitudes generate damped
vibrations. Low values of the applied frequencies ! 2 of subharmonic zones
of vibrations (excitation frequency is doubled in comparison to the shell vibration
frequency) are mixed with the zones of periodic vibrations. An increase of the
excitation frequency implies an extension of these zones into higher frequencies,
and they are interlaced with rather large chaotic/turbulent zones.
While constructing the shell vibration charts chaotic dynamics occurred already
after the second or even first Hopf bifurcation. Additionally, the wavelet spectra
imply that these bifurcations appear even at fixed amplitude and frequency of the
excitation for t t0 (they depend only on time giving rise to intermittency). In
the numerical example with parameters kx1 D 12, kx2 D 0 (cylindrical panel),
!p D 8:4, p0 D 8:5, t 2 Œ0I 300, using both Fourier and wavelets spectra, it
is evidently demonstrated how power spectrum essentially changes in time. For
instance, at t < 150, the first Hopf bifurcation takes place .!1 D !p =2/, and
at t > 150 the second Hopf bifurcation appears .!2 D !p =4/ (Fig. 3). A more
detailed analysis is provided by the wavelet analysis which allows us to monitor
local phenomena of the studied signal.
It should be emphasized that the charts reported in this section (Fig. 2) allow us to
control the investigated continuous mechanical systems. The choice of the control
Fig. 2 Charts of the shell vibration regimes versus geometric (shell curvature) parameters
parameters should be made in a way to keep the system within a safe (periodic)
zone. Otherwise, transition into the chaotic zone implies the loss of system stability
and its catastrophe.
Fourier spectrum (all time) Wavelet spectrum (all time)

0 10
S ωp
ω1 8
−2
ω 6
ω2 ω3
−4
4
−6 2
ω 50 100 150 200 250

−8
0 2 4 6 8 t
Fourier spectrum t < 150 Fourier spectrum t > 150
0 0
S ωp S ωp
ω1 ω2 ω1
−2 ω3
−4 −5
−6
ω ω
−8 −10
0 2 4 6 8 0 2 4 6 8
Fig. 3 Fourier and wavelet spectra kx1 D 12, kx2 D 0, !p D 8:4, p0 D 8:5
3 Wavelet Versus Fourier Analysis
3.1 Governing Equations and Results
It is well known that the data provided by numerical experiments are presented in
the time domain. In other words, we take time as an independent coordinate and
amplitude as a dependent coordinate, and the studied signal is analyzed through
its amplitude–time representation. However, in order to understand profoundly
nonlinear continuous systems subjected to various types of load actions and in order
to fully understand the occurring dynamics, we have to apply the information hidden
in the spectral signal characteristics. Predominantly, the Fourier transformation has
been applied for a long time. However, it has been demonstrated that the Fourier
analysis (FFT) is reliable only for the study of frequency components of stationary
processes, i.e., the processes which through the whole period of investigation keep
constant frequency components in time. It happens that, in particular, the dynamics
of continuous mechanical systems may exhibit quite complicated output and their
frequency characteristics may change strongly in time. This is why in spite of the
standard Fourier approach the wavelet analysis is applied, allowing us to detect and
understand many interesting nonlinear phenomena of the mentioned mechanical
systems.
One of the first important tasks to be solved is associated with the choice of
a suitable wavelet, which fits well with the stated problem. In order to solve this
query we have studied the non-stationary signal (Fig. 4) obtained via the numerical
experiment as an output of the mathematical model of the rectangular flexible
isotropic shell subjected to the periodic shear load acting in the shell volume unit.
The mathematical model is as follows [29]:
1 4 @2 w @w @2 w
2
r w L.w; F /C 2 C" q.x1 ; x2 ; t /C2S D0;
12.1 / @t @t @x1 @x2 (4)
1
r4 F C L.w; w/ D 0;
2
4 4 4 2 2 2 2 2 2
where r4 D 12 @x
@
4 C
2 @
@x24
C 2 @x@2 @x 2 , L.w; F / D @@xw2 @@xF2 C @@xw2 @@xF2 2 @x@1 @x
w @ F
2 @x1 @x2
1 1 2 1 2 2 1
are the known nonlinear operators, whereas w and F stand for the plate deflection
and Airy’s function, respectively.
System (4) is reduced to the non-dimensional form using the following non-
dimensional parameters: D a=b; x1 D ax 1 , x2 D bx 2 are the non-dimensional
parameters regarding x1 and x2 , respectively; w D 2hw is the deflection; F D
4
E.2h/3 F is Airy’s function; t D t0 t is the time; q D E.2h/ a2 b 2
q is the external load;
3
" D .2h/" is the damping coefficient of the surrounding medium; and S D E.2h/ ab
S
is the external shear load. In the equations bars over the non-dimensional quantities
have already been omitted. The following notation is introduced: a; b are the shell
dimensions regarding x1 and x2 , respectively; is Poisson’s coefficient. Zero-
value initial conditions and the following boundary value conditions are attached
to system (4):
@2 w @2 F
w D 0I D 0I F D 0I D 0 for x1 D 0I 1;
@x12 @x12
(5)
@2 w @2 F
w D 0I D 0I F D 0I D 0 for x2 D 0I 1:
@x22 @x22
The external harmonic shear load has the form S D s0 sin !p t . PDEs governing
dynamics of our investigated shell are reduced to the ODEs via the FDM with
the approximation O.h2 / regarding spatial coordinates. Next, ODEs are solved via
the fourth-order Runge–Kutta method, and additionally on each of the iterations
a large system of linear algebraic equations should be solved with respect to the
stress (Airy’s) function. Time integration step has been chosen using the Runge
rule. The partition number of spatial coordinates is n D 14 while applying FDM.
Validity and reliability of the obtained results regarding the number of partitions
2
have been discussed by Awrejcewicz et al. [6]. The term 2S @x@1 @x w
2
introduced in
the governing equations exhibits the action of shear stresses located in shell middle
plane and it influences essentially nonlinear dynamics of the investigated shell. The
numerical simulation indicates that the output signal (shell vibrations) may change
in time rapidly. We apply this signal to choose a methodology suitable for the
investigation of non-stationary processes, and in addition we illustrate advantages
and disadvantages of the standard Fourier approach versus the wavelet transform
procedure. The studied signal has been obtained, analyzing the system with the
following fixed parameters: s0 D 8:4 and !p D 26. We show that frequency
characteristics taken in different time intervals essentially differ from each other.
It should be emphasized that the system stability loss is due not only to the change
of chosen control parameters but is caused even by keeping all of them fixed.
The system time evolution may lead to shell instability. In the first time interval
t 2 Œ50I 56 the shell exhibits two quasiperiodic frequency vibrations. Instead of
the vanished excitation frequency, two independent frequencies appear. A further
long-time evolution of chaotic vibrations with the exhibition of a few dependent
frequencies is observed. In the Fourier spectrum the excitation frequency is not
visible. The last studied time interval corresponds to periodic vibrations, which is
also in agreement with the Fourier power spectrum (Fig. 4).
Fourier spectrum Fourier spectrum Fourier spectrum

50 < t < 56 130 < t < 225 230 < t < 289
(quasi-periodic vibrations) (chaotic vibrations) (periodic vibrations)
0 0
S,db 0 ωp
S,db S,db
ωp
−5 −5 −5
ωp ω ω
−10 −10 −10
0 10 20 0 10 20 0 10 20
Signal
1 w
0.5
−0.5
t
−1
60 80 100 120 140 160 180 200 220 240 260 280
Fig. 4 Fourier spectra and signal

Wavelet spectrum Wavelet spectrum Wavelet spectrum

Meyr Daubechies 16 complex Morlet
25 25 25
20 20 20
ω15 ω 15 ω 15
10 10 10
5 5 5
100 150 200 250 300 100 150 200 250 300 100 150 200 250 300
Wavelet spectrum Wavelet spectrum Wavelet spectrum

Morlet Gauss 16 Gauss 16 complex
25 25 25
20 20 20
ω 15 ω 15 ω 15
10 10 10
5 5 5
100 150 200 250 300 100 150 200 250 300 100 150 200 250 300
Fig. 5 Wavelets spectra
We constructed wavelet spectra to study the mentioned signal. We applied the

following wavelets: Haar, Shannon–Kotelnikov, Meyer, Daubechies from db2 up to
db 16, Coiflets and Symlet, as well as the Morlet and Gauss (real and imaginary),
on the basis of the derivatives from 2 to 16. The Haar and Shannon–Kotelnikov
wavelets appeared to be unsuitable for the analysis of shell structures. The first
one is badly localized regarding frequency, whereas the second one, contrary to
the previous wavelets, is badly localized in time. On the other hand, the analysis
based on the Daubechies wavelets, as well as Coiflets and Symlet wavelets, implies
an increase of the frequency resolution, assuming that the filter properties are
increased. Neglecting differences in the wavelet forms and the associated filters, the
wavelet spectra obtained through the Daubechies wavelets as well as Coiflets and
Symlet wavelets are practically identical. However, their localization with respect
to frequency is not suitable for the analysis of continuous systems dynamics. In the
case of the Gauss functions, an increase of their derivative order implies an increase
of the frequency resolution.
Figure 5 gives results associated with the application of different wavelets
(Meyer, Morlet, complex Morlet, real and complex Gauss with 16 derivative orders,
Daubechies) to analyze nonlinear shell vibrations.
One may conclude from Fig. 5 that the localization regarding frequency increases
with an increase of the number of the wavelet zero moments.
This section has shown that the complex Morlet and Gauss wavelets have better
localization with respect to frequency than their real analogues, but time localization
is better exhibited for the real wavelets. Therefore, one may apply either real or
complex Morlet and Gauss wavelets of the order bigger than 16 while studying
plates/shells dynamics.
4 Turbulence of Flexible Multilayer Shells with Gaps

Between Layers
4.1 Problem Statement
Significant results have been achieved recently during the analysis of chaotic
dynamics of simple dynamic systems yielded by truncated original ODEs or the
recurrence equations (maps). It should be emphasized, however, that in these
simple systems we may observe only temporal chaos. This may be helpful while
analyzing the beginning of turbulence, when velocity field starts with the evolution
of chaotic velocities, but keeping all the time the well self-organized space. In this
work we illustrate that both chaotic phenomena, i.e., temporal and spatial chaos,
exist simultaneously. Spatiotemporal chaos (weak turbulence) has been detected
while analyzing the Ginzburg–Landau equation [11], but as far as we know this
phenomenon has not been detected so far with respect to the shell structural
members. In this section nonlinear multilayer shells are studied, where in spite of
the geometric nonlinearitywe also take into account a design-type nonlinearity. It
means that on each time step we need to solve a contact problem yielded by the
interaction of shell members.
We apply a classical nonlinear theory to study the two-layer spherical flexible
isotropic elastic rectangular shell with constant stiffness and density and subjected
to the action of longitudinal time periodic load. The load is applied only to the upper
shell layer. The layers exhibit hybrid dynamics: they are either contacting between
each other without friction or they vibrate without a contact. The occurrence of stick
zones is rather unlikely because the contact pressure is small. Contact conditions
between the layers may depend on the coordinates and may include all kinds of the
one-sided contact. Welding conditions regarding normal and tangential directions
are not considered in this work. The behavior of layers agrees with the Kármán-
Vlasov theory and is the same for all layers. The function describing the contact
pressure is removed from the number of unknowns. Since the order of the governing
differential equations is a product of the number of layers and the system equations
order, we deal with the Bolotin–Novikov model (Fig. 6). In this work we analyze
two-layer panels:
˝1 D fx1 ; y1 ; z1 j.x1 ; y1 / 2 Œ0I a Œ0I b; z1 2 Œh1 ; h1 g ;
˝2 D fx2 ; y2 ; z2 j.x2 ; y2 / 2 Œ0I a Œ0I b; z2 2 Œh2 ; h2 g ; 0 t < 1:

Fig. 6 Computational scheme of a two-layer panel
We study the following equations of the theory of shallow shells presented in the
non-dimensional form [15, 29]:
@2 wm @2 wm
r 4 wm C L.wm ; Fm / C r 2 Fm C q1 C pxm 2
C p y m 2
˙
@ym @xm
2 (6)
@ wm @wm
˙K.w1 hk w2 / D 2
C "1 ;
@t @t
r Fm D 2 L.wm ; wm / r wm ;
4 1 2
where
@2 wm @2 Fm @2 wm @2 Fm @2 wm @2 Fm
L.wm ; Fm / D 2 @y 2
C 2 @y 2
2 ;
@xm m @xm m @xm @ym @xm @ym
1 @4 @4 @4
r 4 D 2 4 C 2 4 C 2 2 2 ; (7)
@xm @ym @xm @ym
2 2 2 2
@ wm @ wm @ wm
L.wm ; wm / D 2 @y 2
; i D 1; 2:
@xm m @xm @ym
Here D 12 Œ1 C sign.w1 hk w2 /, wm and Fm are the functions of deflection and

stresses, respectively, where the layers m D 1; 2 and K D 17; 500 are the stiffness
coefficients of the transversal contact zone. If w1 > w2 C hk , the contact between
layers occurs, which means that D 1 and otherwise D 0, w1 ; w2 deflection of
the upper and lower panel, respectively.
The system of (6) is transformed to the non-dimensional form through the
following relations:
a2 b2
xm D ax m ; ym D by m ; k x m D kx m ; k y m D ky m ;
hm hm
1 1 Em h4 Em h3m
kx m D ; ky m D ; qm D q m 2 2m ; pxm D p xm ; (8)
rxm rym a b b2
3 r
Em hm ab m a
py m D p ym ; m D ; 1 D ;
a2 hm Em gm b
where a; b are the sides of a cylindrical panel regarding xm and ym , respectively; hm

is the shell thickness; gm is the acceleration due to gravity; m D m hm , where m
is the specific material density; rxm , rym is the radius of the middle surface curvature
regarding xm and ym , respectively. In addition, t is the time; "m is the damping
coefficient of the surrounding medium; is Poisson’s coefficient (for an isotropic
material D 0:3); Em is the elasticity modulus; pxm .t /, pym .t / are the longitudinal
loads; and qm .x; y; t / is the transversal load. Further, bars over non-dimensional
quantities are omitted.
Equations (6) are supplemented with the following boundary conditions:
@2 wm @2 Fm
wm D 0; 2
D 0; Fm D 0; 2
D 0 for xm D 0I 1;
@xm @xm
(9)
@2 wm @2 Fm
wm D 0; 2
D 0; Fm D 0; 2
D 0 for ym D 0I 1;
@ym @ym
and the initial conditions:
@wm
wm .xm ; ym /jtD0 D '1 .xm ; ym /; D '2 .xm ; ym /: (10)
@t
In order to reduce PDEs (6) to a system with lumped parameters, the FDM is applied
with the approximation O.h2 / with respect to spatial variables xm and ym .
The Cauchy problem is solved by the fourth-order Runge–Kutta method. On
each of the time steps we need to solve a large system of linear algebraic equations
(yielded by the second equation of the governing ones with respect to the stress
function) using the inverse matrix method. Time step follows from the Runge
principle.
4.2 Numerical Experiment
We consider nonlinear dynamics of a two-layer cylindrical panel with curvatures

kxm D 12, kym D 0. First, the panel is subjected to the action of time periodic
load px1 .t / D py1 .t / D p1 sin.!p t / with frequency !p D 8:4, excitation
amplitude p1 D 7:74, and shell clearance hk D 0:5. The numerically obtained time
characteristics include the signal wm .t /, wavelet spectrum, and Fourier spectrum,
whereas the spatial characteristics include the curves of equal deflections and
contact pressure (see Fig. 7). While analyzing signal (Fig. 7m) and applying the
wavelet analysis (Fig. 7b) three phases are clearly distinguished. After a transitional
state the signal becomes stationary, i.e., periodic (Fig. 7c, d). Owing to the Fourier
analysis carried out in time interval t 2 .30I 60/ (Fig. 7f), first periodic vibrations
are observed, and then the first Hopf bifurcation occurs for t 2 .100I 250/, which
is reported in Fig. 7g. As it can be seen in the drawings, in the case of stationary
vibrations (periodic vibrations and vibrations with the frequency after the Hopf
bifurcation), the contact pressure (Fig. 7i, k) and isoclines regarding deflections
(Fig. 7j, l) are symmetric with respect to the axes. Next, the system jumps into a
turbulent regime.
Stiff stability loss is exhibited by the system, and the deflection amplitude
suddenly increases (Fig. 7e); both contact pressure (Fig. 7m) and deflection isoclines
(Fig. 7n) (spatial characteristics) become nonsymmetric. Power spectrum (Fig. 7h),
which exhibits time properties of vibrations, is a broad band implying the occurrence
of chaos for t 2 .250I 500/.
The so far described scenario from regularity to chaos for our investigated
structural member has been reported for the first time. It allows us to conclude that
both temporal and spatial chaos appears simultaneously, yielding the shell turbulent
behavior.
5 Turbulent Behavior of Two-Layer Cylindrical Shells

Coupled via Boundary Conditions and Subjected
to Local Periodic Transversal Load Action
5.1 Problem Formulation
In this section we study the mathematical model of a two-layer cylindrical shell

(with clearance between both shells) having constant stiffness and density and
subjected to a periodic transversal load action. The derived mathematical model
allows us to study nonlinear dynamics of both structural members with regard
to external load and internal interaction between the two layers, as well as
the force action coming from the inside of the second cylinder. The developed
mathematical model includes geometric nonlinearity of both cylinders and their
contact interactions. Differential equations governing the dynamics of both shells
are solved by the Bubnov–Galerkin higher-order approximation method, whereas
the obtained Cauchy problem is solved using the fourth-order Runge–Kutta method.
Convergence of the Bubnov–Galerkin method versus the number of approximating
series terms is considered. Namely, we solve the problem of two embedded
cylindrical shells including their interaction as systems with an infinite number of
degrees of freedom.
We consider cylindrical shells as closed 3D objects embedded into space R3 with
the curvilinear coordinates x 1 ; y 1 ; z1 and x 2 ; y 2 ; z2 introduced in the following way:
in the shell body we fix the middle surface for z1 D 0 and z2 D 0; axes 0x1 , 0x2 and
0y1 , 0y2 are directed along the main curvatures of this middle surface, whereas axes
0z1 , 0z2 go into the curvature centers (Fig. 8). In the given system of coordinates the
following 3D subspaces ˝1 and ˝2 are defined:
a b c
10
9 3
5 wm (0.5, 0.5, t)
8
7
6
5 0
4
0 3
2
1 −3
−3 t
30 100 200 300 400 500 100 200 300 400 500 50 51 52 53
d e f
3 3 0
S,db
wp
0 0
−5
−3 −3 w
200 201 202 203 350 351 352 353 0 5 10
g h i
0 0
S,db S,db wp
wp
wp x1
−5 2
−5
w w y1
−10
0 5 10 0 5 10
j k l
x1
y1
m n
x1
y1
Fig. 7 System characteristics. (a) Signal spectrum. (b) Wavelet. (c) Periodic vibrations
wm .0:5I 0:5I t /. (d) Hopf bifurcation wm .0:5I 0:5I t /. (e) Chaotic vibrations wm .0:5I 0:5I t /. (f)
Power spectrum t 2 .30I 60/. (g) Power spectrum t 2 .100I 250/. (h) Power spectrum
t 2 .250I 500/. (i) Contact pressure. (j) Deflection isoclines w1 .x1 I y1 I 35/. (k) Contact
pressure. (l) Deflection isoclines w1 .x1 I y1 I 150/. (m) Contact pressure. (n) Deflection isoclines
w1 .x1 I y1 I 500/
Fig. 8 Computational scheme
( " #)
h1 h1
˝1 D x 1 ; y 1 ; z1 =.x 1 ; y 1 ; z1 / 2 Œ0I a Œ0I b ; ;
2 2
( " #)
h2 h2
˝2 D x 2 ; y 2 ; z2 =.x 2 ; y 2 ; z2 / 2 Œ0I a Œ0I b ; ;
2 2
and bars refer to dimensional quantities here.

The PDEs given below govern nonlinear dynamics of the shell system presented
in Fig. 8, and they are derived taking into account the following hypotheses:
two-layer shells are made from an isotropic, homogeneous, and elastic material
satisfying the Kirchhoff–Love hypothesis of each of the two layers:
1 @2 F1 @2 w1 @w1
2
w1 L.F 1 ; w1 / k y C C "1
12.1 /
2 1
@x1 2 @t 2 @t
q1 .t / C K.w1 hk w2 / D 0;
@2 w1
2 F1 C L.w1 ; w1 / C ky1 D 0;
@x12
2 (11)
1 @ F2 @2 w2 @w2
2
w2 L.F 2 ; w2 / k y C C "2
12.1 /
2 2
@x2 2 @t 2 @t
q2 .t / C K.w1 hk w2 / D 0;
@2 w2
2 F2 C L.w2 ; w2 / C ky2 D 0;
@x22
where
1 @4 2 @
4
@4
2 D C C 2 ;
2 @xi4 @yi4 @xi2 @yi2
@2 wi @2 Fi @2 wi @2 Fi @2 wi @2 Fi
L.wi ; Fi / D 2 2
C 2 2
2 ;
@xi @yi @yi @xi @xi @yi @xi @yi
2
@2 wi @2 wi @2 wi
L.wi ; wi / D ; i D 1; 2:
@xi2 @yi2 @xi @yi
Non-dimensional quantities are as follows:

2
wi D hi wi ; F i D E i hi Fi ; t D t ; "i D "i = ;
4
1 Eh
x i D axi ; y i D r i yi ; k y D ; qDq 2
;
r a2 b (12)
s y i
ar i i a
D ; D ;
hi E i gi ri
where wi stands for deflection; Fi is the stress function; t is the time; "i is the
damping coefficient; a and ri D ryi are the length and radius of the closed
cylindrical shell regarding xi and yi , respectively; hi is the shell thickness; gi is
the Earth acceleration; i denotes the specific material density; and ryi is the radius
of the middle surface. In addition, is Poisson’s coefficient of the isotropic material;
Ei is the Young modulus; and q.xi ; yi ; t / is the external transversal load.
The following boundary and initial conditions are attached to (11)
@2 wi @2 Fi
wi D 0; D 0; F i D 0; D 0 for x D 0I 1;
@xi2 @xi2
@2 wi
wi D gi .xi ; yi ; t /; D ri .xi ; yi ; t /; Fi D ui .xi ; yi ; t /; (13)
@yi2
@2 Fi
D zi .xi ; yi ; t / for yi D 0I 2;
@yi2
@wi
wi .xi ; yi /jtD0 D 0; D 0; i D 1; 2: (14)
@t
We apply the Bubnov–Galerkin method to reduce PDEs to ODEs, and the latter ones
are solved via the fourth-order Runge–Kutta method. The numerical computation
step is given by the Runge rule.
Fig. 9 Computational scheme
5.2 Computational Results
Reliability and validity of the obtained results are illustrated and discussed else-
where (Fig. 12); we consider vibrations of one-layer cylindrical shell under the
action of the local transversal load regarding x1 and y1 (Fig. 9).
The maximum shell deflection amplitude versus load amplitude wmax .q0 / is
shown in Fig. 8, where the so-called vibration scales are also included. Fourier
spectra for each of the control parameters have been computed in order to construct
the scales. Below, we analyze the influence of both the loading surface and
coordinates of the load action on the vibration regimes of the studied cylindrical
shells.
In Fig. 10 functions wmax .q0 / and scales of dynamic regimes for various cases
of loading are applied. We apply the square loading surface, excluding the case
reported in Fig. 10e, where the load along y1 is distributed around a disc. Analysis
of the obtained results implies that an increase of the loading surface yields an
increase of the chaotic vibrations. In addition, location of the load has also an
essential influence on the properties of chaotic vibrations. Longitudinal shift of the
concentrated load into the shell center is accompanied by an increase of chaotic
vibration zones. However, if the external load is applied to the shell edges (Fig. 10a),
there are chaotic vibrations, since the so-called boundary layer effect plays a crucial
role here.
We study the spatiotemporal (turbulence) characteristics of the shell behavior
in two-frequency quasiperiodic and chaotic regimes in the case of the action of
transversal local load x1 2 .0:2I 0:4/. Figures 11 and 12 show wave forms versus
wmax wmax wmax

0.0 0.5 1.0
0.0 0.5 1.0

0.6 12.5 14
0.0 0.5 1.0

q0 q0 q0
0 0 0
0 0.65 1.29 0 0.33 0.65 0 0.3 0.6
a b c
wmax wmax
wmax 25
25
28 0.0 0.5 1.0
0.0 0.5 1.0 0.0 0.5 1.0
12.5 12.5
14
q0 q0 q0
0 0
0
0 0.23 0.45
0.05 0.15 0.25 f 0 0.35 0.7
d e
Notation
Independent frequencies
Periodic vibrations
and their linear combinations
Hopf bifurcations Chaotic vibrations
Fig. 10 Functions wmax .q0 /, scales of dynamic regimes and load action schemes. (a) x1 2
.0:0I 0:2/. (b) x1 2 .0:2I 0:4/. (c) x1 2 .0:4I 0:6/. (d) x1 2 .0:3I 0:7/. (e) x1 2 .0:1I 0:5/. (f)
x1 2 .0:4I 0:6/, y1 2 .0I 2/
time w1 .x1 I y1 I t /, 0 x1 1, 0 y1 2, signal w.0:3I 0:0I t / in the loading

center, phase portrait w1 .wP 1 /, Poincaré map w1t .w1tCT /, and the frequency power
spectrum S.!/. In the case of quasiperiodic vibrations with two frequencies, a1
and !p , all shell points move in a regular way, and the points under the action of
the load move inside the shell curvature, whereas free points regarding the load
action move up and vice versa. Maximum deflections along the circle direction are
located in the external loading zones. In the case of turbulent vibrations a complex
movement of all shell points is observed. Maximum deflections are distributed onto
the whole shell surface, whereas shell points under the load and free of the load
action may move in an unpredictable manner (Fig. 12e, t D 3; 7; 10, and t D 2; 4).
Step changes of the deflection signs are exhibited, which has not been observed in
the case of quasiperiodic vibrations. In the latter case the sign changes appeared in
a smooth way.
The so far reported study allows for a conclusion that the spatial chaos and
temporal chaos appear simultaneously, which validates our earlier observations on
the shell transition into turbulent shell vibrations (see Sect. 4).
a w1 b w1
2 2
1 5
2
0 0
3 4
−2 t −2 w1
16 17 18 −50 0 50
c w1t+T d
S
1.272 0
wp
1.269 −5
b1
a1
w1t w
1.266 −10
1.266 1.269 1.272 0 10 20 30
t=1 t=2 t=3
t=4 t=5
Fig. 11 Spatiotemporal characteristics of the shell quasiperiodic dynamics. (a) w1 .0:3I 0:0I t /.
P 1 /. .c/ w1t .w1t CT /. .d/ S.!/. .e/ w1 .x1 I y1 I t /; 0 x1 1; 0 y1 2
.b/ w1 .w
a
20
w1 b w1
20
3 7 8
5 10
1
5 5
9
2 4 6
−10 t −10
w1
16 17 18 −500 0 500
c w1t+T d S
20 2
wp
5 −2
−10
w1t −6 w
−10 5 20 0 10 20 30
e
t=1 t=2 t=3
t=4 t=5 t=6
t=7 t=8 t=9 t = 10
Fig. 12 Spatiotemporal characteristics of the shell turbulent dynamics. (a) w1 .0:3I 0:0I t /. (b)
P 1 /. (c) w1t .w1t CT /. (d) S.!/. (e) w1 .x1 I y1 I t /; 0 x1 1; 0 y1 2
w 1 .w
6 Analysis of the Lyapunov Exponents for Nonlinear

Dynamics of Plates and Shells
6.1 Governing Equations
This section aims at a description of a novel method for the estimation of Lyapunov
exponents using neural networks. In the attempts described earlier, many approaches
and algorithms have been proposed to compute the Lyapunov exponents, but in
general almost all of them apply the algorithm proposed already by Benettin [7].
However, while studying nonlinear dynamics of structural members (beams, plates,
and shells) it is of great importance to estimate the spectrum of Lyapunov exponents.
We also illustrate and demonstrate the efficiency of the neural network approach
to study a flexible plate with an infinite length. The governing equations are within
the Kirchhoff hypothesis and they have the following non-dimensional form [29]:
@2 u @2 u
C L 3 .w; w/ D0
@x 2 @t

2
(15)
4
1 1 @w @2 w @w
C L 1 .u; w/ C L 2 .w; w/ C q " D 0;
2 12 @x 4 @t 2 @t
@2 u @w @u @2 w
L1 .u; w/ D C ;
@x 2 @x @x @x 2
2
3 @w @2 w
L2 .w; w/ D ;
2 @x @x 2
@w @2 w
L3 .w; w/ D ;
@x @x 2
where L1 .u; w/, L2 .w; w/, L3 .w; w/ are the nonlinear operators; w.x; t / is the
plate element bending in normal direction; u.x; t / is the plate element longitudinal
displacement; "is the dissipation coefficient; E is the Young modulus; h is the height
of the transversal panel cross section; is the specific plate material gravity; g is the
acceleration due to gravity; t is the time; and q D q0 sin.!p t / is the external load.
The non-dimensional parameters are as follows:
a w ua t
D ; w D ; u D 2 ; x D xa ; t D ;
h hs h
a Eg a qa4 (16)
D ; pD ; "D ; qD 4 ;
p p hE
and bars over the non-dimensional quantities have already been omitted in (15).
We demonstrate how to determine four first Lyapunov exponents applying pinned
boundary conditions:
w.0; t / D w.a; t / D u.0; t / D u.a; t / D w00xx .0; t / D w00xx .a; t / D 0;

(17)
w.0; t / D w.1; t / D u.0; t / D u.1; t / D w00xx .0; t / D w00xx .1; t / D 0;
and the following initial conditions
w.x; 0/ D w.x;
P 0/ D u.x; 0/ D uP .x; 0/: (18)
The boundary value problem (15), (17), (18) is reduced to the Cauchy problem via
FDM of the second-order accuracy. The obtained ODEs are solved by the Runge–
Kutta method of the fourth and sixth orders. Validity and reliability of the obtained
numerical results are confirmed by the FEM results. The initial problem of infinite
dimension is substituted by that of finite dimension via partition of the interval x 2
Œ0; 1 into 120 parts.
One of the ways to compute the spectrum of Lyapunov exponents is the neural
network approach based on the generalized Benettin algorithm. It includes the
following steps: (i) choice of the appropriate time delay via tests; (ii) computation
of an embedding space dimension; (iii) reconstruction of pseudo-phase trajectories
using the method of time delays; (iv) neural network approximation; (v) teaching
of neural networks to compute successive iteration vectors; (vi) computation of the
spectrum through the generalized Benettin algorithm with the help of the neural
networks approach.
We apply the neural network with the following properties: it is an analogue
network regarding the input data (information is delivered through real numbers);
it is self-organized with respect to its teaching aspects (output space of solutions is
defined only through the input data); it belongs to the neural networks of straight
signal distributions (all neural network couplings come from the input neurons and
go to the output neurons); the neural network has dynamic couplings (control and
improvement of synaptic couplings is carried out during the neural network learning
process .d W =dt ¤ 0/, where W stands for the net weight coefficients).
6.2 Generalization of the Benettin Algorithm
Let point x0 belong to attractor A of a dynamic system. The trajectory of evolution

of point x0 is called the unperturbed trajectory. We choose the positive quantity
" essentially lesser than the attractor dimension. Next, we choose an arbitrary
(perturbed) point xQ 0 in a way to satisfy the relation kxQ 0 x0 k D ". Then, we monitor
the evolution of the chosen x0 and xQ 0 in time interval T , and the corresponding new
points obtained after that time interval are denoted as x1 and xQ 1 , respectively. Vector
x1 D xQ 1 x1 is called the perturbation vector. We are ready to estimate
1 kx1 k
Q 1 D ln : (19)
T "
Time interval T is chosen in a way to keep the perturbation amplitude lesser than
the linear dimensions of the space nonhomogeneity as well as lesser than the
attractor dimension. We consider the unit normalized perturbation vector x10 D
"x1 =kx1 k and the corresponding new perturbation point xQ0 1 D x1 C x10 . We
extend the so far described approach by using points x1 and xQ 1 instead of x0 and xQ 0 ,
respectively. We repeat the described procedure M times, and we may estimate as
an average arithmetic quantity Q i of those obtained on each computation step. The
proposed approach has been tested using the standard classical examples including
that of the Henon map [14], the Lorenz system [22], and the logistic map [28].
We consider vibrations of our mechanical object with the following fixed
parameters: D 50, " D 1, !p D 7, q D q0 sin.!p t /, and for the following
amplitudes of the harmonic excitation: q0 D 0:125 103 ; 5 103 ; 7 103 . In order
to study chaotic dynamics of flexible plates, we need to monitor and analyze the
following output characteristics: time histories (a), phase (c), and modal portraits;
phase portraits yielded by the neural networks approach (d); Fourier power spectra
(b); wavelet spectra, Poincaré sections (e); spectra of Lyapunov exponents, where d
stands for the fractional part of dimension and h is the Kolmogorov–Sinai entropy
(f); and autocorrelation functions (some of them are reported in Fig. 13). Analysis
of the obtained results implies that for q0 D 0:125 103 periodic vibrations appear,
whereas for 5 103 a spatiotemporal chaos is exhibited, and for 7 103 the hyper-
spatio-temporal chaos occurs, since two of the Lyapunov exponents spectrum are
positive.
The obtained validated spectra of the Lyapunov exponents allow for the estima-
tion of Kaplan–Yorke dimension, Sinai–Kolmogorov entropy, and velocity of the
phase-space compression.
We studied numerically two-dimensional solid objects represented by shells and

plates putting emphasis on high-dimensional spatiotemporal chaos versus the weak
wave turbulence framework. Although all analyzed nonlinear PDEs were reduced
to ordinary differential equations and possibly algebraic equations, all of the studied
mechanical objects were modeled as infinite dimensional systems using the robust
discretization methods and then by a careful checking of validation and reliability
of the results. Next, we briefly summarized the obtained novel results.
In Sect. 2 spatiotemporal chaotic parametric vibrations of the flexible rectangular
plate with constant stiffness and density periodically loaded were investigated.
Fig. 13 Plate characteristics

Its 2D model was constructed using the Kirchhoff–Love hypotheses and taking into
account nonlinear relations between deformations and displacements in the form
proposed by von Kármán. The governing PDEs were reduced to ODEs and AEs via
application of the FDM with approximation O.h2 / regarding spatial coordinates.
Among others a sudden transition from periodic (laminar) into chaotic (turbulent)
shell dynamics was reported without any other transitional zones. It was also shown
how low values of the applied frequencies .! 2/ of subharmonic zones were
interlaced with the zones of periodic vibrations. The excitation frequency increase
implies extension of these zones into higher frequency spectrum and finally they
started mixing with large chaotic zones strengthening the shell spatiotemporal
chaotic dynamics. We reported also another transition when chaotic dynamics
occurred already after the second or even first Hopf bifurcation.
Section 3 deals with the irregular dynamics of the rectangular flexible isotropic
shell subjected to the periodic shear load. The governing PDEs were reduced
to ODEs via the FDM with approximation O.h2 / regarding spatial coordinates
and then they were solved using the fourth-order Runge–Kutta method. The
reported result had two important aspects. First, the standard Fourier approach
(being often used) might lead even to erroneous results. The reason was that the
independent variable (time) might play a role as a parameter. Namely, it was
shown that in the first time interval the shell exhibited two-frequency quasiperiodic
dynamics, then a few linearly dependent frequencies appeared, making the signal
more complicated, then chaotic dynamics was exhibited, where finally .230 <
t < 286/ periodic/laminar shell dynamics appeared. It is clear that the different
Fourier frequency spectra were obtained depending on the applied time intervals.
Furthermore, assuming that one takes the FFT regarding the whole studied time
interval .0 < t < 286/, the obtained results will have no physical interpretation.
This is why another supplemented technique should be applied, namely the wavelet-
based approach. Another important point yielded by our numerical simulation is a
question if the so far reported scenario violates the theoretical prediction of the so-
called wave (or weak) turbulence relying on the relaxation of strong assumptions
exhibited by the fully developed turbulence and, in particular, where the presence
of the intermittency is removed [31]. Second, different wavelets (Meyer, Morlet,
complex Morlet, real and complex Gauss, Daubechies) were analyzed with respect
to their application to study nonlinear dynamics of shells.
In Sect. 4 we apply the Bolotin–Novikov model to study two-layer panels with
a gap. A sudden jump into a turbulent regime is reported accompanied by a stiff
stability loss. The previous symmetry of a contact pressure and deflection isoclines
is also suddenly broken. It means that this sudden scenario of transition into a
spatiotemporal chaos is associated with a catastrophe and symmetry breaking.
Furthermore, we observed that spatial and temporal chaos appeared simultaneously,
yielding shell turbulent dynamics.
Next, Sect. 5 focuses on a study of contact interaction of two-layer cylindrical
shells coupled via boundary conditions and subjected to local periodic transversal
load action. The governing PDEs are solved via the Bubnov–Galerkin higher-
order approximation method, whereas the obtained ODEs are solved through
the fourth-order Runge–Kutta method. Again a weak shell turbulent behavior is

observed, and again the spatial and temporal chaos appears simultaneously.
Finally, Sect. 6 deals with nonlinear dynamics of plates/shells governed by the
Kirchhoff–Love PDEs. The stated boundary value problem was reduced to ODEs
through two different approaches (FDM and FEM), which allowed us to achieve
validity and reliability of the results. From the theoretical/numerical point of view
we propose here a novel method for computation of the Lyapunov exponents within
the neural network framework. From a physical point of view, a weak turbulent plate
behavior was observed.
Acknowledgements This work has been supported by the grant RFBR No 12-01-31204 and the
National Science Centre of Poland under the grant MAESTRO 2, No. 2012/04/A/ST8/00738, for
years 2012–2015.
References
1. Alvarez Diaz, L., Martin, M.T., Vampa, V.: Daubechies wavelet beam and plate finite elements.
Finite Elem. Anal. Des. 45, 206–209 (2009)
2. Amaratunga, K., Williams, J.R., Quian, S., Weiss, J.: Wavelet-Galerkin solutions for one
dimensional partial differential equations. Int. J. Natl. Meth. Eng. 37, 2703–2716 (1994)
mechanical systems. Part 1: mathematical models and solution methods. Chaos Solitons
Fractals 45, 687–708 (2012)
mechanical systems. Part 2: modelling transitions from regular to chaotic dynamics. Chaos
Solitons Fractals 45, 709–720 (2012)
mechanical systems. Part 3: the Lyapunov exponents, hyper, hyper-hyper and spatial-temporal
chaos. Chaos Solitons Fractals 45, 721–736 (2012)
6. Awrejcewicz, J., Krylova, E.Y., Papkova, I.V., Krysko, V.A.: Wavelet-based analysis for
the regular and chaotic dynamics of rectangular flexible plates subjected to shear-harmonic
loading. Shock Vib. 19, 979–994 (2012)
7. Benettin, G., Galgani L., Giorgilli, A., Strelcyn, J.M.: Lyapunov characteristic exponents for
smooth dynamical systems and for Hamiltonian systems: a method for computing all of them.
Part I: theory. Part II: numerical application. Meccanica 15, 21–30 (1980)
8. Boudaut, A., Cadot, O., Odille, B., Touzé, C.: Observation of wave turbulence in vibrating
plates. Phys. Rev. Let. 100, 234504 (2008)
9. Connaughton, C.: Numerical solutions of the isotropic 3-wave kinetic equation. Physica D 238,
2282–2297 (2009)
10. Ghanem, R., Romeo, F.: A wavelet-based approach for the identification of linear time-varying
dynamical systems. J. Sound Vib. 234, 555–576 (2000)
11. Ginzburg, V.L., Landau, L.D.: On the theory of superconductivity. ZhETF 20, 1064–1081
(1950) (in Russian)
12. Hans, S., Ibraim, E., Pernot, S., Boutin, C., Lamarque, C.-H.: Damping identification in
multi-degree-of-freedom system via a wavelet-logarithmic decrement-Part 2: Study of a civil
engineering building. J. Sound Vib. 235, 375–403 (2000)
13. Hariharan, G.: Solving finite length beam equation by the Haar wavelet method. Int. J. Comput.
Appl. 9, 27–24 (2010)
14. Henon, M.: A two-dimensional mapping with a strange attractor. Comm. Math. Phys. 50,
69–77 (1976)
15. Kantor, B.Ya.: Problems of Nonlinear Theory of Rotational Shells. Naukova Dumka, Kiev
(1990) (in Russian)
16. Kornichin, M.S.: Non-Linear Problems of the Theory Plates and Shallow Shells and Methods
of Their Solutions. Nauka, Moscow (1964) (in Russian)
17. Lamarque, C.-H., Pernot, S., Cuer, A.: Damping identification in multi-degree-of-freedom
system via a wavelet-logarithmic decrement-Part I. Theory. J. Sound Vib. 235, 361–374 (2000)
18. Lepik, U.: Impulsively loaded fully clamped elastic-plastic beams by Galerkin’s method. Int.
J. Impact Eng. 15, 17–23 (1994)
19. Lepik, U.: Vibrations of elastic-plastic fully clamped beams and flat arches under impulsive
loading. Int. J. Non-Linear Mech. 29, 613–623 (1994)
20. Lepik, U.: Dynamic response of elastic-plastic beams with axial constraints. Int. J. Impact Eng.
15, 3–16 (1994)
21. Liew, K.M., Wang, Q.: Application of wavelet theory for crack identification in structures. J.
Eng. Mech. 124, 152–157 (1998)
22. Lorenz, E.N.: Deterministic non-periodic flow. J. Atm. Sci. 20, 130–141 (1963)
23. Mordant, N.: Are there waves in elastic wave turbulence? Phys. Rev. Let. 100, 124505 (2008)
24. Mordant, N.: Fourier analysis of wave turbulence in a thin elastic plate. Eur. Phys. J. B 76,
537–545 (2010)
25. Staszewski, W.J., Tomlinson, R.G.: Application of the wavelet transform to fault detection in a
spur gear. Mech. Syst. Signal Process 8, 289–307 (1994)
26. Touze, C., Thomas, O., Amabili, M.: Transition to chaotic vibrations for harmonically forced
perfect and imperfect circular plates. Int. J. Non-linear Mech. 46, 234–246 (2011)
27. Touze, C., Bilbao, S., Cadot, O.: Transition scenario to turbulence in thin vibrating plates. J.
Sound Vib. 331, 412–433 (2012)
28. Verhulst, P.-F.: Recherches mathématiques sur la loi d’accroissement de la population. Nouv.
Mém. Acad. Roy. Sci. Belles-Lett. Bruxelles 18, 1–41 (1845) (in French)
29. Volmir, A.S.: Nonlinear Dynamics of Plates and Shells. Nauka, Moscow (1972) (in Russian)
30. Wang, W.J., McFadden, P.D.: Application of wavelets to gearbox vibration signals for fault
detection. J. Sound Vib. 192, 927–939 (1996)
31. Zakharov, V.E., Lvov, V.S., Falkovisch, G.: Kolmogorov Spectra of Turbulence I: Wave
Turbulence. Springer, Berlin (1992)
Using Bifurcation Diagrams for Controlling
Chaos
Daniel Morcillo, Daniel Burbano, Fabiola Angulo, and Gerard Olivar
Abstract In this chapter, a methodology to compute chaos controllers based on

bifurcation diagrams is proposed and applied to PWM-controlled power converters.
The technique is based on an adaptive control where the offset of the T -periodic
sawtooth signal is modified. Basically, the sawtooth is redefined as a function of
the output and reference voltages. This control technique reduces the percentage
of regulation error as well as eliminates orbits of period greater than one and the
chaotic behavior when the voltage source is varied. Finally, numerical results are
obtained to validate the performance of the proposed scheme.
1 Introduction
Power converters are widely used in industrial and household appliances. Many
researchers have focused their attention to the study of control techniques for
switching power supplies, specially the buck converter which is one of the most
common converters used in a variety of applications [10]. Due to the presence
of the PWM, the power converters exhibit a plethora of complex behaviors.
These behaviors strongly affect the performance of the converter. Nevertheless,
many techniques have been developed to counteract chaotic regimes and another
undesired behaviors [1, 4, 9]. In order to eliminate high-period orbits and chaos two
techniques have been mainly proposed: OGY [7] and TDAS [8]. Both methods
work properly when they are used in the buck converter and an example can be found
in [3]. However, these techniques require complex schemes to be implemented as
well as they need digital devices such as FPGAs, DSPs, microcontrollers, and so on.
D. Morcillo () • F. Angulo • G. Olivar

Department of Ingeniería Eléctrica, Electrónica y Computación, Universidad Nacional de
Colombia, Campus La Nubia, Manizales, Colombia
e-mail: jdmorcillob@unal.edu.co; fangulog@unal.edu.co; golivart@unal.edu.co
D. Burbano
Department of Systems and Computer Sciences, University of Naples Federico II, Via Claudio
21, 80125 Napoli, Italy
e-mail: danielalberto.burbanolombana@unina.it

78 D. Morcillo et al.
Deane et al. [2] proposed a basic voltage-mode control applied to a DC–DC

buck converter based on a P W M , using a simple ramp as a T -periodic signal.
The circuit is depicted in Fig. 1a. This simple technique makes that the system
regulates for some Vi n values, but orbits of period greater than one and chaotic bands
appear when Vi n varies. After the knowledge of these phenomena, some authors
have developed controllers for this power converter. In [5] a method which is called
“Vi n feed-forward” is discussed; this method changes the amplitude of the ramp in
such a way that the amplitude is proportional to Vi n values. On the other hand, in
[6] are presented different ways to automatically adjust the slope of the ramp to
avoid output variations or subharmonic oscillations. However, these papers do not
provide bifurcation analysis to observe clearly the performance improvement in a
wider operation range of Vi n , and they do not consider the offset voltage of the ramp
as a variable.
With the aim to develop a control technique simple to implement, in this paper
it is proposed a novel control scheme which suppresses chaotic bands and orbits
of period greater than one in a wide range of power source, reference voltage,
and load value for the buck power converter controlled by ramp. Essentially, this
method, which is named “adaptive ramp control,” consists in adapting the ramp
waveform (V r.t /) to the control signal (Vco.t /) in such a way V r.t / becomes
similar to Vco.t /. In this way the ramp signal can change its offset voltage over
the time. In addition, with this control technique, the frequency of the ramp signal
is not altered; thus the buck converter continues as a periodically forced system.
Besides, after applying the proposed technique, the performance of the system has
been proven numerically. It was proven that 1T -periodic orbits remain stable within
the Vi n range Œ13; 40 V.
Fig. 1 (a) Simplified circuit of DC–DC buck converter. (b) Bifurcation diagram. Vi n is used as
bifurcation parameter
Using Bifurcation Diagrams for Controlling Chaos 79
2 DC–DC Buck Converter
A simplified schematic diagram of the P W M voltage-controlled buck converter is

depicted in Fig. 1a and is described by Eq. (1):

VPC 1=RC 1=C VC 0

D C u (1)
IPL 1=L Ri n =L IL Vi n =L
VC and IL are the state variables and correspond to capacitor voltage and inductor
current. Switches S1 and S2 are operating in a complementary way. R, L, and C are
the load resistance, the inductor, and the capacitor, respectively. Ri n is the equivalent
resistance of the current sensor and inductor internal resistance. Vi n is the value of
the power source. u takes values into the discrete set f0, 1g and its value is given
according to the controller.
The buck converter controlled by ramp consists in a control signal Vco.t / which is
compared with a T -periodic sawtooth waveform V r.t / which is given by Eq. (2):
t
V r.t / D VL C .Vu VL / (2)
T
where VL and Vu are constant voltages (lower and upper voltage values) and T is
the switching period. The control signal Vco.t / is proportional to the output error,
which is the difference between the reference voltage V ref and the output voltage
VC . The amplifier A1 with gain a is used to obtain Vco.t /, and A2 is as comparator
which has infinite gain, and it is used to obtain the control action u. Vco.t / and u
are computed according to the following equations:
Vco.t / D a .VC .t / V ref / (3)

1 if Vco.t / < V r.t /
uD (4)
0 if Vco.t / > V r.t /
To perform numerical analysis SimPowerSystems tool has been used. Table 1

contains all the parameter values used in this work. In Fig. 1b a bifurcation diagram
as Vi n varies is shown. This bifurcation scenario was firstly reported in 1990 by
Deane and Hamill in [2]. Bifurcation diagram observed in Fig. 1b exhibits the same
behaviors as it has been described in other works. Using parameters of Table 1, close
to Vi n D 28 V a period doubling orbit is detected and after Vi n D 37:5 V chaos is
presented. With the aim to eliminate the chaotic behavior and orbits with period
greater than one, a control technique based on adaptive ramp is proposed.
Table 1 Parameter values Parameter Value

used in simulations
Input voltage (Vi n ) (20–40) V
Reference voltage (V ref ) 11.3 V
Inductance (L) 20 mH
Inductor resistance (Ri n ) 3.8
Capacitance (C ) 47 F
Load (R) 22
Lower voltage (Vu ) 3.5 V
Upper voltage (VL ) 8.1 V
Ramp period (T ) 400 s
A1 gain (a) 7.8
A2 gain (C omp) 1
3 Methodology to Compute the Adaptive Ramp
Analysis of the signal Vco.t / shows that as Vi n increases Vco.t / also increases
and Vco.t / is unable to interact with the voltage of ramp V r.t /, inducing the
bifurcation scenario presented in Fig. 1b. Then, the main idea is to change V r.t /
value depending on Vco.t / value without changing the slope of the ramp which
makes easy to implement the controller. Remaining fixed the slope of the ramp and
avoiding undesired behaviors it is possible to obtain fixed frequency switching and a
simple circuit design. However, in order to find a suitable constant slope, the design
of the controller is based on the bifurcation diagram when the slope changes. In this
way, the basic “adaptive ramp (Var.t /)” in which k and Vi n vary is defined as
Vco.t / Vi n t
Var.t / D C (5)
k k T
where Var.t / corresponds to the signal that will replace V r.t /, k is a constant to be
computed, and Vi n will be fixed to a constant value. Values of Vi n and k are to be
calculated in such a way that undesired behaviors are avoided. From Eq. (5) can be
observed that a change in Vco.t / immediately updates Var.t /. Notice that the term
Vco.t / makes the adaptive-ramp waveform (Var.t /) change its voltage offset. Now,
the control signal is described by

1 if Vco.t / < Var.t /
uD (6)
0 if Vco.t / > Var.t /
In what follows the methodology to compute k and to fix the sawtooth slope value
is explained.
The objective now is to find the values for expression Vi n of Eq. (5) and
constant k which guarantee 1T -periodic solution with good regulation performance.
Bifurcation diagrams have been used to find suitable values for Vi n and k. Initially,
Fig. 2 (a) Two-dimensional bifurcation diagram. Bifurcation parameters are Vi n and k. (b) Three-
dimensional diagram. Bifurcation parameters are Vi n and k, showing the percentage of regulation
error (%e)
Fig. 3 (a) Function that provides suitable values of k according to Vi n . (b) One-dimensional
bifurcation diagram of the buck converter controlled by adaptive ramp taking k D 5:7 for all
Vi n values. VC .t / versus Vi n
a two-dimensional bifurcation diagram (depicted in Fig. 2a) is computed using Vi n

and k as bifurcation parameters. The aim of this diagram is to find a region where the
system presents a 1T -periodic solution, which is obtained if k is properly chosen.
Now, the values for which the system exhibits the lowest regulation error are found.
Figure 2b shows the steady-state regulation error (in percentage) as Vi n and k vary.
In Fig. 3a the lowest percentage error and the stability limit of 1T -periodic orbit
for each combination of Vi n and k are presented; in fact, the limit of the stability
coincides with the lowest error. Curves were smoothed by cubic interpolation.
1T -Periodic orbits taken from Fig. 3a upper are mapped to Fig. 3a lower. Hence,
from Fig. 3a can be observed that it is possible to obtain a 1T -periodic orbit for
Vi n 2 Œ13; 70 V by selecting a proper value of k. According to this figure, the

procedure to compute Vi n and k is explained.
The first step is to define the working range of Vi n , Vi n 2 Œ13; 40 is chosen, and
then Vi n in Eq. (5) is replaced by 40 as it is the last value of the working range. Now,
taking into account stability range and lowest error curves, the second step consists
of choosing k value. In this case k D 5:7; then the final design is
Vco.t / 40 t
Var.t / D C (7)
5:7 5:7 T
4 Numerical Results
In the following, the designed controller will be proven when the input voltage,
reference voltage, and load resistance change. The control design was successful
and bifurcations and chaotic attractor were suppressed. Figures 3b and 4a, b
are bifurcation diagrams numerically computed, using the voltage source, load
resistance, and reference voltage as bifurcation parameters, respectively. In these
bifurcation diagrams it is observed that 1T -periodic orbit is preserved when: (i)
Vi n 2 Œ13; 40 V and R D 22 and V ref D 11:3 V, (ii) R 2 Œ2; 500 and
Vi n D 40 V and V ref D 11:3 V, and (iii) V ref 2 Œ8; 26 V and Vi n D 40 V and
R D 22 .
Finally, three-dimensional bifurcation diagrams (taking V i n, V ref , and R
as bifurcation parameters) have been computed in order to observe the broader
range of 1T -periodic orbits achieved applying the “adaptive-ramp control” to the
buck converter. In Fig. 5a, b are shown the bifurcation diagrams computed from
the system controlled by ramp and for the system controlled by adaptive ramp,
respectively.
Fig. 4 One-dimensional bifurcation diagram. k D 5:7 and Vi n D 40 V. (a) Load resistance R

varies. (b) Reference voltage V ref varies
Fig. 5 Three-dimensional bifurcation diagram of the buck converter (a) controlled by ramp, (b)
controlled by adaptive ramp
5 Conclusions
A new technique for chaos control applied to a PWM voltage-controlled buck

converter has been proposed. The design approach is based on the analysis of
bifurcation diagrams to compute constant parameters which define the controller.
The methodology provides an adaptive ramp whose offset voltage changes as Vco
changes, which prevents complex behaviors within the range Vi n 2 Œ13; 40 V.
The proposed control is easy to implement, does not need many components, and
yields excellent results. Simulation results were used to show the feasibility of the
proposed control method.
Acknowledgements This work was partially supported by Universidad Nacional de Colombia -

Manizales, project 12475, Vicerrectoría de Investigación, DIMA, and Programa Jóvenes Investi-
gadores e Innovadores 2011 (COLCIENCIAS) convenio No. 0043 de 2012.
References
1. Bouzahir, H., El Guezar, F., El Aroudi, A., Ueta, T.: Partial time delayed feedback control
of chaos in a hybrid model of a DC–DC converter. In: Proceedings of 3rd International
Symposium on Communications, Control and Signal Processing, 2008. doi:10.1109/ISCCSP.
2008.4537211
2. Deane, J.H.B., Hamill, D.C.: Analysis, simulation and experimental study of chaos in the buck
converter. In: Proceedings of 21st Annual IEEE Power Electronics Specialists Conference,
PESC’90, 1990. doi:10.1109/PESC.1990.131228
3. El Aroudi, A., Debbat, M., Giral, R., Olivar, G., Benadero, L., Toribio, E.: Bifurcations in
DC-DC switching converters: review of methods and applications. Int. J. Bifurcat. Chaos Appl.
Sci. Eng. 15, 1549–1578 (2005). doi:10.1142/S0218127405012946
4. Giaouris, D., Banerjee, S., Zahawi, B., Pickert, V.: Stability analysis of the continuous-
conduction-mode buck converter Via Filippov’s method. IEEE Trans. Circ. Syst. I Regular
Papers 55(4), 1084–1096 (2008). doi:10.1109/TCSI.2008.916443
5. Karppanen, M., Suntio, T., Sippola, M.: Dynamical characterization of input-voltage-

feedforward-controlled buck converter. IEEE Trans. Ind. Electron 54, 1005–1013 (2007).
doi:10.1109/TIE.2007.892732
6. Liyu, Y., Jinseok, P., Huang, A.Q.: An adaptive external ramp control of the peak current con-
trolled Buck converters for high control bandwidth and wide operation range. In: Proceedings
of 25th Annual IEEE Applied Power Electronics Conference and Exposition, APEC’2010,
2010. doi:10.1109/APEC.2010.5433539
7. Ott, E., Grebogi, C., Yorke, J.A.: Controlling chaos. Phys. Rev. Lett 64, 1196–1199 (1990).
doi:10.1103/PhysRevLett.64.1196
8. Pyragas, K.: Continuous control of chaos by self-controlling feedback. Phys. Lett. A 170,
421–428(1992). http://dx.doi.org/10.1016/0375-9601(92)90745-8
9. Samosir, A.S., Yatim, A.H.M.: Dynamic evolution control for synchronous buck DC-DC
converter: theory, model and simulation. Simul. Model. Pract. Theory 18(5), 663–676 (2010).
http://dx.doi.org/10.1016/j.simpat.2010.01.010
10. Xiaohui, Q., Siu-Chung, W., Tse, C.K.: Resonant assisted buck converter for offline driving
of high brightness LED replacement lamps. In: Proceedings of 2010 IEEE Energy Conversion
Congress and Exposition (ECCE), 2010. doi:10.1109/ECCE.2010.5618054
Shear Waves Dispersion in Cylindrically
Structured Cancellous Viscoelastic Bones
I.V. Andrianov, V.V. Danishevs’kyy, and J. Awrejcewicz
Abstract In this chapter we study anti-plane shear waves propagating through a

cylindrically structured cancellous bone represented by a two-dimensional mesh of
elastic trabeculae filled by a viscous marrow. In the long-wave limit, the original
heterogeneous medium can be approximately substituted by a homogeneous one
characterized by an effective complex shear modulus. The effect of dispersion is
caused by the transmission of mechanical energy to heat due to the viscosity of
the marrow (viscoelastic damping). We derive an approximate analytical solution
using the asymptotic homogenization method; the cell problem is solved by means
of a boundary shape perturbation and a lubrication theory approaches. For short
waves, when the wavelength is comparable to the trabeculae size, the effect of
dispersion is caused by successive reflections and refractions of local waves at the
trabecula-marrow interfaces (Bloch dispersion). Decrease in the wavelength reveals
a sequence of pass and stop frequency bands, so the heterogeneous bone can act like
a discrete wave filter.
I.V. Andrianov ()

Department of General Mechanics, RWTH Aachen University, Templergraben 64,
52056 Aachen, Germany
e-mail: andrianov@gmail.com
V.V. Danishevs’kyy
Department of Structural Mechanics and Strength of Materials, Prydniprovs’ka State Academy of
Civil Engineering and Architecture, Chernyshevs’kogo 24a, Dnipropetrovs’k 49600, Ukraine
e-mail: vdanish@ukr.net
J. Awrejcewicz
Department of Automation, Lodz University of Technology, Biomechanics and Mechatronics,
1/15 Stefanowski St., 90-924 Lodz, Poland
Department of Vehicles, Warsaw University of Technology, 84 Narbutta Str.,
02-524 Warsaw, Poland

86 I.V. Andrianov et al.
1 Introduction
Animal and human bones are heterogeneous materials with a complicated hierar-
chical structure. Bone tissues occur in the two main forms: as a dense solid (cortical
or compact bone) and as a porous medium filled by a viscous marrow (trabecular
or cancellous bone) [17]. The basic mechanical discrepancy between these two
types consists in their relative densities measured by a volume fraction of solids.
Both types can be found in the most bones of the body. A classical example of the
macroscopic bone structure can be given by the long bones (e.g., humerus, femur,
and tibia). They include an outer shell of a dense cortical tissue surrounding an inner
core of a porous cancellous tissue.
The microstructure of cancellous bones is often described by two- or three-
dimensional mesh of interconnected rods and plates [12, 17, 21]. Despite the
obvious simplicity, such idealized models can provide a satisfactory agreement
between theoretical predictions and experimental results for the mechanical prop-
erties of real bones [10, 11, 15, 19, 25]. In the present paper, we shall deal with a
two-dimensional model of cylindrically structured cancellous bones.
Williams and Lewis [27] considered a real 2D section of a trabecular bone and
evaluated its elastic constants using the plane-strain finite elements method. The
developed approach enables one to predict mechanical properties of cylindrically
structured cancellous bones basing on the morphological measurements in the
transverse plane.
A challenging problem consists in the detection of the bone structure using non-
invasive measurements. The inverse homogenization approach (“dehomogenization
procedure”) [7] can help to derive information about the microgeometry of the bone
tissue from the magnitudes of static effective moduli. However, the static moduli
supply only limited morphological data. Much more information can be obtained
studying dynamic response of the bones. Measuring velocities and attenuation
of acoustic waves at different frequencies provide us with additional information
about the microstructure. Generally, frequency-dependent dynamic properties of the
bone may be considered as a kind of “identification portrait”, which is unique for
every sample. The larger is the explored frequency range, the more accurate is the
“portrait” that can be compiled. This should give a possibility to detect even very
small variations of the internal bone texture.
Acoustic waves propagating through cancellous bones undergo dispersion and
damping. There are two different physical effects influencing on the dynamic
properties of the bone: (i) transmission of mechanical energy to heat due to the
viscosity of the marrow (viscoelastic damping and dispersion) [16] and (ii) succes-
sive reflections and refractions of local waves at the trabecula-marrow interfaces
(Bloch dispersion) [2, 20]. From the theoretical standpoint, both effects are realized
simultaneously. However, their intensities are very frequency dependent. For many
real materials the effects of viscoelastic damping and Bloch dispersion are observed
in rather distant frequency ranges. In such a case they can be analyzed separately.
We study propagation of anti-plane shear waves through a 2D cylindrically
structured cancellous bones. In our work, we have shown what kind of dispersion
mechanism is dominant in the propagation of viscoelastic waves in the bone. This
is an important theoretical result, because it allows choosing an adequate simulation
Shear Waves Dispersion in Cylindrically Structured Cancellous Viscoelastic Bones 87
model for the solution of this class of problems. Namely, it is necessary to consider
Bloch dispersion and at the same time viscoelastic effects can be neglected, i.e. use
an elastic model of the media.
The input dynamic problem is formulated in Sect. 2. In Sect. 3, the long-
wave limit is considered. The solution is evaluated by means of the asymptotic
homogenization procedure, and the effect of viscoelastic damping is predicted. In
Sect. 4, the short-wave case is studied and the effect of Bloch dispersion is analyzed
by means of the plane-wave expansion method. Conclusive remarks are presented
in Sect. 5.
2 Input Dynamic Problem
We study transverse anti-plane shear waves propagating in the x1 x2 plane through

a regular cancellous structure consisting of a spatially infinite elastic matrix
(trabeculae) (1) and viscous inclusions (marrow) (2) (Fig. 1). The governing two-
dimensional wave equation is
@2 u
rx .Grx u/ D ; (1)
@t 2
where G is the complex shear modulus, is the mass density, u is the longitudinal
displacement (in the x3 direction), and r x D e1 @/@x1 C e2 @/@x2 , e1 , e2 are the unit
Cartesian vectors.
Due to the heterogeneity of the medium, the physical properties G and are
represented by piecewise continuous functions of co-ordinates:
G .x/ D G .a/ ; .x/ D .a/ ; for x 2 .a/ ; x D x1 e1 C x2 e2 : (2)
Fig. 1 Cancellous structure under consideration

Here and in the sequel the superscript (a) denotes different components of
the structure a D 1, 2. In Eq. (2), G(1) is the real shear modulus of the elastic
matrix and G(2) is the frequency-dependent imaginary shear modulus of the marrow.
Following the linear theory of viscoelasticity, we can set G(2) D i!(2) , where ! is
the frequency of a harmonic wave and (2) is the viscosity of the marrow.
Equation (1) can be written in the equivalent form:
@2 u.a/
G .a/ rxx u D .a/
2 .a/
; (3)
@t 2
ˇ
.2/ ˇ
˚ .1/ ˇ .1/ @u
.1/
.2/ @u
u D u.2/ ˇ@˝ ; G DG ˇ
ˇ ; (4)
@n @n @˝
where r 2xx D @2 /@x21 C @/@x22 , @/@n is the normal derivative to the contour @. From
the physical standpoint, Eqs. (4) means the perfect bonding conditions at the
trabecula-marrow interface @.
3 Long-Wave Approach: Asymptotic Homogenization
3.1 Two-Scale Asymptotic Procedure
We start with the case when the wavelength L is essentially larger than the internal
size l of the cancellous structure l < < L. The original heterogeneous bone can
be approximately substituted by a homogeneous one with a certain homogenized
(effective) complex shear modulus G0 . Such an approach neglects local reflections
and refractions of the waves on microlevel. The effect of dispersion is caused by
the transmission of the mechanical energy of the acoustic wave to heat due to the
viscosity of the marrow.
Let us study the input boundary value problem (3), (4) by the asymptotic
homogenization method [4]. In order to separate macro- and microscale components
of the solution we introduce so-called slow x and fast y co-ordinate variables
x D x; y D "1 x; (5)
where y D y1 e1 C y2 e2 , " D l/L is a natural small parameter, and search the displace-
ment as an asymptotic expansion
.a/ .a/
u.a/ D u0 .x/ C "u1 .x; y/ C "2 u2 .x; y/ C : : : : (6)
The first term u0 of expansion (6) represents the homogenized part of the
solution; it changes slowly within the whole sample of the bone and does not
depend on the fast co-ordinates (@u0 /@y1 D @u0 /@y2 D 0). The next terms, u(a) i ,
i D 1, 2, 3, : : : , provide corrections of the orders "i and describe local variations of
the displacements on microlevel.
The differential operators read
rx D rx C "1 ry ; rxx
2
D rxx
2
C 2"1 rxy
2
C "2 ryy
2
; (7)
where r y D e1 @/@y1 C e2 @/@y2 , r 2xy D @2 /(@x1 @y1 ) C @2 /(@x2 @y2 ), r 2yy D @2 /@y21
C @/@y22 .
Splitting the input problems (3), (4) with respect to " leads to a recurrent sequence
of cell boundary value problems:
.a/
@2 ui2
2 .a/ 2 .a/ 2 .a/
G .a/ rxx ui2 C 2rxy ui1 C ryy ui D .a/ ; (8)
@t 2
n oˇ
.1/ .2/ ˇ
ui D ui ˇ ;
@˝
( ! !) ˇ
.1/ .1/ .2/ .2/ ˇ
@ui1 @u @ui1 @u ˇ
G .1/
C i DG .2/
C i ˇ ; (9)
@n @m @n @m ˇ
@˝
where i D 1, 2, 3, : : : , u(a)
1 D 0, @/@m is the normal derivative to the interface @
written in fast variables.
For a spatially periodic medium, the terms u(a) i have to satisfy the conditions of
periodicity
.a/ .a/
ui .x; y/ D ui x; y C Lp ; (10)
and normalization
D E
.a/
ui D 0; (11)
where Lp D " 1 lp , lp D p1 l1 C p2 l2 , p1 , p2 D 0, ˙ 1, ˙ 2, : : : , l1 , l2 are the funda-

mental translation vectors of the cancellous structure (see Fig. 1),
0 1
“ “
1 BB
C
hi D @ ./ dy1 dy2 C ./ dy1 dy2 C
A (12)
S0
.1/ .2/
˝0 ˝0
is the homogenizing operator over the unit cell domain 0 D (1) (2)
0 C 0 (Fig. 2),
and S0 D L is the area of the unit cell in the fast co-ordinates.
2
Conditions (10) and (11) can be approximately replaced by zero boundary

conditions for same functions in the center and along the outer contour @0 of the
unit cell:
Fig. 2 Periodically repeated

unit cell
n oˇ n oˇ
.2/ ˇ .1/ ˇ
ui D 0 ˇ ; ui D 0 ˇ : (13)
x;yD0 @˝0
For 1D problems, Eq. (13) appears entirely equivalent to Eqs. (10) and (11).
For 2D problems, replacing the periodicity conditions by zero boundary conditions
increases stiffness of the system and, thus, provides an upper bound for the effective
properties. Analysis of numerical examples has shown that discrepancy between the
final solutions in both cases is not essential [1, 3], whereas utilizing approximation
(13) leads to a sufficient simplification of the cell problems.
Due to the periodicity of u(a)i (10), Eqs. (8) and (9) can be considered within
only one distinguished unit cell. Solution of the cell problems (8), (9), (13) at i D 1
determines the term u(a)
1 . In order to find the effective modulus G0 , the homogenizing
operator (12) is applied to Eq. (8) at i D 2. The terms u(a)2 are eliminated by means
of Green’s theorem, which together with the boundary conditions (9) and the
periodicity relation (10) implies
D E
2 .a/ 2 .a/
G .a/ rxy ui C ryy uiC1 D 0:
As a result, the homogenized wave equation of the order "0 is obtained:

D E ˝ ˛ @2 u0
2 .a/
G .a/ rxx
2
u0 C rxy u1 D .a/ : (14)
@t 2
Substituting here expressions for u(a)

1 evaluated below we shall come to a
macroscopic wave equation
@2 u0
G0 rxx
2
u0 D 0 ; (15)
@t 2
where 0 D (1 c(2) )(1) C c(2) (2) is the effective mass density, c(2) is the volume
fraction of the inclusions, c(2) D A/S0 , A is the size of the inclusion (Fig. 2). The
effective modulus G0 can be derived after evaluation of the integrals in Eq. (14).
Below we find approximate solutions of the cell problem (8), (9), (13) and
determine the effective shear modulus G0 using a boundary shape perturbation and
lubrication theory approaches.
3.2 The Case c(2) < < 1: Boundary Shape Perturbation
If the volume fraction c(2) of the marrow inclusions is relatively small, the
square shapes of the domains (1) , (2) can be approximately substituted by the
equal circles of radii R1 , R2 so that c(2) D R22 /R21 (Fig. 3). This simplification can
be considered as the first approximation of the method of the boundary shape
perturbation [1, 14].
Let us introduce in the unit cell the polar co-ordinates r2 D y21 C y22 , tan
D y2 /y1 .
Equations (8), (9), and (13) at i D 1 read
.a/ .a/ .a/

@2 u1 1 @u1 1 @2 u1
C C D 0; (16)
@r 2 r @r r 2 @
2
Fig. 3 Simplification of the

unit cell in the case c(2) < < 1
n oˇ
.1/ .2/ ˇ
u1 D u1 ˇ ;
rDR2
ˇ (17)
.1/ .2/ ˇ
G .1/ @u0 @u1
C @r D G .2/ @u0
C
@u1 ˇ ;
@n @n @r ˇ
rDR2
n oˇ n oˇ
.2/ ˇ .1/ ˇ
u1 D 0 ˇ ; u1 D 0 ˇ ; (18)
rD0 rDR1
where @/@n D cos

@/@x1 C sin
@/@x2 .
Solution of the simplified cell problem (16)–(18) is
@u
u1 D C1 r C C2 r 1
.a/ .a/ .a/ 0
; (19)
@n
.2/ .2/
.1/ 1 c .2/ .1/ 1 R22
C1 D .2/ ; C2 D .2/ ;
C 1 c .2/ .2/ 1 C 1 c .2/ .2/ 1
.2/
.2/ 1 1 c .2/ .2/
C1 D .2/ ; C2 D 0;
C 1 c .2/ .2/ 1
where (2) D G(2) /G(1) .

Substituting expressions (19) into the homogenized equation (14), we obtain the
effective modulus G0 in the closed analytical form:

.2/ C 1 C c .2/ .2/ 1
0 D .2/ ; (20)
C 1 c .2/ .2/ 1
where 0 D G0 /G(1) .
It should be noted that solution (20) is precisely the same as can be obtained
by the composite cylinder assemblage model and by the generalized self-consistent
scheme [8].
3.3 The Case c(2) ! 1: Lubrication Theory
In the case of densely packed marrow inclusions, when the volume fraction c(2) is
close to unit, c(2) ! 1, an asymptotic solution of the cell problem can be obtained
using as a natural small parameter the non-dimensional width ı D H/L of the trabec-
ula (Fig. 4). Let us suppose ı < < 1. Being restricted by the O(ı 0 ) approximation,
Fig. 4 Unit cell in the case

c(2) ! 1
for the matrix strips d˝ 1 , d˝ 2 , which separate neighbouring inclusions, one can
show
.1/ .1/ .1/ .1/
@2 u1 @2 u1 @2 u1 @2 u1
2
>> for y 2 d ˝1 ; << for y 2 d ˝2 : (21)
@y1 @y22 @y12
@y22
The physical meaning of estimations (21) is that in the narrow strip d1 the
variation of local stresses in the direction y1 is dominant and, hence, the term
@2 u(1) 2 2 (1) 2
1 /@y2 can be neglected in comparison with @ u1 /@y1 . Vice versa, in the strip
d2 , the dominant variation of the local stress field takes place in the direction
y2 , so the term @2 u(1) 2 2 (1) 2
1 /@y1 can be neglected in comparison with @ u1 /@y2 . Such
a simplification is similar to the basic idea of the well-known lubrication theory,
which was used in the theory of composites for many years [8, 9].
Following estimations (21), in the O(ı 0 ) approximation, Eq. (8) reads
.1/ .1/ .1/

! .1/
@2 ui2 @2 ui1 @2 ui @2 ui2
G .1/
C 2 C D .1/ ; (22)
@xs2 @xs @ys @ys2 @t 2
.2/
@2 ui2
2 .2/ 2 .2/ 2 .2/
G .2/ rxx ui2 C 2rxy ui1 C ryy ui D .2/
@t 2
Solution of the simplified cell problems (9), (13), (22) at i D 1 is
p
1
c .2/ .2/ 1 @u0
.2/
u1 D p ys ; (23)
c
.2/ .2/ 1
.2/ @x s
b
a λ0 1.0
λ0 5 0.9
0.8
4
0.7
0.6
3
0.5
2 0.4
0.3
1 0.2
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
c(2) c(2)
Fig. 5 Effective modulus in the elastic case. Solids—formula (20), dashes—formula (24). (a)
(2) D 5. (b) (2) D 0.2
p
.1/ c .2/ .2/ 1 L @u0
u1 D p ys at ys > 0;
.2/ c .2/ .2/ 1 2 @xs
p
.1/ c .2/ .2/ 1 L @u0
u1 D p C ys at ys < 0;
c
.2/ .2/ 1
.2/ 2 @x s
where s D 1, 2.
For the effective shear modulus we obtain
p p
.2/ c .2/ 1 c .2/ .2/ 1
0 D p : (24)
.2/ c .2/ .2/ 1
Numerical results, calculated by formulas (20) and (24), are very close (except
the case (2) < 1, c(2) ! 0). This is illustrated at Fig. 5 for real values of (2) , which
correspond to elastic materials. Moreover, in the limit c(2) ! 1, the approximate
solutions (20) and (24) exhibit the same asymptotic behaviour and give identical
expansions for 0 until the order O[(1 c)2 ]:
1 .2/ h 2 i
0 D .2/ 1 .2/ C 1 1 c .2/ C O 1 c .2/ at c .2/ ! 1:
2
This fact reveals that for the cancellous structure under consideration, expression
(20), originally obtained for the case c(2) < < 1, provides a reasonable approxima-
tion in the whole region of the inclusions volume fraction 0 c(2) 1.
3.4 Propagation of Long Waves
Let us consider a harmonic wave
u0 D U exp .i x/ exp .i !t / ; (25)
where U is the amplitude, ! is the frequency, and D 1 e1 C 2 e2 is the wave

vector and the direction of propagation is determined by the angle ˛, tan ˛ D 2 /1 .
Separating real R and imaginary I parts of the wave vector D R iI ,
expression (25) reads
u0 D U exp .I x/ exp .iR x/ exp .i!t / : (26)
Here I D jI j is the attenuation factor and R D jR j D 2/L is the wave
number.
For the viscoelastic composite medium, the effective complex modulus G0 , the
attenuation coefficient I and the phase velocity vp D !/R depend on the frequency
of the travelling signal. Substituting expression (26) into the macroscopic wave
equation (15), we obtain
.G0;R C iG0;I / .I C iR /2 D 0 ! 2 ; (27)
where G0,R , G0,I are, respectively, the real and the imaginary part of G0 ,
G0 D G0,R C iG0,I . Collecting in Eq. (27) the terms at 1 and i, after routine
transformations we derive
v20
I D R tan .'0 =2/ ; v2p D ; (28)
cos .'0 =2/2
p
where tan(®0 ) D G0,I /G0,R is the effective loss tangent and v0 D jG0 j =0 is the
effective velocity in the elastic case.
Adopting for G0 the solution (20), we obtain
G .1/ .1c .2/ /Ci!.2/ .1Cc .2/ /

G0 D G .1/ G .1/ ;
.1Cc .2/ /Ci!.2/ .1c .2/ /
(29)
4Gih.1/ c .2/ !.2/
tan .'0 / D h 2 2
i
2 :
1.c .2/ / .G .1/ / C.!.2/ /
In the numerical examples presented below we accept some rough estimations of

the properties of the components following Bryant et al. [6], Guo [13], and Van Riet-
bergen and Huiskes [26]. The shear modulus of the trabeculae is G(1) D 3.85 109 Pa
and the viscosity of the marrow is (2) D 0.15 Pa s (at the room temperature of
20 ı C) and (2) D 0.05 Pa s (at the body temperature of 37 ı C). The trabeculae
volume fraction c(1) D 1 c(2) can vary from 0.05–0.1 for aged osteoporotic bones
to 0.3–0.35 for young normal bones.
m/mR 0.7
0.6
0.5
0.4
0.3
0.2
0.1
8 9 10 11 12 13 14
1g ω, Hz
Fig. 6 Attenuation factor of a normal bone. Solids—(2) D 0.05 Pa s, dashes—(2) D 0.15 Pa s
Dependencies of the attenuation factor I upon the frequency ! are displayed

at Fig. 6 (normal bone, c(1) D 0.3) and Fig. 7 (osteoporotic bone, c(1) D 0.1).
The dispersion effect vanishes (i) at ! ! 0, when the deformation rate is small and
the stiffness of the marrow is negligible, and (ii) at ! ! 1, when the deformation
rate is high, so the marrow acts like a perfectly stiff medium. Decrease of the
trabeculae volume fraction c(1) leads to the intensifying of the dispersion: the
damping frequency region extends and the attenuation factor I grows. Decrease
in temperature (i.e. increase in the marrow viscosity (2) ) leads to a reduction of
the damping frequency. In any case, for physically meaningful values of the bone
properties, the effect of viscoelastic damping can be observed starting from the
frequencies of the order 100 MHz and higher.
It should be noted that in the long-wave limit (l < < L) the cancellous structure
under consideration is transversely orthotropic. The obtained solution for anti-plane
shear waves is isotropic in the plane x1 x2 , so the parameters G0 , ®0 do not depend
on the direction of the wave propagation. The effect of anisotropy is predicted in the
case of short waves (see Sect. 4).
4 Floquet-Bloch Approach: Plane-Wave Expansion Method
When the wavelength L is comparable to the internal size l of the cancellous

structure, the effect of dispersion is caused by successive reflections and refractions
of local waves at the trabecula-marrow interfaces. Decrease in the wavelength
reveals a sequence of pass and stop frequency bands (so-called phononic bands)
m/mR 0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
8 9 10 11 12 13 14
1g ω, Hz
Fig. 7 Attenuation factor of an osteoporotic bone. Solids—(2) D 0.05 Pa s, dashes—

(2) D 0.15 Pa s
[2, 20]. Thus, a heterogeneous bone can act as a discrete wave filter. If the
frequency of the signal falls within a stop band, a stationary wave is excited and the
neighbouring trabeculae vibrate in alternate directions. On macrolevel the amplitude
of the global wave attenuates exponentially, so no propagation is possible.
In order to explore such a case, let us assume the threshold of the first stop band
to be essentially lower than the viscoelastic damping frequencies. The marrow is not
involved into the shear deformation, so we can set G(2) D 0, (2) D 0.
Following the Floquet-Bloch theorem [5], a harmonic wave propagating through
a periodic cancellous structure is represented in the form
w D F .x/ exp .i x/ exp .i !t / ; (30)
where F(x) is a spatially periodic function, F(x) D F(x C lp ).

We use the plane-wave expansion method [20, 23] and express the function F(x)
and the material properties G(x), (x) as infinite Fourier series:
1
X 1
X
2
F .x/ D Ak1 k2 exp i .k1 x1 C k2 x2 / ;
l
k1 D1 k2 D1
X1 X1
2
G .x/ D Bk1 k2 exp i .k1 x1 C k2 x2 / ; (31)
l
k1 D1 k2 D1
X1 X1
2
.x/ D Ck1 k2 exp i .k1 x1 C k2 x2 / ;
l
k1 D1 k2 D1
where
“
2
Bk1 k2 D 1
l2
G .x/ exp i .k1 x1 C k2 x2 / dx1 dx2 ;
l
“
0
2
Ck1 k2 D l2
1
.x/ exp i .k1 x1 C k2 x2 / dx1 dx2 ;
l
0
“
the operator ./dx1 dx2 denotes integration over a distinguished unit cell 0 .
0
Substituting Ansatz (30) and expansions (31) into the wave equation (1) and
collecting the terms exp[i2l 1 (j1 x1 C j2 x2 )], j1 , j2 D 0, ˙ 1, ˙ 2, : : : , we come to
an infinite system of linear algebraic equations for the unknown coefficients Ak1 k2 :
8
ˆ
ˆ
1
X 1
X < 2
Ak1 k2 B k C 1 2 j C 1 C
ˆ j 1 k1 ; l 1 l 1
k1 D1 k2 D1 :̂
j2 k2 9 (32)
>
>
2 2 =
C l k2 C 2 l j2 C 2 C ! 2
D 0:
j 1 k1 ; > >
;
j2 k2
System (32) has a nontrivial solution if and only if the determinant of the matrix
of the coefficients is zero. Equating the determinant to zero, we derive a dispersion
relation for ! and . It should be noted that the plane-wave expansion method
does not use explicitly the bonding conditions (4), whereas they are “embedded”
implicitly into Eq. (1) and expansions (31).
To illustrate the appearance of phononic band gaps, let us rewrite Ansatz (30)
separating real R and imaginary I parts of the wave vector D R iI :
w D F .x/ exp .I x/ exp .iR x/ exp .i !t / : (33)
The imaginary part I jI j represents the attenuation factor. Frequency regions
where I ¤ 0 correspond to stop bands (signal (30) attenuates exponentially), while
regions where I D 0 correspond to pass bands.
In numerical examples the dispersion relations are calculated approximately by
the truncation of the infinite system (32) supposing jmax js jmax . The number of
the kept equations is (2jmax C 1)2 . We expect that increase in jmax shall improve the
accuracy of the solution. From the physical point of view, such a truncation means
cutting off the higher frequencies.
Figure 8 displays dispersion curves for a normal bone with the following prop-
erties of the trabecular tissue: G(1) D 3.85 109 Pa, (1) D 1900 kg/m3 , c(1) D 0.3.
Calculations are performed at jmax D 3. The diagram consists of two (left and
Fig. 8 Dispersion curves of a normal bone
right) parts separated by a vertical dash line. The right part displays a solution
for the orthogonal direction (˛ D 0) and the left part for the diagonal direction
(˛ D /4) of the wave propagation. The results for the frequency ! are normalized
to ! 0 D v0 R D 2v0 /L. We can observe that in the long-wave case (! ! 0, l/L ! 0)
the solution is isotropic. However, with the increase in ! and decrease in L, the
cancellous structure exhibits an anisotropic behaviour.
Shaded areas at Fig. 8 indicate the threshold of the first stop bands. Let us
estimate the corresponding values ! s of the frequency. We obtain ! s l/(! 0 L) 0.29
at ˛ D 0 and ! s l/(! 0 L) 0.41 at ˛ D /4. The typical lengthp of trabeculae is
about l 10 3 m. Taking into account ! 0 D 2v0 /L, v0 D G0 =0 , we derive
! s 2.0 MHz at ˛ D 0 and ! s 2.8 MHz at ˛ D /4.
5 Conclusions
For anti-plane shear waves the effect of Bloch dispersion, caused by the hetero-
geneity of cancellous bones, appears at essentially lower frequencies than the effect
of viscoelastic damping, caused by the viscosity of the marrow. Bloch dispersion
is expected to play the primary role in the processes of ultrasonic diagnostic,
which usually deals with acoustic waves in the regions 1–10 MHz. The viscoelastic
damping can be neglected until the frequency of about 100 MHz. Obtained results
may be used for the development of new methods of non-invasive testing and
diagnostic.
In the present paper, a perfectly regular arrangement of marrow inclusions is
investigated. It is clear that the microstructure of real bones is not regular. At the
same time, it has been shown in a number of studies [18, 22, 24] that regular
structures exhibit the narrowest stop bands in comparing to disordered systems.

Thus, the obtained solutions may be treated as theoretical bounds for the stop band
thresholds that appear in randomly disordered bone tissues.
Acknowledgments This work is supported by the Alexander von Humboldt Foundation (Insti-
tutional academic co-operation programme, grant no. 3.4-Fokoop-UKR/1070297), the German
Research Foundation (Deutsche Forschungsgemeinschaft, grant no. WE 736/30-1).
References
1. Andrianov, I.V., Danishevs’kyy, V.V., Guillet, A., Pareige, P.: Effective properties and micro-
mechanical response of filamentary composite wires under longitudinal shear. Eur. J. Mech.
A/Solids 24, 195–206 (2005)
2. Andrianov, I.V., Bolshakov, V.I., Danishevs’kyy, V.V., Weichert, D.: Higher-order asymptotic
homogenization and wave propagation in periodic composite materials. Proc. Roy. Soc.
A Math. Phys. Eng. Sci. 464, 1181–1201 (2008)
3. Andrianov, I.V., Bolshakov, V.I., Danishevs’kyy, V.V., Weichert, D.: Asymptotic study of
imperfect interfaces in conduction through a granular composite material. Proc. Roy. Soc.
A Math. Phys. Eng. Sci. 466, 2707–2725 (2010)
4. Bakhvalov, N.S., Panasenko, G.P.: Homogenization: Averaging Processes in Periodic Media.
Mathematical Problems in Mechanics of Composite Materials. Kluwer, Dordrecht (1989)
5. Brillouin, L.: Wave Propagation in Periodic Structures: Electric Filters and Crystal Lattices.
Dover, New York (2003)
6. Bryant, J.D., David, T., Gaskell, P.H., King, S., Lond, G.: Rheology of bovine bone marrow.
Proc. Inst. Mech. Eng. H J. Eng. Med. 203(2), 71–75 (1989)
7. Cherkaev, E., Ou, M.-J.Y.: Dehomogenization: reconstruction of moments of the spectral
measure of the composite. Inverse Probl. 24, 065008 (2008)
8. Christensen, R.M.: Mechanics of Composite Materials. Dover, Mineola (2005)
9. Frankel, N.A., Acrivos, A.: On the viscosity of a concentrated suspension of solid spheres.
Chem. Eng. Sci. 22, 847–853 (1967)
10. Gałka, A., Telega, J.J., Tokarzewski, S.: A contribution to evaluation of effective moduli of
trabecular bone with rod-like microstructure. J. Theor. Appl. Mech. 37, 707–727 (1999)
11. Gałka, A., Telega, J.J., Tokarzewski, S.: Application of homogenization to evaluation of
effective moduli of linear elastic trabecular bone with plate-like structure. Arch. Mech. 51,
335–355 (1999)
12. Gibson, J.L., Ashby, M.P.: Cellular Solids: Structure and Properties. Pergamon, Oxford (1988)
13. Guo, X.E.: Mechanical properties of cortical bone and cancellous bone tissue. In: Cowin, S.C.
(ed.) Bone Mechanics Handbook, pp. 10-1–10-23. CRC, Boca Raton (2001)
14. Guz, A.N., Nemish, Y.N.: Perturbation of boundary shape in continuum mechanics. Soviet
Appl. Mech. 23(9), 799–822 (1987)
15. Hollister, S.J., Fyhire, D.P., Jepsen, K.J., Goldstein, S.A.: Application of homogenization
theory to the study of trabecular bone mechanics. J. Biomech. 24, 825–839 (1991)
16. Hughes, E.R., Leighton, T.G., Petley, G.M., White, P.R., Chivers, R.C.: Estimation of critical
and viscous frequencies for Biot theory in cancellous bone. Ultrasonics 41, 365–368 (2003)
17. Jee, W.S.S.: Integrated bone tissue physiology: anatomy and physiology. In: Cowin, S.C. (ed.)
Bone Mechanics Handbook, pp. 1-1–1-68. CRC, Boca Raton (2001)
18. Jensen, J.S.: Phononic band gaps and vibrations in one- and two-dimensional mass-spring
structures. J. Sound Vib. 266, 1053–1078 (2003)
19. Kasra, M., Grynpas, M.D.: Static and dynamic finite element analyses of an idealized structural
model of vertebral trabecular bone. J. Biomech. Eng. 120, 267–272 (1998)
20. Kushwaha, M.S., Halevi, P., Martinez, G., Dobrzynski, L., Djafari-Rouhani, B.: Theory of
acoustic band structure of periodic elastic composites. Phys. Rev. B 49, 2313–2322 (1994)
21. Odgaard, A.: Qualification of cancellous bone architecture. In: Cowin, S.C. (ed.) Bone
Mechanics Handbook, pp. 14-1–14-19. CRC, Boca Raton (2001)
22. Schraad, M.W.: On the macroscopic properties of discrete media with nearly periodic
microstructures. Int. J. Solids Struct. 38, 7381–7407 (2001)
23. Sigalas, M.M., Economou, E.N.: Elastic and acoustic wave band structure. J. Sound Vib. 158,
377–382 (1992)
24. Thorp, O., Ruzzene, M., Baz, A.: Attenuation and localization of wave propagation in rods
with periodic shunted piezoelectric patches. Smart Mater. Struct. 10, 979–989 (2001)
25. Tokarzewski, S., Telega, J.J., Gałka, A.: Prediction of torsional rigidities of bone filled with
marrow: the application of multipoint Padé approximants. Acta Bioeng. Biomech. 2, 560–566
(2000)
26. Van Rietbergen, B., Huiskes, R.: Elastic constants of cancellous bone. In: Cowin, S.C. (ed.)
Bone Mechanics Handbook, pp. 15-1–15-24. CRC, Boca Raton (2001)
27. Williams, J.L., Lewis, J.L.: Properties and an anisotropic model of cancellous bone from the
proximal tibial epiphysis. J. Biomech. Eng. 104(1), 50–56 (1982)
Quasi-regular and Chaotic Dynamics of Postural
Sway in Human
Natalya Kizilova, Elena Karpinska, and Michael Karpinsky
Abstract Parameters of different vertical two-leg stances are studied by the force
platform measurements on a control group of young healthy individuals and two
groups of patients with spine and joint pathologies. Body sway patterns are classi-
fied and analyzed and some new indexes are proposed for clinical diagnostics of the
spine and joint pathologies. The dependencies between the body accelerations in the
frontal and sagittal planes are proposed for the estimation of the resulting feedback
force that provides stable vertical stance. The spectral and wavelet analyses are
used for the detection of the quasi-regular and chaotic behavior in the dynamics
of postural sway. It is shown that the amplitudes and power spectral densities of the
center of the pressure coordinates are subjected to periodic oscillations, which are
not in phase for different frequencies. The quasi-regular behavior is proper to the
young healthy volunteers, while most of the patients with combined spine and joint
impairments demonstrate the chaotic dynamics of the body sway.
1 Introduction
Body sway parameters are important for medical diagnostics of the musculoskeletal,
neurological, visual, and balance system impairments. Steady stance of a human
is provided by vestibular, visual, auditory, and tactile information relevant for the
balance control. Degradation in the sensory integration is determined by age-related
variations or specific diseases gradually leading to balance disorders, sudden fall,
and trauma.
N. Kizilova ()
Department of Theoretical and Applied Mechanics, Kharkov National University,
Svobody sq., 4, 61022 Kharkov, Ukraine
Interdisciplinary Centre for Mathematical and Computational Modeling, Warsaw University, ul.
Krakowskie Przedmieście 26/28, 00-927 Warszawa, Polska
e-mail: n.kizilova@gmail.com
E. Karpinska • M. Karpinsky
Department of Biomechanics, M.I. Sytenko Institute of Spine and Joint Pathology, Pushkinska st.,
80, 61024 Kharkov, Ukraine

104 N. Kizilova et al.
Computerized posturography is used for assessment, treatment, rehabilitation,

and management of individuals with balance problems and implanted prosthesis.
Computerized posturography is a simple and biomechanically tractable instrument
for early diagnosis of the vestibular and musculoskeletal pathology. The tech-
nique assesses how well a given individual integrates muscular torque in joints
and visual somatosensory and other information needed for the balance control.
Computerized
n! posturography is based on automatic measurements of the reaction
on
forces R i in n separate movable parts of the force platform or in films
iD1
with continuously distributed pressure sensors. In this study the computer-assisted
force platform “Statograph M006” that allows simultaneous measurement of four
reaction forces (see Fig. 1a) and computes the coordinates (xp , yp ) of the center
of pressure (COP) was used. Since the body is composed of
200 bones and
>700 muscles controlled by nervous, humoral, and local tissue regulation systems,
the physiological cycles of different characteristic times are involved into the
balance control and the trajectories (xp (t), yp (t)) are quite complex [6]. Physiological
systems exhibit regular quasi-periodical or chaotic behavior. The former might
be deterministic chaos or random behavior [9]. Quasi-periodicity is proper to
physiological systems with complex spectra of oscillations at different scales, which
is the case for human locomotor, cardiovascular, and nervous systems. Deterministic
chaos is frequently observed in the physiological systems which are responding
and adapting to external stimuli. For instance, for the musculoskeletal system, the
visual, sensory, tactile, mechanical, and heat are the most important signals. Random
behavior corresponds to uncontrolled and uncoordinated systems [9].
Some parameters of the posturographic curves and integral indexes like sway
amplitudes in the frontal (0y) and sagittal (0x) planes and sway asymmetry are
widely used in medicine for diagnostics of the musculoskeletal and nervous system
disorders, balance control, and inner ear pathologies [8, 10]. Force platform is a
useful tool for the stimulation of the coordination and balance functions in children,
sportsmen, and elderly people during training or rehabilitation, because coordinated
movement demands considerable practice at special conditions. Stabilography has
gained widespread acceptance in the rehabilitation of thepatients with trauma,
stroke, Parkinson’s disease, and cerebral palsy for recovery and improvement of
the locomotor function and speech after the stroke and for the development of the
individual training regimes and optimal individual sport positions of biathlonists,
weight lifters, figure skaters, and shooters [3–5]. Progressive decrease in the sway
amplitude and body stabilization is observed during the force platform training of
the sportsmen and the patients with balance impairments.
The posture stability may be maintained with respect to a moving preference
point rather than a stationary one as it was suggested in [11, 12]. The sway
amplitude is bigger and the sway pattern is more complicated at unstable one-
leg stances and two-leg balance on the unstable or moving support. A method of
decomposition of the sway into two time series, which are called rambling and
trembling trajectories, can be used for checking the hypothesis [12]. The rambling
component represents migration of the reference point, with respect to which the
Quasi-regular and Chaotic Dynamics of Postural Sway in Human 105
Fig. 1 An n-link model of a

human upright stance on the
force platform (a) and four
measured reaction forces (b)
equilibrium is maintained. When the horizontal ground reaction force equals zero,
the so-called instant equilibrium point is reached. Then the reference position at
each time interval may be estimated by calculation of the coordinates (xp , yp ) of
COP [11]. The rambling trajectory may be obtained by spline approximation of
the consecutive instant equilibrium points. The trembling component represents the
oscillations of the COP relatively to the rambling trajectory. Analyses of the two
sway components give new insight to understanding the balance control in normalcy
and pathology. In the paper the results of the computer-assisted posturographic
measurements on a control group of young healthy individuals and two groups of
patients with spine and joint problems are analyzed and both chaotic and quasi-
regular behaviors are detected and studied on the developed mathematical model.
2 Materials and Methods
A control group of 40 healthy subjects (weight 66.4 ˙ 19.7 kg, height

1.73 ˙ 0.14 m, and age 20–29 years) without marked neuromuscular disorders have
been asked to keep 30-s quiet two-leg vertical stance (test I) on the force platform
Fig. 2 Sketches of the

stances studied
experimentally on the force
platform: symmetric two-leg
stances (a); two-leg stances
with body weight shifted onto
the left (b) and right (c) foot;
a step forward off from the
force platform on the plate of
the same thickness (d)
o4
n!
(Fig. 2a). The reaction forces R i (Fig. 1b) have been measured for each
iD1
foot and oscillations of COP (xp (t), yp (t)) have been automatically computed. The
second set of tests (test II/III) was 40-s two-leg stance with body weight transferred
onto the right/left leg (Fig. 2b, c correspondingly). The last set of tests was a step
forward off the force platform on the plate of the same thickness (Fig. 2d). The
healthy individuals were also asked to keep the balance standing on one (left/right,
test IV) foot with 30-s rest between the tests.
A group of patients with spine and joint impairments such as osteochondrosis,
arthritis, arthrosis, and trauma (weight 69.6 ˙ 27.5 kg, height 1.69 ˙ 0.11 m, and
age 52–69 years) has been tested on the same series of experiments as the control
group.
The calculated time series (xp (t), yp (t)) have been amplified and the
low (f < 0.01 Hz) and high (f > 10 Hz) frequency components have been
subtracted using the second-order Butterworth filter. Trend of the basic line
has been eliminated by shifting the curves x(t) and y(t) relatively to the mean
values < x(t) > and < y(t)>. The first and the last 5-s portions of the data series have
been deleted to eliminate the experimental errors [8].
3 Experimental Results and Analyses
The computed COP trajectories data have been classified in several types depending
on the direction of the COP transfer. Different types of COP trajectories (xp (t), yp (t))
are presented in Fig. 3. The trajectories in the central box correspond to test I, while
the trajectories in the left/right boxes correspond to the body mass shift onto the
left/right leg (tests II/III). Very few individuals exhibited symmetric stance with
xp
0, yp
0. In most cases a shift backward (Fig. 2b) has been observed, though
Fig. 3 Dynamic COP trajectories for the two-leg stances of healthy patients (a,b), a patient with
coxartrosis (c) and osteochondrosis (d). The central box in each image allocates the normal stance,
while the left and right boxes correspond to the stance with the body weight shift onto the left and
right leg accordingly
some individuals demonstrated a shift forward (Fig. 3a) which is more proper to the
military.
Young healthy individuals exhibited zero or small shift forward/backward while
their body weight was transferred to one of the legs (Fig. 3a, b). It implies that
body shift was not accompanied by body rotation or bending. The sway amplitudes
were always lower during the first test; it means any two-leg stance with shifted
body weight corresponds to a one-leg stance with additional support which is more
unstable than the stance with full support of both feet, in spite of the total contact
area remaining the same in test I and tests II and III (see Fig. 2a–c), though some
exceptions have been observed (Fig. 3). In tests II and III the sway amplitudes could
increase in comparison with test I in the sagittal only (Ax ), frontal only (Ay ), or
in both (Ax and Ay ) planes, either symmetrically or asymmetrically. Here Ax and
Ay are the width and the height of the rectangular circumscribed around the sway
trajectory (the boxes in Fig. 3). The corresponding asymmetry indexes
AIIx;y AIII
x;y
JIIx;y D ; x;y D
JIII
AIx;y AIx;y
give valuable information on the postural balance of the individual during the
change of the type of support. The ratio Jl/r D JIIx,y : JIII x,y describes the left/right
asymmetry of the body balance, which might be normal or pathological depending
on the value of Jl/r . For instance, patients with spastic hemiparesis exhibit noticeable
shift of COP toward the unaffected limb and keep this asymmetry while walking [7].
The directions of the shift differ in right- and left-sided hemispheric patients, and
the changes correlate with the degree of the stance asymmetry Jl/r . More detailed
diagnostic information can be obtained by gradual elimination of the visual (closed
eyes [4]), auditory (soundless or high noise conditions [3]), and other kinds of
control with consequent comparison of the trajectories and sway indexes.
The groups of patients with spine and joint pathology exhibited more complex
behavior during the body weight transfer onto one of the legs (Fig. 3c, d). During
the simple shift their bodies noticeably moved forward or backward in the same
(Fig. 3c) or opposite (Fig. 3d) directions experiencing certain rotation and/or bend. It
might be connected with asymmetric insufficiencies in the ligamentous apparatus or
defense reaction of the body aimed at preventing the discomfort/pain in the injured
joint or spine. As a measure of the disease the angles
r ,
l between the straight
lines connecting the centers of the corresponding rectangular and vertical axis 0y
can be used [12]. Since the spine and joint pathologies are frequently combined, the
differential diagnostics can be done using the indexes JII,IIIx,y , values, and signs of the
angles
r and
l .
The trajectories (xp (t), yp (t)) are helpful for the estimation of the sway amplitudes
and body shift in the frontal and sagittal planes, but more detailed information

on the body balance can be obtained from the phase curves xR p xP p ; yR p yP p and
acceleration curves yR p xR p . Since the acceleration being multiplied by the body
!
mass gives the resulting force R † produced by the musculoskeletal system to turn
the body back to the equilibrium position, the curves yR p xR p correspond to the
feedback forces that provide a stable posture. Young healthy volunteers exhibited
smooth circles around the center that is oscillating along the xP p axes (Fig. 4a).
The moving centers and the cycles correspond to the gambling and trembling
components of the body sway accordingly. The group of patients with spine and
joint disorders
mostly exhibited irregular dynamics when the cycles in the xR p xP p
and yR p yP p planes become incomplete and non-smooth with rapid variations in the
sway velocity and acceleration (Fig. 4b).
!

The curves yR p xR p illustrate the respond R † of the body to an arbitrary
displacement ! u of COP. The simplest balance control function !
R D k! u used
P † P
in theoretical models has not been detected either in the control or in the two groups
!
!
of patients. Instead of the restoring forces R † .t/ D R † ! u P .t & / ; !
v .t & /
P
where − is the time delay, the COP acceleration, which differs from the force by the
body mass multiplier, can be computed and analyzed. The young healthy volunteers

Fig. 4 Dependencies xR p xP p for a young healthy volunteer (a) and an elderly patient with
coxarthrosis (b)

Fig. 5 Dependencies yR p xR p for a patient with coxarthrosis of two (a) and one (b) hip joints
demonstrated uniform distribution of the accelerations in different

directions, while
the patients with joint problems exhibited the asymmetric yR p xR p with one (Fig. 5b)
or two (Fig. 5a) principal directions depending on the type of the disorder. Those
directions depicted in Fig. 5a, b by the arrows have been computed as the most
probable direction of the body deflection averaged over the time series. The balance
control strategy can be based on the “scanning” of the area around the equilibrium
point in different directions intentionally diverging
the COP in the vicinity of the
equilibrium point [2]. It means that the yR p xR p asymmetry corresponds to the
disorders when the balance control system avoids scanning some directions to avoid
the pain/discomfort or exhibit the sway
anisotropy.
Typical phase trajectories xR p xP p (the same for yR p yP p ) computed for a healthy
young volunteer with two-leg stance with a body shift to one of the legs (Fig. 6a)
and the one-leg stance on the same leg (Fig. 6b) differ in the values of velocities
and accelerations. The stance on one leg is more unsteady than any two-leg stance
and all the subjects experienced larger velocities and accelerations (forces). The
maximal values of the velocities andaccelerations
inthe sagittal and coronal planes
may be easily estimated on the xR p xP p and yR p yP p curves, while it is hard to do
on the COP trajectories (Fig. 3). The phase curves can be recommended for clinical
diagnostics of the body balance problems.
The spectral and wavelet analyses have been carried out to detect the regular
and chaotic dynamics of the body sway. The discrete short-time Fourier transform

Fig. 6 Dependencies xR p xP p for a healthy subject standing on two legs with body mass transferred
onto his left leg (a) and standing on his left leg (b)
(DSTFT) of the time series gn
C1
X
G .m; f/ D gn Wnm exp .i!n/;
1
where m corresponds to the time shift and W[n m] is the discrete windows
function, allows simultaneous analysis of the amplitudes G or power spectral density
(PSD) in the frequency and time domains. The Gaussian function was chosen as
W[n m]. Some results of DSTFT of the xp (t) and yp (t) time series are presented in
Fig. 7.
The discrete wavelet transform (DWT)
C1
1 X
.; s/ D p gn k
jsj 1
where k D (n m)/s, s is the scale, and k is the wavelet function.

DSTFT has a constant window size at any time and for any harmonics, and the
problem of exactness of the simultaneous determination of frequency and time exists
for it. DWT applied windows of different scaled sizes and thus allows study of
appearance of the high and low harmonics in different times more precisely [1].
Some results of the DWT of the xp (t) and yp (t) time series are presented in Fig. 8.
It was found that 87 % of the healthy individuals exhibit quasi-periodic time
oscillations of the PSD of different frequencies and the variations are slower in the
frontal plane than in the sagittal plane. In the case presented in Fig. 7 two periods
of the PSD oscillations in the frontal plane (Fig. 7a) and three periods in the sagittal
plane (Fig. 7b) are visible at high frequencies f D 17.5 Hz. The same picture one can
obtain by making slices of the function PSD(t, f) at different frequencies, though
the oscillations at different frequencies are not in phase. Spectral analysis revealed
Fig. 7 Power spectral density of a healthy individual body sway in the frontal (a) and sagittal (b)
planes
Fig. 8 Discrete wavelet transform (, s) of the postural sway of healthy (a, b) and sick (c, d)
individuals in the frontal (a, c) and sagittal (b, d) planes
three main frequencies of the body sway [4] and DSTFT revealed the quasi-periodic
behavior of the body sway, at least in the young healthy volunteers (see Fig. 7).
DWT confirms the slow variations of the amplitudes at upper scales. Two and
four cycles at the basic scale (Fig. 8a, b correspondingly) are presented and the
oscillations are slower in the coronary plane than in the sagittal plane. The patients
with spine and joint problems demonstrate more complex chaotic dynamics with
Fig. 9 COP trajectories of the step off forward from the force platform measured on the healthy
individual (a) and patients with osteochondrosis (b) and coxarthrosis of the left (c) and right (d)
knees
spike-type changes in the amplitudes (Fig. 8c, d) even at the basic scale. The spikes
are not computational errors, and they did not disappear after iterative filtering of
the signals.
The chaotic dynamics have also been detected in the trajectories of COP
produced during the step off forward from the force platform. The test is very
important because it includes a transition from the control strategy over the steady
posture to the control over gait. In healthy individuals the trajectory starts with
almost straight line connecting the COP positions for the two-leg stance and stance
on the support foot followed by the body transfer forward onto another foot (Fig. 9a).
The trajectories are reproducible during either short-term or long-term studies
exhibiting steady personal body dynamics.
Patients with spine problems demonstrate different dynamics of the COP dis-
placement during the step forward. First of all, they move their COP backward
releasing the corresponding leg, then start to move along the straight direction to
the final destination and on the almost half-way change the direction of motion
toward the support foot transferring their body mass on it (Fig. 9b). The patterns
of the COP trajectories are also patient-specific. The heights of the “harps” on the
COP trajectories (h in Fig. 9b; the same sign on the other curve is not depicted for
simplicity) and the ratio hr : hl measured on the yp (xp ) curves for the steps forward
with right and left legs can be recommended for diagnostics of the spine pathologies.
When patients with trauma or joint impairments who suffer of acute pain make
a step with support on the affected leg, the COP demonstrates irregular trajectories
with unpredictable acceleration and deceleration of the body in both frontal and
sagittal planes (left foot in Fig. 9c and right foot in Fig. 9d). In this case the COP
trajectories are irreproducible even being repeatedly measured during consequent
steps in a short-term study. It implies that the control strategy is based on sudden
changes in the locomotor programs depending on occasional combination of the
space positions of the body segments resulting in either normal or painful reactions.
4 Conclusions
Human postural sway can exhibit quasi-regular and chaotic dynamics depending
on the age, locomotor impairments, and balance system state. COP trajectories
during the two-leg stances with different body weight shifts are characterized by
variations of the body sway in the frontal and sagittal planes and/or variations in
the sway asymmetry. The corresponding indexes together with the displacement of
the COP forward or backward are helpful for the diagnostics of spine, joints, and
combined pathology. Analysis of the phase curves is useful for understanding the
quasi-regular and chaotic body sway dynamics and body control strategy. The quasi-
regular behavior is more proper to the young healthy volunteers, while most of the
patients with combined spine and joint impairments exhibited the chaotic dynamics
with certain asymmetry of the body acceleration and the returning force produced
by the muscle synergy. Some new indexes on the COP trajectories during the step
forward off from the force platform are proposed for differential diagnostics spine
and joint pathology.
References
1. Akansu, A.N., Smith, M.J.T.: Subband and Wavelet Transforms: Design and Applications.
Kluwer Academic Publishers, Boston (1995)
2. Godi, M., Grasso, M., Guglielmetti, S., Nardone, A., Schieppati, M.P.: Changes of balance and
gait in spastic hemiparesis are correlated with asymmetry of quiet stance. Gait Posture 28, s6
(2008)
3. Griskevicius, J., Jarmaliene, E., Sesok, A., Daunoraviciene, K., Kizilova, N.: Evaluation of
human postural balance in quiet standing by direct measurement of human body center of
mass acceleration. J. Vibroeng. 11, 556–561 (2009)
4. Karpinsky, M.Y., Kizilova, N.: Computerized posturography for data analysis and mathemat-
ical modelling of postural sway during different two-legged and one-legged human stance.
J. Vibroeng. 9, 118–124 (2007)
5. Kizilova, N., Karpinsky, M., Griskevicius, J., Daunoraviciene, K.: Posturographic study of the
human body vibrations for clinical diagnostics of the spine and joint pathology. Mechanika 80,
37–41 (2009)
6. Latash, M.L., Zatsiorsky, V.M. (eds.): Classics in Movement Sciences. Human Kinetics,
Champaign (2001)
7. Mochizuki, L., Duarte, M., Amadio, A.C., Zatsiorsky, V.M., Latash, M.L.: Changes in
postural sway and its fractions in conditions of postural instability different postural control
mechanisms. J. Appl. Biomech. 22, 51–60 (2006)
8. Norr, M.E., Forrez, G.: Posture testing (posturography) in the diagnosis of peripheral vestibular
pathology. Arch. Otorhinolaryngol. 243, 186–189 (1986)
9. Sharma, V.: Deterministic chaos and fractal complexity in the dynamics of cardiovascular
behavior: perspectives on a new Frontier. Open Cardiovasc. Med. J. 3, 110–123 (2009)
10. Sologubov, E.G., Yavorskii, A.B., Kobrin, V.I., Nemkova, S.A., Sinel’nikova, A.N.: Use of
computer stabilography and computer-assisted biomechanical examination of gait for diagnosis
of posture and movement disorders in patients with various forms of infantile cerebral
paralysis. Biomed. Eng. 34, 138–143 (2000)
11. Zatsiorsky, V.M., Duarte, M.: Instant equilibrium point and its migration in standing tasks:
rambling and trembling components of the stabilogram. Motor Control 3, 28–38 (1999)
12. Zatsiorsky, V.M., King, D.L.: An algorithm for determining gravity line location from
posturographic recordings. J. Biomech. 31, 161–164 (1998)
Modelling of Ropes with Consideration of Large
Deformations and Friction by Means
of the Rigid Finite Element Method
Iwona Adamiec–Wójcik, Jan Awrejcewicz, Lucyna Brzozowska,

and Łukasz Drag
˛
Abstract This paper presents a model of a rope, which is an example of a slender

system. The method used for modelling is the modified rigid finite element method,
which enables us to consider not only bending flexibility but also longitudinal
flexibility. A new approach presented in the paper is validated by comparison of
our own results with an analytical solution for a catenary line. For this problem
the influence of both the number of elements into which the link is discretised
and various values of stiffness coefficient is analysed. Numerical simulations of the
dynamics of a rope of an offshore crane lifting a load from a vessel are presented.
1 Introduction
Slender systems such as lines, ropes and pipes are very often used in offshore
engineering applications. The analysis of these systems requires the use of methods
enabling large deformations to be considered. Methods of discretising such systems
have been developed over many years, yet the most popular is the finite element
method [4, 10, 19]. Application of the finite segment method to modelling of planar
systems is discussed in [18], while [9] presents the lumped mass method used for
modelling spatial risers. A method containing features of both the lumped mass and
finite segment methods is the rigid finite element method (RFEM) [15]. This enables
slender links to be discretised by dividing them into rigid finite elements which
assume the mass (inertial) characteristics of the link and massless, dimensionless
spring-damping elements which reflect the flexible characteristics. The method is
also used to analyse various systems in offshore engineering [16]. In the classical
I. Adamiec–Wójcik () • L. Brzozowska • Ł. Drag ˛

University of Bielsko–Biała, ul. Willowa 2, 43-309 Bielsko–Biała, Poland
e-mail: i.adamiec@ath.bielsko.pl; lbrzozowska@ath.bielsko.pl; ldrag@ath.bielsko.pl
J. Awrejcewicz
Lodz University of Technology, ul. Stefanowskiego 1/15, 90-924 Lodz, Poland
Warsaw University of Techonology, ul. Narbutta 84, 02-524 Warsaw, Poland
e-mail: awrejcew@p.lodz.pl

116 I. Adamiec–Wójcik et al.
formulation of RFEM each rigid finite element (rfe) has six DOFs for a spatial
model and three DOFs for a planar model. A modified formulation [1, 12, 17] limits
considerations to the analysis of bending and torsional vibrations for the spatial
systems and bending in one direction for the planar systems.
In modelling of cranes it is often assumed that the rope is nondeformable [5, 14]
or that the rope is replaced by a substitute longitudinal stiffness [6–8]. In the analysis
of planar vibrations of a rope fixed at one end the authors of [3] consider bending
flexibility. However, wave phenomena in the rope can have an essential influence on
the displacements of the load when considering ropes with large lengths. A complex
model of the rope is necessary in order to enable us to consider such phenomena.
This paper presents a planar model of the rope for the analysis of both bending and
longitudinal vibrations.
2 Formulation of the Model
This section presents the formulation of the modified rigid finite element method
(MRFEM) for planar slender systems in which both bending and longitudinal
flexibilities are considered, without the need to connect rfes by means of constraint
equations. Such an approach eliminates constraint reactions and reduces the size of
the final system to 2(n C 2) equations.
Figure 1 shows a flexible link divided into n C 1 elements.
It is assumed that the motion is described by x0 , y0 , coordinates of point A0 ; and
i , ' i , generalised coordinates of rfe i(i D 0, 1, : : : , n).
The number of elements of the vector of generalised coordinates is defined by
the formula
Fig. 1 Division of a flexible link into rfes and sdes

Modelling of Ropes with Consideration of Large Deformations and Friction. . . 117
m D 2 C 2 .n C 1/ D 2n C 4: (1)
Coordinates of point Ai can be calculated in the following way:
X
i1
X
i1

xi D x0 C lj C j cos 'j ; yi D y0 C lj C j sin 'j ; (2)
j D0 j D0
where lj are lengths of rfes preceding rfe i and j , ' j are defined in Fig. 1.
2.1 Kinetic Energy of a Link
The kinetic energy of rfe i can be expressed as follows:
1 .1/ h .1/ i2 1 .1/ 2 1 .2/ h .2/ i2 1 .2/ 2

Ei D m vc;i C Jc;i 'Pi C mi vc;i C Jc;i ; 'Pi (3)
2 i 2 2 2
(j) (j)
where mi is the mass of part (j) of rfe i, vc,i is the velocity of the centre of mass of
(j)
part (j) of rfe i, and J c,i is the central inertial moment of part (j) of rfe i, j D 1, 2 (see
Fig. 2).
Fig. 2 Rigid finite element (rfe i) in an undeformed state

Coordinates of the centre of mass of rfe i can be defined by means of the formulae
.1/ .1/ .1/ .1/

xc;i D xi C ai cos 'i ; yc;i D yi C ai sin 'i (4)
and

.2/ .2/ .2/ .2/
xc;i D xi C ai C i cos 'j ; yc;i D yi C ai C i sin 'i : (5)
Differentiation of (4) and (5) yields
h i2 h i2 h i2
.1/ .1/ .1/
vc;i D xP c;i C yPc;i D
h i2
.1/ .1/
D xP i2 C yPi2 C ai 'Pi2 2ai 'Pi ŒxP i s'i yPi c'i ;
h i2 h i2 h i2
.2/ .2/ .2/
vc;i D xP c;i C yPc;i D (6)
2
D xP i2 C yPi2 C P 2 C a.2/ C i 'P 2 C 2 P i .xP i c'i C yPi s'i /
i
i i
.2/
2 ai C i 'Pi .xP i s'i yPi c'i / ;
where s' i D sin ' i , c' i D cos ' i .

Substituting (6) into (3), the kinetic energy of rfe i can be obtained:
2
.2/ P 2 1 .2/ .2/
Ei D 12 mi xP i2 C 12 mi yPi2 C 12 mi i C 2 mi ai C i 'Pi2 C 12 Ji 'Pi2
h i (7)
.2/ .2/ P
C ŒxP i s'i C yPi c'i 'Pi ˛i C mi i C mi P i c'i C yPi s'i ;
i Œx
where mi D m(1) (2) (1) (1) 2 (1) (2) (1) (1) (2) (2)
i C mi , Ji D mi [ai ] C J i C J i , ˛ i D mi ai C mi ai .
Having differentiated (2), the following is calculated:
X
i1 X
i1

xP i D xP 0 C P j c'j
lj C j 'Pj s'j ;
j D0 j D0
(8)
X
i1 Xi1

yPi D yP0 C P j s'j C
lj C j 'Pj c'j
j D0 j D0
and
X
i1 X
i1

xR i D xR 0 C R j c'j
lj C j 'Rj s'j
j D0 j D0
X
i1 X
i1

2 P j 'Pj s'j
lj C j 'Pj2 c'j ;
j D0 j D0
(9)
X
i1 X
i1

yRi D yR0 C R j s'j C
lj C j 'Rj c'j C
j D0 j D0
X
i1 X
i1

C2 P j 'Pj c'j
lj C j 'Pj2 s'j :
j D0 j D0
In view of (8) the kinetic energy of rfe i depends on x0 , y0 , 0 , ' 0 , : : : , i , ' i and
their derivatives.
The Lagrange operators for kinetic energy Ei of the rfe can be written in the
following form:
d @Ei @Ei X i
"x0 .Ei / D D mi xR 0 C R j c'j
mij
dt @xP 0 @x0 j D0
X
i X
i X
i1
nij 'Rj s'j 2 P j 'Pj s'j
mij nij 'Pj2 c'j ;
j D0 j D0 j D0
(10)
d @Ei @Ei X i
"y0 .Ei / D D mi yR0 C R j s'j C
mij
dt @yP0 @y0 j D0
X
i X
i X
i
C nij 'Rj c'j C 2 P j 'Pj c'j
mij nij 'Pj2 s'j
j D0 j D0 j D0
for j D 1, : : : , i–1:
d @Ei @Ei
"j .Ei / D D mi xR 0 c'j C mi yR0 s'j C
P
dt @j @j
X
i
X i

C R k cos 'j 'k C
mik nik 'Rk sin 'j 'k C
kD0 kD0
Xi
X
i

C2 P k 'Pk sin 'j 'k
mik nik 'Pk2 cos 'j 'k ;
kD0 "
kD0
d @Ei @Ei
"'j .Ei / D D lj C j mi xR 0 s'j C mi yR0 c'j
dt @'Pj @'j
X
i
Xi

R k sin 'j 'k C
mik nik 'Rk cos 'j 'k C
kD0 kD0 #
Xi
X i

C2 P
mik k 'Pk cos 'j 'k C nik 'Pk sin 'j 'k ;
2
kD0 kD0
( (
mi for k < i mi .lk C k
/ for k < i
mik D .2/ ; nik D .2/ ;
mi for k D i ˛i C mi k for k D i
"
d @Ei @Ei .2/
"i .Ei / D D mi xR 0 c'i C yR0 s'i C
dt @ Pi @i
(11)
X i X i
C R k cos .'i 'k / C .lk C k / 'Rk sin .'i 'k /C
kD0 kD0 #
X i X i
C2 P
k 'Pk sin .'i 'k / .lik C i / 'Pk cos .'i 'k / ;
2
kD0 kD0
d @Ei @Ei
.1/
"'i .Ei / D D ˛i C mi i .xR 0 s'i yR0 c'i /
dt @'Pi @'i
X
i X
i
R k sin .'i 'k / C
˛ik ˇik 'Rk cos .'i 'k /C
kD0 kD0
Xi Xi
C2 P k 'Pk cos .'i 'k / C
˛ik ˇik 'Pk2 sin .'i 'k /;
( kD0 (kD0
lk C k for k < i ˛i C m i i for k < i
lik D .2/ ; ˛ik D .2/ ;
ai for k D i m2i ai C j for k D i
8
< ˛i C m.2/ i i .lk C k / for k < i
ˇik D 2 :
:m a C
.2/ .2/
C J for k D i
i k i i
Formulae (11) can be recast to the following compact form for j D 1, : : : , i–1, i:
X
i

"j .Ei / D mij xR 0 c'j C mij yR0 s'j C R k cos 'j 'k C
mij k
kD0
X
i
Xi

C nij k 'Rk sin 'j 'k C 2 P k 'Pk sin 'j 'k
mij k
kD0 kD0
X
i

nij k 'Pk2 cos 'j 'k ;
kD0
(12)
X
i

"'j .Ei / D nij xR 0 s'j C nij yR0 c'j R k sin 'j 'k C
˛ij k
kD0
X
i
Xi

C ˇij k 'Rk cos 'j 'k C2 P k 'Pk cos 'j 'k C
˛j ik
kD0 kD0
Xi

C ˇij k 'Pk2 sin 'j 'k ;
kD0
( (
mik for j < i nik for j < i
mij k D .2/ ; nij k D .2/ ;
mi for j D i mi .lik C i / for j D i

lj C j mik for j < i lj C j nik for j < i
˛ij k D ; ˇij k D :
˛ik for j D i ˇik for j D i
Since the kinetic energy of the link is a sum of the kinetic energies of rfes 0, : : : ,n,
i.e.:
X
n
ED Ei ; (13)
iD0
and bearing in mind that Ei D Ei (x0 , y0 , 0 , ' 0 , : : : , i , ' i ), one gets
X
n X
n
"x0 .E/ D "x0 .Ei /; "y0 .E/ D "y0 .Ei / (14)
iD0 iD0
and for j D 0, 1, : : : , n:
X
n X
n
"j .E/ D "j .Ei /; "'j .E/ D "'j .Ei /: (15)
iDj iDj
Taking into account
X
n X
i X
n X
n
Aik ak D ak Aik ;
iD0 kD0 kD0 iDk
(16)
Xn X i X
n X
n
Aij k ak bj k D ak bj k Aij k ;
iDj kD0 kD0 iDmaxfj;kg
the following relations are obtained:
X
n X
n
"x0 .E/ D mxR 0 C R k Ak c'k
'Rk Bk s'k
kD0 kD0
Xn X
n
2 P k 'Pk Ak s'k
'Pk2 Bk c'k ;
kD0 kD0
X
n X
n (17)
"y0 .E/ D myR0 C R k Ak s'k C
'Rk Bk c'k C
kD0 kD0
Xn Xn
C2 P k 'Pk Ak c'k
'Pk2 Bk s'k ;
kD0 kD0
X
n X
n X
n
where Ak D mik ; Bk D nik ; mD mi and for j D 0, 1, : : : , n:
iDk iDk iD0
X
n

"j .E/ D xR 0 Aj c'j C yR0 Aj s'j C R k ajk cos 'j 'k C

kD0
X
n
Xn

C 'Rk bj k sin 'j 'k C 2
P k 'Pk ajk sin 'j 'k

kD0 kD0
X
n

'Pk2 bjk cos 'j 'k ;
kD0
X
n (18)

"'j .E/ D xR 0 Bj s'j C yR0 Bj c'j R k a' sin 'j 'k C
jk
kD0
X
n
Xn

C
'
'Rk bj k cos 'j 'k C 2 P k 'Pk a' cos 'j 'k
jk
kD0 kD0
X
n
'
'Pk2 bj k sin 'j 'k ;
kD0
X
n X
n
'
X
n
'
where ajk D mij k ; bjk D nij k ;aj k D ˛ij k ; bj k D
iDmaxfj;kg iDmaxfj;kg iDmaxfj;kg
X
n
ˇij k :
iDmaxfj;kg
Let us introduce the following notations:
2 3
r
6 7
6 q0 7
6 7
6 7
6 q1 7 " # " #
6 7 x0 j
6 7
q D 6 7; rD ; qj D ; (19)
6 7 y0 'j
6q 7
6 i7
6 7
6 7
4 5
qn
where q is a vector of generalised coordinates of the link, and

2 3
©r .E/
6 © .E/ 7
6 0 7
6 7
6 ©1 .E/ 7

6 :: 7
6 7 "x0 .E/ "j .E/
©q .E/ D 6 : 7; ©r .E/ D ; ©j .E/ D : (20)
6 7 "y0 .E/ "'j .E/
6 ©i .E/ 7
6 7
6 :: 7
4 : 5
©n .E/
Finally, the following second-order ODEs are obtained
©q .E/ D MqR C h; (21)
where
2 3 2 3
Mrr Mr0 : : : Mrj : : : Mrn hr
6M M00 : : : M0j : : : M0n 7 6h 7
6 0r 7 6 07
6 : :: :: 7 6 :7
6 : 7 6 :7
6 : : : 7 6 :7
MD6 7; h D 6 7;
6 Mj r Mj 0 : : : Mjj : : : Mj n 7 6 hj 7
6 7 6 7
6 :: :: :: :: 7 6 :: 7
4 : : : : 5 4 :5
Mnr Mn0 : : : Mnj : : : Mnn hn

m 0 Ak c'k Bk s'k

Mrr D ; Mrk D MTkr D
0m Ak s'k Bk c'k
" #
ajk cos 'j 'k bjk sin 'j 'k
Mj k D ' ;
aj k sin 'j 'k bj k cos 'j 'k
2 3
Xn Xn
6 2 P
k 'Pk Ak s'k 'Pk Bk c'k 7
2
6 kD0 7
6
hr D 6 X kD0 7;
n Xn 7
4 2 P k 'Pk Ak c'k
'Pk Bk s'k 5
2
kD0 kD0
2 3
Xn
X n

62 P k 'Pk aj k sin 'j 'k

'Pk bj k cos 'j 'k 7
2
6 kD0 7
6
hj D 6 X kD0 7:
n
Xn
7
42 P k 'Pk a cos 'j 'k

'
'Pk2 bj k sin 'j 'k 5
'
jk
kD0 kD0
Elements of matrix M depend on the generalised coordinates q of the link, while

the elements of vector h depend on q and q. P It should be noted that if the longitudinal
flexibility is left out of consideration (j 0, for j D 0, 1, : : : , n), then the formulae
for elements of matrix M and vector h (having erased respective rows and columns)
coincide with those presented in [17].
2.2 Potential Energy of the Link
The potential energy of gravity forces of rfe i can be presented in the form
g .1/ .1/ .2/ .2/

Vi D mi gyc;i C mi gyc;i ; (22)
where g stands for the gravity acceleration.

Bearing in mind (4), (5) and (2), one finds that

g
@Vi 0
@r
D Gir D ;
mg
8 i
ˆ
ˆ mi gs'j
ˆ
< mi g lj C j c'j for j < i (23)
" #
g
@Vi
D Gij D .2/
@qj ˆ
ˆ h mi gs'i i
:̂ .1/ .1/ .2/ .2/ for j D i:
g mi ai C mi ai C i c'i
Since the potential energy of the link is a sum of the energies of all rfes
X
n
g
Vg D Vi ; (24)
iD0
then using the following relations
@V g X @V
n g
@V g X @V n g
D i
; D i
; (25)
@r iD0
@r @qj iDj
@qj
one obtains
3 2
Gr
6G 7
6 07
6 :7
g 6 :7
@V 6 :7
DGD6 7; (26)
@q 6 Gj 7
6 7
6 :: 7
4 :5
Gn

X
n
0
where Gr D ; Gj D Gij for j D 0, 1 , : : : , n.
mg iDj
The energy of spring deformation of the link in the case of linear characteristics
of sdes can be written in the following form:
1X b 1X l 2
n n
Vs D ci .'i 'i1 /2 C c : (27)
2 iD1 2 iD0 i i
Thus, the following is obtained:
@V s @V s
D 0; D Sj ; (28)
@r @qj

" #
c0l 0 cjl j
where S0 D ; Sj D ;
c0b .'1 '0 / cjb 'j 'j 1 cjb C1 'j C1 'j

cnl n
for j D 0, 1, : : : , n–1, Sn D b .
cn .'n 'n1 /
The above relations can be written in the equivalent compact form as follows:
32
0
6S 7
6 07
6 :7
s 6 :7
@V 6 :7
D S D 6 7: (29)
@q 6 Sj 7
6 7
6 :: 7
4 :5
Sn
The motion of the link discretised in the way described above can be caused
by either external forces (force input) or by definition of the motion of point A0
(kinematic input). In the next section, both approaches are presented.
2.3 Synthesis of the Equations
In the case of a force input the equations of motion follow:
MqR D f; (30)
where M is defined in (21) and

@V g @V s
f D h C Q:
@q @q
Here Q denotes the vector of generalised forces resulting from the external
forces.
On the other hand, in the case of kinematic input, the following relations hold:

x0 D x0 .t /; xP 0 D xP 0 .t /; xR 0 D xR 0 .t /;
(31)
y0 D y0 .t /; yP0 D yP0 .t /; yR0 D yR0 .t /;
and thus the components of the vector
r D r.t / (32)
and its derivatives with respect to time are known. Then equations (30) can be cast
to the following form:
X
n
@V g
Mrr rR C Mrk qR k D fr D hr @r
C R0 ;
kD0
Xn (33)
@V g @V s
Mj r rR C Mj k qR k D fj D hj @Rqj
@Rqj
; j D 0; 1; : : : ; n;
kD0
Rx0
where R0 D y are the components of an unknown force acting at point A0
R0
ensuring the realisation of (31) and (32).
Since vector rR is known in the case of kinematic input, the second part of (33)
can be written in the form
X
n
Mj k qR k D fk Mj r rR D Pk (34)
kD0
or equivalently
MqR D P; (35)
2 3 3
2 3 2
M00 : : : M0j : : : M0n q0 P0
6 :: :: :: 7 6 :: 7 6 :: 7
6 : : : 7 6 :7 6 :7
6 7 6 7 6 7
where M D 6 Mj 0 : : : Mjj : : : Mj n 7
6
7; qD6 7
6 qj 7 ; PD6 7
6 Pj 7.
6 : : : 7 6 :7 6 :7
4 : : :
: : 5
: 4 :: 5 4 :: 5
Mn0 : : : Mnj : : : Mnn qn Pn
Having calculated qR from (35), the components of reaction vector R0 can be

calculated from the first part of (33):
X
n
@V g
R0 D Mrr rR C Mrk qR k C hr C : (36)
@r
kD0
Thus, in the case of kinematic input, the calculation of the acceleration vector
requires the solution of the system of 2(n C 1) linear algebraic equations. In the
case of force input the number of unknowns in system (30) is larger by 2 and is
equal to 2(n C 1) C 2.
3 Validation of the Model
If in equations (33) we omit components connected with accelerations and veloci-

ties, we obtain a system of nonlinear algebraic equations enabling the static problem
to be solved. In order to verify the algorithms and computer programs elaborated,
calculations of the shape of the line fixed and loaded as in Fig. 3 are carried out.
Coordinates of point P from the analytical solution of the problem, when bending
flexibility is omitted, are calculated according to the formulae [9]:

FxE s FxE FyE -wLCws FyE -wL

x.s/ D EA
C w
arcsinh F E arcsinh F E ;
x x
y.s/ D EAs
FyE wL C 12 ws C
8" 9 (37)
< 2 # 12 " 2 # 12 =
F E F wLCws
E F wL
E
C wx 1 C y FE 1 C yF E ;
: x x ;
Fig. 3 Catenary line

and force T takes the form

12
2 2
T .s/ D FxE C FyE wL C ws ; (38)
where F Ex , and F Ey are vertical and horizontal components of the force acting on the
line at point E, E is the Young modulus of the line, w D A g is the weight of
the length unit of the line, L is the length of the line and A is the area of the cross
section.
In order to calculate the shape of the catenary line shown in Fig. 3, we
use discretisation by means of the MRFEM presented above. Forces F Ex and F Ey
are introduced into the equations of motion (30) as the generalised forces. The
coordinates of point E are as follows:
n
X n
X
xE D x0 C li C i c'i ; yE D y0 C li C i s'i : (39)
iD0 iD0
Therefore, we get

Qx0 FxE ; FyE D FxE ; Qy0 FxE ; FyE D FyE (40)
and

Qi FxE ; FyE D FxE c'i C FyE s'i ;
(41)
Q'i FxE ; FyE D FxE .li C i / s'i C FyE .li C i / c'i :
In order to define the accuracy of the method, calculations have been carried
out for a varying number n of elements into which the line with the parame-
ters L D 100 m, A D 0.25D2 , D D 0.035 m, E D 1011 N/m2 , D 2,000 kg/m3 ,
F Ex D 1,000 N, F Ey D 2,000 N has been divided.
Table 1 presents the comparison of results obtained according to formulae (37)
and the MRFEM when it has been assumed that
Cib D 0; for i D 1; : : : ; n; (42)
which means that bending flexibility is omitted. Quantity e % is defined by the

formula
ˇ ˇ
ˇ wa wm ˇ
e% D ˇ ˇ ˇ 100%; (43)
wa ˇ
where wa is calculated analytically and wm is calculated using MRFEM.

Table 1 Comparison of displacements and axial force of the catenary line

xP
s D 25 m s D 50 m s D 75 m s D 100 m
n Value e % Value e % Value e % Value e %
Analytical xP 23.478 – 42.828 – 58.175 – 70.540 –
solution
(37), (38)
yP 8.046 – 23.743 – 43.436 – 65.149 –
T 1,158.2 – 1,454.5 – 1,826.2 – 2,236.1 –
10 xP 23.493 0.366 42.760 0.158 58.181 0.101 70.541 0.014
yP 8.070 3.323 23.869 0.530 43.462 0.265 65.216 0.116
T 1,158.2 0 1,454.5 0 1,826.2 0 2,236.1 0
20 xP 23.460 0.075 42.812 0.038 58.163 0.022 70.541 0.003
yP 8.090 0.559 23.751 0.136 43.462 0.059 65.166 0.029
T 1,158.2 0 1,454.5 0 1,826.2 0 2,236.1 0
40 xP 23.473 0.018 42.826 0.009 58.172 0.005 70.540 0.001
yP 8.049 0.138 23.751 0.034 43.443 0.015 65.154 0.007
T 1,158.2 0 1,454.5 0 1,826.2 0 2,236.1 0
80 xP 23.477 0.005 42.827 0.002 58.175 0.001 70.540 0
yP 8.049 0.035 26.745 0.008 43.438 0.004 65.150 0.002
T 1,158.2 0 1,454.5 0 1,826.2 0 2,236.1 0
160 xP 23.478 0.001 42.828 0.001 58.175 0 70.540 0
yP 8.047 0.009 23.743 0.002 43.437 0.001 65.149 0
T 1,158.2 0 1,454.5 0 1,826.2 0 2,236.1 0
The analysis of results from Table 1 allows to conclude that for the MRFEM an
exactness of 0.1 % is obtained even for the division of the line into n D 40 elements.
It is characteristic, however, that the smaller the parameter s, the larger the error.
Figure 4 presents the shape of the catenary according to the analytical solution
with the points representing values obtained by the MRFEM for n D 20. Very good
compatibility of the results can be observed.
Numerical simulations have also been carried out in the case when the stiffness
coefficient of bending is defined according to the following formula:
EI
Cib D ; for i D 1; 2; : : : ; nI (44)
d
where I is inertial moment of the cross-sectional area of the rope and d D L/n.
The calculations have been carried out for n D 400, which means that the line has
been divided into segments of 0.25 m in the primary division.
Table 2 presents the results obtained when formula (42) has been used so that the
bending flexibility of the line has been omitted (line W), with the consideration of
the bending stiffness using formulae (44) (line P). For the second case it has been
additionally assumed that ' 0 0 by introduction of the generalised force:
Fig. 4 Catenary line - comparison of results
Table 2 Influence of bending stiffness

xP yP T
Cib D Cib D Cib D
Cbi D 0 EI
d
Cbi D 0 EI
d
Cbi D 0 EI
d
s [m] line W line P e% line W line P e% line W line P e%
0.25 0:248 0:250 0.63 0:028 0:005 87:2 1,006.8 1,002.8 0.40
0.50 0:497 0:500 0.64 0:058 0:013 77:0 1,007.4 1,004.0 0.33
1.00 0:993 0:999 0.65 0:121 0:039 67:8 1,008.6 1,005.2 0.23
2.00 1:983 1:996 0.64 0:260 0:121 53:3 1,011.2 1,010.1 0.11
4.00 3:955 3:976 0.57 0:593 0:388 34:6 1,017.5 1,017.2 0.08
8.00 7:856 7:889 0.42 1:473 1:219 17:3 1,034.1 1,034.1 0
16.00 15:429 15:468 0.26 4:035 3:759 6:8 1,082.4 1,082.4 0
32.00 29:349 29:346 0.18 11:858 11:561 2:5 1,230.1 1,230.1 0
64.00 51:851 51:922 0.14 34:439 34:121 0:9 1,656.4 1,656.4 0
100.00 70:540 70:632 0.13 65:149 64:820 0:5 2,236.1 2,236.1 0
Q'0 D Cib '0 : (45)
Percentage error e% has been calculated as

ˇ ˇ
ˇ wP wW ˇ
e% D ˇˇ ˇ 100%;
ˇ (46)
w P
Fig. 5 The shape of slender line W (Cbi D 0) and flexible line P (Cib D EI
d
)
where wP 2 fxP , yP , Tg is calculated for the rope with the assumption Cib D EI
d
and
wW 2 fxP , yP , Tg is calculated for the rope with the assumption Cbi D 0.
Results presented in Table 2 show that models W and P give considerably
different results for coordinate yP when coordinate s is small, while for s > 1/2 L,
those differences decrease to 1 %. Part of the graph for s 16 m showing the shapes
of lines W and P is presented in Fig. 5.
From the results presented above, it can be concluded that the consideration of
bending flexibility for long ropes is not necessary. It may be of importance in the
case of detailed analysis.
4 Numerical Example
One of the potential applications of the models presented is the analysis of failures,
which may occur during the work of cranes mounted on drilling platforms (Fig. 6).
The motion of the vessel due to the sea waves or floating away despite
disconnecting the load from the crane can cause slippage of the load from the deck.
As a result this can cause large loading of the rope system considerably larger than
nominal loading. A simplified model of such a system is presented in Fig. 7. The
system consists of:
(i) A rope with length Lr mounted to a fixed point A0 , where the origin of the
inertial coordinate system A0 xy is placed
Fig. 6 Crane lifting a load from a supply vessel
Fig. 7 Phases of motion of the load: (a) static problem; (b) movement of the load with the vessel;
(c) free vibrations of the rope with the load
(ii) A sling with length Ln connecting end E of the rope with load L, treated as a
massless spring-damping element (sde)
(iii) A load with mass mL treated as a lumped mass
For numerical analysis three essential phases are distinguished.
Phase 1 In the situation presented in Fig. 7a the initially vertical rope is displaced
under the influence of horizontal force F Ex , so that coordinate xE of the end of the
rope achieves given value x0L . In this phase mass mL is motionless, and the load is
not connected with end E of the rope. As a result, the value y0L is obtained, assuming
that
yL0 D jyE jxE Dx 0 : (47)

L
Fig. 8 Force transferred by the sling: (a) inertial force; (b) force F Ly
The task is solved as a static problem by the solution of a system of nonlinear

algebraic equations.
Phase 2 A sling connects the load with end E of the rope. Mass mL moves together
with the vessel (Fig. 7b). This phase lasts until one of the following conditions is
satisfied:
T N .case P1/ ; N 0 .case P2/ : (48)
In the first case (P1) the load and the rope are subjected to free vibrations as a
result of the friction slip. In the second case (P2) the phase of free vibrations starts,
when the vertical contact of the load to the vessel is lost. Forces N and T can be
calculated from the relations (Fig. 8):
N D mL g FyL ; T D mL vP L FxL ; (49)
where vP L D ddtv is the acceleration of the vessel and F Lx and F Ly are components of
the force in the sling.
Phase 3 At this stage (Fig. 7c) the load and rope connected by the sling are
subjected to free vibrations caused by the initial tension of the system due to the
forces of gravity.
Force FL and its components F Lx , and F Ly can be calculated in the following way:
P EL ; xE xL yE yL
F L D cL EL C bL FxL D F L ; FyL D F L ; (50)
jELj jELj
h i 12
where EL D jELj Ln , jELj D .xE xL /2 C .yE yL /2 I xE , yE are defined
in (39); xL , yL are coordinates of the load; cL , bL are stiffness and damping
P EL D d jELj :
coefficients of the sling; dt
During the second phase, coordinates xL , yL and xP L ; yPL can be calculated from
the motion of the vessel, while during the third phase these are the generalised
coordinates of the load. Hence, the equations of motion take the form
Table 3 Numerical parameters of the system

Notation Description Value Unit
L Length of the rope 80 m
D Diameter of the rope 0.04 m
E The Young modulus 11011 N/m2
Density 6,500 kg/m3
mL Mass of the load 5,000 kg
x0L Horizontal displacement 2D m
v Velocity of the vessel t 20t > 20 –1/4 t–5 m/s m/s
cL Stiffness coefficient of the sling 106 N/m
bL Damping coefficient of the sling 2102 Ns/m
Fig. 9 Trajectory of the load: (a) P1, (b) P2
mL xR L D FxL ; mL yRL D FyL : (51)
The equations of motion presented earlier have to be modified in order to take

into account force FL . Numerical parameters of the system analysed are reported in
Table 3.
Calculations in both cases, P1 and P2, have been carried out for the division of
the rope into n D 40 elements. The equations of motion have been integrated by the
Runge–Kutta method of the fourth order with the constant integration step 210–4 s.
Trajectories of mass mL in both cases are shown in Fig. 9.
The degree of rope loading during the motion of the load is well represented by
the dynamic coefficient as presented in [8]:
FL
.b/ D ; (52)
ŒmL C w .L s/ g
where FL is the force in the rope, w is the unit mass and s is the coordinate of the
position of the point.
Fig. 10 Time history of function (L): (a) case P1, (b) case P2
Fig. 11 Position of the system at different time instants: (a) P1, (b) P2
This coefficient relates the actual value of the force in the rope to the nominal
static load. Table 4 gives maximal values of coefficient for s D 0 (for point A0 )
and for s D L (point E). For both cases simulation time is T D 50 s.
Table 4 Maximal values of Case (0) (L)

dynamic coefficient
P1 3.860 3.715
P2 2.861 2.813
The values of coefficient are larger for case P1. The reason is that when the
friction force exceeds N, the vertical component of force FL is still small, and
thus, when the load loses contact to the deck of the vessel, a significant decrease in
mass mL occurs. In case P2 the force F Ey is balanced by the gravity force of the load.
Figure 10 presents fragments of time histories of function for both cases.
Figure 11 shows the system at the end of phases 1 and 2, and at time instant
t D 20 s in phase 3.
The model of discretisation of slender links is a modification of models presented

up to now and enables us to consider both bending and longitudinal flexibilities
at the same time. The model is a generalisation of the MRFEM presented in [17]
and used for vibration analysis of cranes, manipulators, linkage mechanisms and
offshore systems. Computer implementation enabled us to carry out validation
by comparing our own results with an exact analytical solution for the catenary
line. Good compatibility of results has been achieved. The model is then used for
vibration analysis in the case of sudden loss of contact of the load lifted from the
vessel deck by a crane or A–frame.
A disadvantage of the method is a full mass matrix, which makes it impossible to
use quick algorithms for solutions of linear algebraic equations with sparse matrices.
The formulation presented in [2] is without this disadvantage.
It is important to note that in the modified formulation of the RFEM it is easy
to take into account physical nonlinearities, which is essential in the case of pipe-
laying process [13]. In such a case the procedure described in detail in [11, 12, 16]
can be used.
Acknowledgments This work has been supported by National Centre of Science under the grant
MAESTRO 2, No. 2012/04/A/ST8/00738 for the years 2012–2015.
References
1. Adamiec–Wójcik, I.: Modelling Dynamics of Multibody Systems, Use of Homogenous

Transformations and Joint Coordinates. Lambert Academic Publishing, Köln (2009)
2. Adamiec–Wójcik, I., Wittbrodt, E., Wojciech, S.: Rigid finite element method in modelling of
bending and longitudinal vibrations of ropes. Int. J. Appl. Mech. Eng. 12(3), 665–676 (2012)
3. Fritzkowski, P., Kamiński, H.: A discrete model of a rope with bending stiffness or viscous
damping. Acta Mech. Sinica 27(1), 108–113 (2011)
4. Kane, T.R., Levinson, D.A.: Dynamics: Theory and Applications. McGraw Hill, New York
(1985)
5. Kłosiński, J.: Swing–free stop control of the slewing motion of a mobile crane. Control Eng.
Pract. 13(4), 451–460 (2005)
6. Maczyński, A.: Positioning and Stabilisation of Load of Jib Cranes. Publishing Office of the
University of Bielsko–Biała, Bielsko–Biała (2005) (in Polish)
7. Maczyński, A., Wojciech, S.: Dynamics of a mobile crane and optimisation of the slewing
motion of its upper structure. Nonlinear Dyn. 32(3), 259–290 (2003)
8. Maczyński, A., Wojciech, S.: Stabilization of load’s position in offshore cranes. J. Offshore
Mech. Arct. Eng. 134, 1–10 (2012)
9. Raman-Nair, W., Baddour, R.: Three-dimensional dynamics of a flexible marine riser undergo-
ing large elastic deformations. Multibody Syst. Dyn. 10, 393–423 (2003)
10. Shabana, A.A.: Dynamics of Multibody Systems. Cambridge University Press, Cambridge
(1998)
11. Szczotka, M.: Pipe laying simulation with an active reel drive. Ocean Eng. 37, 539–548 (2010)
12. Szczotka, M.: The Rigid Finite Element Method in Modelling of Nonlinear Offshore Systems.
Gdańsk University of Technology Publishing Office, Gdańsk (2011) (in Polish)
13. Szczotka, M., Wojciech, S., Maczyński, A.: Mathematical model of a pipelay spread. Arch.
Mech. Eng. 54(1), 27–46 (2007)
14. Tomski, L., Szmidla, J., Podgórska–Brzd˛ekiewicz, I., Sochacki, W., Kukla, S., Posiadała, B.,
Przybylski, J., Uzny, S.: Vibrations and Stability of Slender Systems. WNT, Warsaw (2004) (in
Polish)
15. Wittbrodt, E., Adamiec–Wójcik, I., Wojciech, S.: Dynamics of Flexible Multibody Systems:
Rigid Finite Element Method. Springer, Berlin (2006)
16. Wittbrodt, E., Szczotka, M., Maczyński, A., Wojciech, S.: Rigid Finite Element Method in
Analysis of Dynamics of Offshore Structures. Springer, Berlin (2013)
17. Wojciech, S.: Dynamics of Planar Linkage Mechanisms with Considerations of Both Flexible
Links and Friction as Well as Clearance in Joints. Publishing Office of the Technical University
of Łódź, Łódź (1984) (in Polish)
18. Xu, X., Wang, S.: A flexible-segment-model-based dynamics calculation method for free
hanging marine risers in re-entry. China Ocean Eng. 26, 139–152 (2012)
19. Zienkiewicz, O.C., Taylor, R.L.: The finite element method. In: Solid Mechanics, vol. 2.
Butterworth–Heinemann, Oxford (2000)
Dynamical Response of a Van der Pol–Duffing
System with an External Harmonic Excitation
and Fractional Derivative
Arkadiusz Syta and Grzegorz Litak
Abstract We examine the Van der Pol–Duffing system with external forcing and a
memory possessing a fractional damping term. The system exhibits broad spectrum
of nonlinear behavior including transitions from the periodic to nonperiodic motion.
Replacing a first-order derivative damping term by a fractional damping one, we
include to the system memory effect which increases the dimension of the dynam-
ical system. As a consequence of such assumptions, the quantitative nonlinear
analysis meets some limitations in this case. Instead of the well-known Lyapunov
exponent treatment, we advocate to use the 0–1 test that combines both statistical
and frequency properties of the attractor but does not depend on the dimension of
the state space. The results have been confirmed by quantitative nonlinear analysis:
bifurcation diagrams, phase portraits, Poincare sections, and the maximal Lyapunov
exponent estimated in the limited two-dimensional phase space.
1 Introduction
Recently, dynamical systems with fractional damping have brought the attention of
scientists from different research fields with the expectation to describe complex
system behavior and/or complex material dynamical responses [2, 6, 13–15, 18–20].
This approach differs from the linear and nonlinear damping given by a function
of velocity [3, 18]. Introducing fractional damping to a dynamical system is made
by replacing an integer order derivative with a fractional operator. There are many
definitions of fractional operator but the most common are Grünwald–Letnikov
and Riemman–Liouville. Both of them are derived from general fractional q order
operator; more precisely, the first one represents the q order derivative, while
A. Syta ()
Lublin University of Technology, Institute of Technological Systems of Information,
Nadbystrzycka 36, PL-20-618 Lublin, Poland
e-mail: a.syta@pollub.pl
G. Litak
Department of Applied Mechanics, Lublin University of Technology, Nadbystrzycka 36,
PL-20-618 Lublin, Poland
e-mail: g.litak@pollub.pl

140 A. Syta and G. Litak
the second represents the q fold integral. It is worth to note that the Grünwald–
Letnikov definition requires the function to be m C 1 continuously differentiable
while the Riemman–Liouville definition, only to be integrable which may play the
role in scope of application. However, for function defined by Grünwald–Letnikov
class both definitions are equivalent. Below, we show briefly how to construct
the fractional derivative in the meaning of Grünwald–Letnikov definition which is
fairly straightforward from a numerical point of view. Let us consider the Leibniz
definition of the first and second derivative:
f .t / f .t h/
f 0 .t / D lim (1)
h!0 h
f 0 .t / f 0 .t h/
f 00 .t / D lim
h!0 h
f .t / 2f .t h/ C f .t 2h/
D lim
h!0 h2
Now, basing on Eq. (1) one can write a general form of the nth order derivative:
!
1 X n
n
f n .t / D lim n .1/j f .t j h/ n 2 N: (2)
h!0 h j
j D0
By generalizing the integer order in the derivative in Eq. (2) one can obtain the
q
Grünwald–Letnikov definition of fractional operator a Dt f .t / [17]:
Œ ta !
h
d qf 1 X q q
D lim q .1/j f .t j h/ Da Dt f .t /; (3)
dt q h!0 h
j D0
j
where q > 0, and binomial coefficients can be extended to complex numbers by use
of the Euler Gamma function:
!
q qŠ .q C 1/
D D I (4)
j j Š.q j /Š .j C 1/ .q j C 1/
a pair of square brackets Œ: appearing in the upper limit of the sum denotes the
integer part, while a is the length of the memory. Generally, dynamical systems
with memory effects pose long transients, but in this case it is possible to shorten the
length of the memory (a) in Eq. (3) using the short memory principle [16, 17]. Such
a length, introduced to the system, reduces the computational effort of simulations
while generally preserving the system dynamics. Thus, Eq. (3) becomes
!
1 X
ŒN.t/
q j q
L Dt f .t / D lim q .1/ f .t j h/; (5)
h!0 h j
j D0
where N.t / D min. tL

h
; Lh /.
Dynamical Response of a Van der Pol–Duffing System with an External. . . 141
2 The Model
We start from the standard Van der Pol–Duffing equation decomposed into the set
of two differential equations of first degree:
dx
D y.t /;
dt (6)
dy
D ˛x.t / ˇx.t /3 C .1 x.t /2 /y.t / C A cos .!t /;
dt
where ˛ D 0:5; ˇ D 0:5; D 0:1; A D 0:15, and the frequency of external

excitation, !, is a control parameter. This model shows many interesting nonlinear
features, e.g., periodic type solution with period doubling sequence, and transitions
to quasiperiodic or chaotic solutions.
Let us compare the system response for various values of the excitation frequency
! D f0:45; 0:515; 0:58g with same initial conditions .x.0/; y.0// D .0:0; 0:0/. We
calculated the solutions using Runge–Kutta of fourth order with fixed time step
taking 200 periods (each with 200) after the transient with length of the memory
a D 5; 000.
Analyzing Fig. 1 one can see different types of motion: from quasiperiodic
(a) with Poincare points located on circle like curve (closed curve would be obtained
for longer computational time) to nonperiodic (b) with clouded Poincare points and
periodic one (c) with single isolated Poincare point. Now, to study the influence of
fractional damping to the model (6) we replaced first-order derivative damping with
fractional damping:
q
L Dt x.t / D y.t /
(7)
1
L Dt y.t / D ˛x.t / ˇx.t /3 C .1 x.t /2 /y.t / C A cos .!t /:
The set of equations can be written in the discretized form by the following
fractional order Newton–Leipnik algorithm [16]:
X
N 1
.q/
x.tk / D y.tk1 /hq cj x.tkj / (8)
j D1
y.tk / D Œ˛x.tk / ˇx.tk /3 C .1 x.tk /2 /y.tk1 / (9)

C A cos .!tk1 /h C y.tk1 /;
.q/
where h is an integration step and coefficients cj satisfy the following recursive
relations:

.q/ .q/ 1Cq .q/
c0 D 1; cj D 1 cj 1 : (10)
j
Fig. 1 Phase portraits and Poincare sections (as introduced in [1]) of system (6) solutions for
(a) ! D 0:45, (b) ! D 0:52, (c) ! D 0:58
Figure 2 presents bifurcation diagrams obtained for various fractional order damping
q D f0:9; 0:95; 1:0; 1:05; 1:1g and initial conditions .x.0/; y.0// D .0:0; 0:0/.
Comparing the bifurcation diagrams one can see a change of system response
even in a short interval of !: for ! 2 .0:35 W 0:7/ (Fig. 2a, b) there dominate
periodic solutions with narrow nonperiodic regimes. When q increases (Fig. 2c) it is
changing to quasiperiodic (! 2 .0:35; 0:48/) and chaotic motion (! 2 .0:48; 0:55/).
Finally, for q D f1:05; 1:1g (Fig. 2d, e) it becomes quasiperiodic or chaotic.
To quantify obtained results we use the 0–1 test [4, 7, 8, 10–12] with single
time series and the maximal Lyapunov exponent (MLE) with projection into two
dimension [9, 21]. The 0–1 test for chaos detection uses spectral and statistical
properties of the single time series, while MLE uses geometrical properties of
phase space. Both methods can distinguish between periodic and chaotic motion,
but because of the memory effect introduced by fractional damping the phase space
of the system has infinite dimension (a D 5; 000) and in computing MLE only
projection to finite and low dimension can be used. As a consequence we don’t
possess information about system dynamics in higher dimension. In this case the
0–1 test might be considered and we will show a brief description of it in a few
steps. The first one is conversion of the coordinates from .x; y/ to the new set .p; q/
defined as follows:
X
n X
n
p.n/ D xQ j cos .jc/; q.n/ D xQ j sin .jc/; (11)
j D1 j D1
where xQ D ŒxQ 1 ; xQ 2 ; xQ 3 ; : : : is a discrete time series sampled from the originally

simulated x using one forth of excitation period T =4 (T D 2=!) which
corresponds to the nodal autocorrelation function of excitation harmonic term
Acos.!t /. Equation (11) corresponds to the Fourier transform (in the limit of
larger n) for chosen frequency c where c 2 .0; /. The next step is calculation
of the mean square displacement (MSD) in the plane spanned by new coordinates
[p, q]:
1 X˚
nj

MSD.c; j / D Œp.i C j / p.i /2 C Œq.i C j / q.i /2 (12)
n j iD1
where 0 j n (in practice n=100 j n=10). It is bounded for regular

dynamics or unbounded for chaotic dynamics [4, 7, 8, 10–12]. The final quantity
K is obtained as a asymptotic growth rate of MSD (here given by the correlation
method):
CovŒj; MSD.c; j /
K.c/ D p ; (13)
CovŒj; j CovŒMSD.c; j /; MSD.c; j /
where j is based on series of natural numbers, j D n=100; n=100C1 : : : ; n=10, and

Cov[x,y] denotes corresponding covariance of two series where the same arguments
Fig. 2 Bifurcation diagrams with the sampling ! D 0:00025 for different order of the damping:
(a) q D 0:9, (b) q D 0:95, (c) q D 1:0, (d) q D 1:05, (e) q D 1:1. Note that Fig. 2d, e shows
quasiperiodic system response in the whole investigated frequency interval
x D y mean variance while for different (x D j and y D MSD.c; j /) it can be

expressed in terms of the expectation value E[.]:
CovŒj; MSD.c; j / D EŒŒj EŒj ŒMSD.c; j / EŒMSD.c; j /: (14)
The K 0 indicates periodic motion while K 1 indicates nonperiodic behavior.

The value of the quantificator might vary with respect of c, so we choose 100 values
of c equally spaced from the interval .0:1; 0:1/ and computed K as the median;
see Fig. 3.
Comparing the K values (Fig. 2) to the bifurcation diagrams (Fig. 3) one can
see that this indicator stays close to 0 for periodic or quasiperiodic motion while it
oscillates near 1 for chaotic motion. Interestingly, for fractional order of the damping
q D f1:05; 1:1g in considered interval of excitation frequency (! 2 .0:35; 0:85/),
the motion is quasiperiodic.
We have also calculated MLE which is commonly used to distinguish between
different types of response of dynamical systems, but in the case of our system this
indicator has no direct meaning because the history effect provided by fractional
damping makes the dimension of the system undetermined. Having that in mind,
one cannot use the well-known Wolf [21] method (with linearized system) nor
Kantz [9] algorithm (with phase space reconstructed from time series). Instead of
that we measured the distance d.i / between reference and test orbits projected
in two dimension .x; y/ and disturbed by some small initial perturbation d0 .i /,
i D 1; : : : ; N , where i denotes the subsequent interval which is smaller compared
to the excitation period. In that case, the value of MLE can be approximated by
1 X
N
MLE D ln .d=d0 / : (15)
t iD1
The corresponding results are presented in Fig. 4.

By analyzing MLE values, one notices fairly good agreement with K values
obtained for q D f0:9; 0:95; 1:0; 1:05g, but for q D 1:1, MLE is grater then 0:2
for all ! 2 .0:35; 0:85/ which indicates chaotic motion and is contrary to the K
values. The reason might be connected with lack of information about the distance
between trajectories in additional dimensions introduced by the memory effect of
the fractional damping term. For better clarity, Fig. 5 shows .x; y/ coordinates of
the phase space for fractional damping order q D 1:05.
3 Conclusions
Our results show that in the case of Van der Pol–Duffing potential with a fractional
damping, various bifurcations can occur. However, the most common solutions are
quasiperiodic with short intervals of chaotic solutions. These effects agree with
Fig. 3 K versus ! with the sampling ! D 0:00025 for (a) q D 0:9, (b) q D 0:95, (c) q D 1:0,
(d) q D 1:05, (e) q D 1:1. K 0 indicates regular while K 1 chaotic behavior
Fig. 4 MLE versus ! with the sampling ! D 0:00025 for (a) q D 0:9, (b) q D 0:95, (c) q D
1:0, (d) q D 1:05, (e) q D 1:1. Note that MLE > 0 implies chaos. In many cases, it is not
consistent with bifurcation diagrams (Fig. 2) and K values (Fig. 3)
Fig. 5 Phase portraits and Poincare sections of Eq. (7) solutions for (a) ! D 0:45, (b) ! D 0:52,
(c) ! D 0:58 and q D 1:05
expectations in fairly large dimension of the dynamical system with delays [5]. It is
not surprising that the simply defined MLE detects chaos wrongly in particular cases
(see Fig. 4e). This must be an effect of higher dimension. Note that for the same
quasiperiodic solution, the 0–1 test gives the correct state identification (see Fig. 3e).
The results have been also confirmed by phase portraits and Poincare series. On the
other hand, in the system with memory the role of transient solutions increases. In
the specific situations transient chaotic motion can be also stabilized. However, to
tell more about such a tendency, more systematic studies should be done to explain
the evolution of the corresponding basins of attractions.
Acknowledgements This paper was supported by the Polish National Science Center under the
grant agreement No. 2012/05/B/ST8/00080.
References
1. Awrejcewicz, J., Olejnik, P.: Stick-slip dynamics of a two-degree-of-freedom system. Int.

J. Bifurcat. Chaos 13, 843–861 (2003)
2. Barbosa, R.S., Machado, J.A.T., Vinagre, B.M., Calderon, A.J.: Analysis of the Van der Pol
oscillator containing derivatives of fractional order. J. Vib. Control. 13, 1291 (2007)
3. Borowiec, M., Litak, G., Syta, A.: Vibration of the Duffing oscillator: effect of fractional
damping. Shock Vib. 14, 29 (2007)
4. Falconer, I., Gottwald, G.A., Melbourne, I., Wormnes, K.: Application of the 0–1 test for chaos
to experimental data. SIAM J. Appl. Dyn. Syst. 6, 395 (2007)
5. Farmer, J.D.: Chaotic attactors of an infinite-dimensional dynamical system. Physica D 4, 366
(1982)
6. Gao, X., Yu, J.: Chaos in the fractional order periodically forced complex Duffing’s systems.
Chaos Solitons Fractals 24, 1097 (2005)
7. Gottwald, G.A., Melbourne, I.: A new test for chaos in deterministic systems. Proc. Roy. Soc.
A 460, 603 (2004)
8. Gottwald, G.A., Melbourne, I.: Testing for chaos in deterministic systems with noise. Physica
D 212, 100 (2005)
9. Kantz, H.: A robust method to estimate the maximal Lyapunov exponent of a time series. Phys.
Lett. A 185, 77 (1994)
10. Litak, G., Syta, A., Wiercigroch, M.: Identification of chaos in a cutting process by the 0–1
test. Chaos Solitons Fractals 40, 2095 (2009)
11. Litak, G., Syta, A., Budhraja, M., Saha, L.M.: Detection of the chaotic behaviour of a bouncing
ball by the 0–1 test. Chaos Solitons Fractals 42, 1511 (2009)
12. Litak, G., Schubert, S., Radons, G.: Nonlinear dynamics of a regenerative cutting process.
Nonlinear Dyn. 69, 1255 (2012)
13. Machado, J.A.T., Silva, M.F., Barbosa, R.S., Jesus, I.S., Reis, C.M., Marcos, M.G., Galhano,
A.F.: Some applications of fractional calculus in engineering. Math. Prob. Eng. 2010, 639801
(2010)
14. Mikens, R.E., Oyedeji, K.O., Rucker, S.A.: Analysis of the simple harmonic oscillator with
fractional damping. J. Sound Vib. 268, 842 (2003)
15. Padovan, J., Sawicki, T.: Nonlinear vibration of fractionally damped systems. Nonlinear Dyn.
16, 321–336 (1998)
16. Petras, I.: Fractional-Order Nonlinear Systems: Modeling, Analysis and Simulation. Springer,
New York (2010)
17. Podlubny, I.: Fractional Differential Equations. Academic, San Diego (1999)
18. Rossikhin, Y.A., Shitikova, M.V.: Application of fractional calculus for dynamic problems of
solid mechanics: novel trends and recent results. Appl. Mech. Rev. 63, 010801 (2010)
19. Seredynska, M., Hanyga, A.: Nonlinear differential equations with fractional damping with
application to the 1dof and 2dof pendulum. Acta Mech. 1760, 169 (2005)
20. Sheu, L.J., Chen, H.K, Tam, L.M.: Chaotic dynamics of the fractionally damped duffing
equation. Chaos Solitons Fractals 32, 1459 (2007)
21. Wolf, A., Swift, J.B., Swinney, H.L.: Determining Lyapunov exponents from a time series.
Physica D 16, 285 (1985)
Vortex Structure Around the Cylinder at a Flow
of Viscous Fluid
Rustyam G. Akhmetov and Ruslan R. Kutluev
Abstract The problem of a stationary viscous incompressible fluid flow around

the cylinder has been analyzed by means of the asymptotic methods. The fluid
flow equations in the variables “stream function-a vortex” are considered. One
component of a vortex vector has remained in case of two-dimensional motion.
Having applied the matching method the vortex asymptotics has been investigated in
an interface near the cylinder. The equation of an interior boundary layer for stream
function has been made with the help of using the method of matched asymptotic
expansions and a matching condition with the solution for vortex. The received
equation is investigated by means of numerical methods for great values of Reynolds
number.
1 Introduction
In case of two-dimensional motion equation of the fluid flow is set down in the
variables “stream function-a vortex.’ The equation of incompressibility is carried
out as the result of introduction of stream function . So for a stationary problem in
the dimensionless variables we have the equations (see, e.g., [14], Chap. XI, p. 471,
Eq. (135), (136))
1 @˝ @˝
˝ D u Cv ; (1)
Re @x @y
D ˝; (2)
where is the Laplace operator, Re is the Reynolds number, and ˝ is the scalar
function:
R.G. Akhmetov () • R.R. Kutluev

Bashkortostan State Pedagogical University, Faculty of Physics and Mathematics,
ul. Oktya’brskoy Revolutsii 3a, 450000 Ufa, Russia
e-mail: akrust@mail.ru; rus4652@yandex.ru

152 R.G. Akhmetov and R.R. Kutluev

@v @u @ @ @ @ @ @
˝ D ˝z D D D ; u D ; vD :
@x @y @x @x @y @y @y @x
One of the ways of equations researching (1), (2) is the transition to polar
coordinates r; '. For stream function we specify conditions
@
D D 0 for r D 1 (3)
@r
and on infinity we specify conditions
1@ @
! cos '; ! sin ' for r ! 1: (4)
r @' @r
For a vortex on infinity
˝ ! 0 for r ! 1 (5)
is obtained. Tom or Woods approximate conditions are usually set at numerical

calculations (see [14], Chap. XI, p. 472, formulas (137), (138)). Problems analogous
to (1)–(5) and a broader class of problems were considered by many authors (e.g.,
see [1, 5, 6, 10–12] and references).
2 Boundary Layer Near the Cylinder
In view of this work the method of the research consists in the following. In the
capacity of the first approximation of stream function outside the wake behind the
cylinder (in area r sin2 ' > const ) we will consider

1 3 ln r ln r
1 .r; '/ D r C 2 sin '; where (6)
r r r
1 D O.r 3 sin '/; as r ! 1 and 1 .1; '/ D. 0

1 /r .1; '/ D 0:
Thus, Re D U1 d=, where d is the diameter of the cylinder and is the viscosity
coefficient.
Then, we are to consider that Re 2 Œ10; A. Further, calculations show that
A D O.104 /. We investigate a vortex asymptotics ˝ applying a method of matched
asymptotic expansions [15] near the cylinder.
Equations
(1) and (2) are considered in
[3], when stream function is .r; '/ D r r 1 sin ' (in area r sin2 ' > const ).
The small parameter 3 D 1=Re is introduced for convenience, and (1) is
rewritten as
Vortex Structure Around the Cylinder at a Flow of Viscous Fluid 153

@2 1 @ 1 @2 1 @ @ 1 @ @
3
C C C D 0; (7)
@r 2 r @r r 2 @' 2 r @' @r r @r @'
! 0; r ! 1; (8)
where .r; '/ D ˝.x; y/. The function 1 (see (6)) at r D 1 has the asymptotic
expansion in a series

1 D .r 1/2 C O .r 1/3 sin ': (9)
The asymptotics is sought in the form of a series
D 0 .; '/ C 1 .; '/ C ; where D .r 1/=:
Then from (7) in variables ; ' we receive the equation for the main member
@2 0 @0 @0
2 cos ' C 2 sin ' D 0: (10)
@2 @ @'
p R p
In (10) we should substitute variables x D sin ', D 1
2 ' sin ydy (see [8],
Chap. 3, p. 110, 111). Then the equation
@2 0 @0
2
x D0 (11)
@x @
is obtained.
The solution of (11) satisfying condition (8) has the form

1 1 x3
0 D c 1 ; ; where (12)
3 3 9
Z 1
.˛; x/ D e y y ˛1 dy
x
and the constant c is determined from the boundary conditions for the 0 where
r D 1. To find the approximate condition at the border, where r ! 1 for 0 , we
should use (2) in the vicinity of the boundary. Then, taking into account formulas
(9), (12), we have
@2 1 .1; '/
D 2 sin ' 0 .0; / D c at r ! 1:
@r 2
It is followed the formula is:

1 1 1 x3
0 .x; / 2 sin ' ; at ! 0 .' ! 0/; where (13)
3 3 9
Z p
1
0 D sin ydy 1; 19814:
2 0
3 Interior Boundary Layer
Behind the cylinder there is an additional boundary layer. The asymptotics of the
solution is regarded in terms of the variables , s D r 1. And for the main
member we receive the equation
@
D 0: (14)
@s
From (14) we receive D . /. Then from (2), we have
D . /: (15)
Matching condition with solution (13) for . / is represented as

!
1=2
1 .r; '/

. / 0 D O 2 for ' 2 O./: (16)
.90 /1=3
Thus it is natural to obtain the explicit function . /

3=2
1 1
. / D 2 1 ; 3 for ' 2 O./: (17)
3 3 90
Equation (15) and formula (17) in the interior boundary layer for the flow function
are written out in work [2].
The equation similar to (15) is usually written out during the study of vortex
motions of ideal fluid D f . / (see, e.g., [13], Chap. 7, Sect. 165). Besides
f . / is an arbitrary function. Our consideration is different as the (15) is written out
in an internal around the cylinder by a streamlined flow of a viscous incompressible
fluid boundary layer, using the method of matched asymptotic expansions [15],
system (1), (2), and matching condition (16) with solution (13).
4 Numerical Solution
For the solution of (15) where the right part looks like (17), taking into account (3),
(4), it is necessary to set additional boundary conditions @ =@' D 0; for ' D 0,
and matching condition with solution (13) at r 1 D O.Re 1=3 /, ' D O.Re 1=3 /.
Let us consider a special case, when D .r/. In this case from (15) we receive
the equation
00 1 0
.r/ C .r/ D . .r//: (18)
r
Taking into account expression (16) and setting initial conditions in a point r0
0
.r0 / D 01 ; .r0 / D 02 : (19)
We receive the solution by numerical methods (see Figs. 1, 2, 3, 4, and 5).

0
The existence of a solution to problem (18) and (19) at .r/ D 0 follows
from Poincare–Bendixson theorem (see [9], Chap. I, p. I.16). Phase trajectories of
the solution to problem (18) and (19) (see Fig. 1) are natural, taking into account
0
existence in the left part of the component 1r .r/ (see, e.g., [4], Chap. 3, Sect. 10).
This is a special point of focus type.
Fig. 1 Phase trajectories for r0 D 1; 1, 01 D 0; 4, 02 D 0; 01, Re D 300

Fig. 2 Dependence of the solution on r at Re D 300
Fig. 3 Phase trajectories for r0 D 1; 1, 01 D 3, 02 D 0; 1, Re D 10
Remark
Equations (1) and (2) are written in dimensionless variables (see [14], Chap. XI,
p. 472, formulas (137), (138)). It follows that function (13) is also dimensionless for
.r/ D O. 2 /.
In the work [5], the stationary problem flow of the circular cylinder for Reynolds
numbers in an interval from 1/8 to 100 is investigated. The results of numerical
calculations are presented. Formula (13) gives a quite satisfactory approximation
compared with the results of numerical calculation [5] outside the wake of a cylinder
Re 2 Œ10I 100 (see, e.g., [5], p. 195, Fig. 8 and Re D 100).
Fig. 4 Phase trajectories for r0 D 1; 1, 01 D 0; 8, 02 D 0; 01, Re D 100
Fig. 5 Phase trajectories for r0 D 1; 1, 01 D 0; 02, 02 D 0; 005, Re D 2 104
In this work [6, 7] a two-dimensional problem near the cylinder for Reynolds
numbers in an interval from 2 104 to 106 was considered. The phenomenon of the
pressure crisis in the boundary layer and the separation of the boundary layer were
analyzed. The formation of the boundary layer (and the separation of the boundary
layer) starts near the midship section of the cylinder (see [7], Fig. 2a). It is noted
that because of the rise of Reynolds number, the boundary layer becomes thinner in
case first vortex ([7], Fig. 2). This article demonstrates that the increase of Re leads
to a reduction of the size of the boundary layer (see Figs. 3, 4, and 5, component ).
Conclusion
Outside the wake of a cylinder the flow function is known and it corresponds to the
flow around the cylinder by the viscous fluid; the equation for the vortex component
near the cylinder is investigated. It is illustrated that there is an interior boundary
layer where the vortex component part depends only on the flow function. The
structure of the vortex component is found by means of the method of matched
asymptotic expansions. The interior boundary layer equation for the flow function
is investigated by means of numerical methods at great values of Reynolds number.
The properties of the flow function are studied in this boundary layer. The solution
in this boundary layer has a special point of focus type.
References
1. Afendikov, A.L., Babenko, K.I.: Mathematical modelling of turbulence in flows of a viscous

incompressible fluid (Russia). Math. Model. Moscow 1(8), 45–74 (1989)
2. Akhmetov, R.G.: About a matching method in the problem of viscous fluid flow around a
cylinder (Russia). In: Theory of Functions, Its Applications and Related Issues. Proceedings
of International Conference, 22–28 Aug 2013. Proceedings of the Mathematical Center of N.I.
Lobachevsky, vol. 46, pp. 106–108. Kazan University, Kazan (2013)
3. Akhmetov, R.G., Kutluev, R.R.: About vortex structure at the cylinder flow viscous fluid
(Russia). In: Fluxes and Structures in Fluids. Proceedings of International Conference, pp.
10–13, 25–28 June 2013. MAKS Press, Saint-Petersburg (2013)
4. Arnold, V.I.: Additional Chapters of the Theory of the Ordinary Differential Equations
(Russia). Nauka, Moscow (1978)
5. Babenko K.I., Vvedenskaya N.D., Orlova M.G.: Calculation of the steady flow of a viscous
fluid past a circular cylinder. Zhurnal Vychislitel’noi Matematiki and Matematicheskoi Fiziki
15(1), 183–196 (1975) [Translate from Russian to English in USSR Comput. Math. Math.
Phys. 15(1), 176–190 (1975)]
6. Dynnikova, G.Ya.: Calculation of flow around a circular cylinder on the basis of two-
dimensional Navier-Stokes equations at large Reynolds numbers with high resolution in a
boundary layer. Doklady Akad. Nauk 422(6), 755–757 (2008) [Translate from Russian to
English in Doklady Phys. 53(10), 544–547 (2008)]
7. Dynnikova, G.Ya.: Fast technique for solving the N-body problem in flow simulation by vortex
methods. Zhurnal Vychislitel’noi Matematiki and Matematicheskoi Fiziki 49(8), 1458–1465
(2009) [Translate from Russian to English in Comput. Math. Math. Phys. 49(8), 1389–1396
(2009)]
8. Gupalo, Yu.P., Polyanin, A.D., Ryazantzev, Yu.S.: Mass and Heat Exchange of Reactive
Particles with the Flow (Russia). Nauka, Moscow (1985)
9. Hairer, E., Norsett, S.P., Wanner, G.: Solving Ordinary Differential Equations, Nonstiff
Problems. Springer, Berlin (1987)
10. Krasnikov, Yu.G., Solovyev, V.R.: Finding of approximate analytical solutions of the equations
of Navier-Stokes for a stationary flow of the cylinder incompressible liquid (Russia). News
Acad. Sci. Mech. Liquid Gas 4, 22–33 (1999)
11. Ladyzhenskaya, O.A.: Mathematical Problems of the Dynamics of a Viscous Incompressible
Fluid (Russia). Nauka, Moscow (1970)
12. Ladyzhenskaya, O.A.: Sixth Problem of the Millennium: Navier–Stokes Equations, Existence
and Smoothness, vol. 58(2), pp. 45–78. Uspekhi Matematicheskikh Nauk, Moscow (2003)
[Translate from Russian to English in Russian Math. Surv. 58(2), 251–286 (2003)]
13. Lamb, G.: Hydrodynamics. Gostekhizdat, Moskow (1947) [English original text in Cambridge
University Press, New York (1932)]
14. Loytsyansky, L.G.: Mechanics of liquids and gases (Russia). Bustard, Moskow (2003)
15. Van Dyke, M.: Perturbation methods in fluid mechanics. World, Moskow (1967) [English
original text in Academic Press, New York (1964)]
Asymptotic Analysis and Limiting Phase
Trajectories in the Dynamics of Spring
Pendulum
Jan Awrejcewicz, Roman Starosta, and Grażyna Sypniewska-Kamińska
Abstract Spring pendulum is a widely discussed two degree-of-freedom (DOF)

mechanical systems in numerous references. In this paper the asymptotic approach
and limiting phase trajectories (LPT) have been applied to analyze the two DOF
mathematical model of a spring pendulum. The LPT and multiple timescale
(MTS) methods are effective tools of the investigation of non-linear systems. Some
interesting and important aspects of dynamics of the system are discussed. The
main attention is focused on the non-steady-state vibrations when the energy is
intensively exchanged. Then with increasing values of the selected parameters, a
sudden change in the character of vibrations is observed. These phenomena are
very well described by the LPT. The method allows to determine the critical values
of the parameters responsible for the mentioned transitions. Our analytical studies
are verified by numerical calculations.
1 Introduction
The steady-state vibrations are, in general, mainly observed in engineering practice.

However, in some cases of sharp resonance, the transient stage of the oscillatory
process and its relaxation can last a long time. Energy exchange and non-stationary
processes appear in many dynamical systems and they are of great interest of many
researchers. This problem has been widely discussed in [2, 6]. It is usually studied
numerically due to occurred essential mathematical difficulties [2]. However, in
recent years, one may observe a great interest in successful application of modern
asymptotic methods to engineering-oriented problems [1, 4]. In particular, a novel
J. Awrejcewicz
Department of Automatics and Biomechanics, Technical Universityof of Łódź,
ul. Stefanowskiego, 90-924 Łódź, Poland
R. Starosta • G. Sypniewska-Kamińska ()
Poznań University of Technology, Institute of Applied Mechanics,
ul. Piotrowo 3, 60-965 Poznań, Poland
e-mail: roman.starosta@put.poznan.pl; grazyna.sypniewska-kaminska@put.poznan.pl

idea for an effective study of non-linear dynamical systems is linked with a concept
of the so-called limiting phase trajectories (LPT) (see [3]).
The analysis of the non-linear spring pendulum is carried out in the paper.
The unsteady-state oscillations near resonance are discussed. The pendulum-type
mechanical systems with non-linear and parametric interactions exhibit a rich
behaviour, and hence their understanding and prediction are important both from
a point of view of the theory and application. Pendulums are relatively simple
systems; neverthelessthey can be used to simulate the dynamics of a wide variety of
engineering devices and machine parts. The coupling of the equations of motion
causes possibility of autoparametric excitation and is connected to the energy
exchange between modes of vibrations [8]. The energy transfer is well known in
dynamics of multi-degree-of-freedom systems and is widely discussed by many
authors [5, 7]. A key role either for theoretical- or application-oriented analysis is
played by prediction and determination of thresholds (critical set of parameters),
where transitions of system dynamics take place from a periodic quasi-linear
to strongly non-linear behaviour. It can be observed in the neighbourhood of a
resonance. Such critical value of non-linear parameter of the spring pendulum is
determined in the paper.
2 Formulation of the Problem
Let us consider the planar motion of a mass attached to the massless non-linear
spring. The examined system is shown in Fig. 1.
The Lagrangian of the system is given by
1 1 1
L D mg cos .L0 C Z/ k1 Z 2 k2 Z 4 C m ZP 2 C .L0 C Z/2 P 2 (1)
2 4 2
where m is the mass of the pendulum, L0 is the length of the nonstretched spring, k1
and k2 are the stiffness coefficients, g is the Earth’s acceleration and Z(t) and (t) are
generalized coordinates (see Fig. 1). The magnitudes of the forces F1 and F2 acting
on the mass along and transversally to the pendulum are F1 (t) D F1 cos(1 t) and
F2 (t) D F2 cos(2 t). Forces of linear viscous damping are considered to be present
in both longitudinal and swing motions of the pendulum (C1 and C2 are viscous
coefficients).
The equations of motion have been obtained using Lagrange equations of the
second type. Their non-dimensional form follows
zRCc1 zPCzC˛ z3 C 3˛ z2r z C 3˛zr z2 C w2 .1 cos '/ .z C 1/ 'P 2 Df1 cos .p1 /
(2)
Asymptotic Analysis and Limiting Phase Trajectories in the Dynamics. . . 163
Fig. 1 The spring pendulum

.z C 1/ .z C 1/ 'R C w2 sin ' C 'P .c2 C 2Pz/ D .z C 1/ f2 cos .p2 / (3)
where z D Z/L,p L D L0 C Zr , c1 D C1 /m! 1 , c2 D C2 /L2 m! 1 , w D ! 2 /! 1 , !2 D

p
g=L, !1 D k1 =m; p1 D 1 /! 1 , p2 D 2 /! 1 , f1 D F1 /Lm! 21 , f2 D F2 /Lm! 21 and
dimensionless time D t! 1 . Now z and are functions of , whereas zr denotes the
elongation of the spring at the static equilibrium position and fulfils the equation
˛ z3r C zr D w2 (4)
The second equation of motion (3) gives, among others, a trivial solution z D 1
which has no physical meaning and should be rejected.
Vibrations of the system are investigated in the neighbourhood of the equilibrium
position; hence, the trigonometric functions can be substituted by their power series
approximations
sin ' ' ' 3 =6; cos ' 1 ' 2 =2 (5)
which limit the angle about to /6 with precision of four significant digits.
The above remarks lead to a new form of the equations of motion
1
zR C c1 zP C z C ˛ z3 C 3˛ z2r z C 3˛ zr z2 C w2 ' 2 .z C 1/ 'P 2 D f1 cos .p1 /
2
(6)

'3
.z C 1/ 'R C w
2
' C 'P .c2 C 2Pz/ D f2 cos .p2 / (7)
6
Let us assume the homogeneous initial conditions
z.0/ D 0; zP.0/ D 0; '.0/ D 0; '.0/

P D0 (8)
The above initial problem described by the coupled and non-linear equations is
investigated.
2.1 Complex Representation of the Problem
Let us introduce the phase space coordinates zP ./ D v ./ and 'P ./ D ˇ ./ into
(6)–(8) and then rewrite the problem in the form
1
vP C c1 v C z C ˛ z3 C 3˛ z2r z C 3˛zr z2 C w2 ' 2 .z C 1/ ˇ 2 D f1 cos .p1 / ;
2
(9)

P '3
.z C 1/ ˇ C w '
2
C ˇ .c2 C 2$/ D f2 cos .p2 / ; (10)
6
z.0/ D 0; v.0/ D 0; '.0/ D 0; ˇ.0/ D 0: (11)
Then, the approach proposed in the paper [3] is applied. Introduction of the
complex-valued functions
‰z D v C i z; ‰' D ˇ C i w '; ‰ z D v i z; ‰' D ˇ i w ' (12)
converts the problems (9)–(11) to the complex form

1
‰Pz C‰ P C c1 ‰ C ‰ 1 i ‰ ‰ C 1 i ˛ ‰ ‰ 3
2 z 2 z z 2 z z 8 z z
3 2 1 2
2 ˛ zr ‰z ‰ z 4 ˛ zr ‰z ‰ z 8 ‰' ‰ '
3 2 (13)
2 1
C 14 ‰' C ‰ ' 2
i ‰z ‰ z 1 D f1 cos .p1 / ;

P C 1 ‰ C ‰ c C ‰ C ‰
1
1 12 i ‰z ‰ z ‰P' C‰ ' ' ' 2 z z
2 3
2
(14)
i .‰' ‰ ' / i .‰' ‰ ' /
C w2 48w 3 2w
D f2 cos .p2 / ;
‰z .0/ D 0; ‰' .0/ D 0; ‰ z .0/ D 0; ‰ ' .0/ D 0 (15)
The complex conjugate equations similar to (13) and (14) are also derived. They
and all the consequent formulas are not written for greater clarity.
Afterwards the exponential form of the functions z ( ) D z ()ei and
® ( ) D w ® ( )eiw is postulated which leads to the new form of the governing
equations

P z C 1 i e 4i ˛ e 2i z z 3 C 1 c1 z C z e 2i 3 i ˛ z2r z z e 2i
8
2 2 2
2
34 ˛ zr e 3i z e 2i z 38 w2 e i.1C2w/ ' C 'e
2 4i w

C 14 i e 2i w2 ' ' e 2i z z C e i
2
C 18 i e 2i.1Cw/ w2 e 2i z z e 4iw '2 C ' D f1 e i cos .p1 / ;
(16)

w P ' C 12 i e i w P ' z e 2i z 12 c2 w ' e 2iw ' C 14 e i w .w C 2/ ' z
14 e i .12w/ w .w 2/ ' z 14 e i w .w 2/ ' z C 14 e i .1C2w/ w .wC2/ '
3 z
48 1
i w2 e 4iw ' e 2i w ' Df2 e iw cos .p2 / ;
(17)
with the initial conditions
z .0/ D 0; ' .0/ D 0; z .0/ D 0; ' .0/ D0 (18)
3 Asymptotic Solution
The problems (16)–(18) can be efficiently solved by the asymptotic multiple scale
method. The assumptions of smallness of the parameters are proposed in the form
c1 D cQ1 "2 ; c2 D cQ2 "2 ; zr D zQr "; f1 D fQ1 "3 f2 D fQ2 "3 ; (19)
where " is the so-called small parameter.

Adopting three timescales in the analysis the solutions are searched in the
following form of series with respect to the small parameter:
X
kD3

z . I "/ D "k z k .0 ; 1 ; 2 / C O "4 ;
kD1
(20)
X
kD3

' . I "/ D " 'k .0 ; 1 ; 2 / C O "4 ;
k
kD1
and the differential operator has the form
d @ @ @
D C" C "2 C :::: (21)
d @0 @1 @2
3.1 Motion Near Resonance
Let us focus attention on the case of main resonance p2 w and p1 1. In order to

deal this case the following substitutions have been done
p1 D 1 C 1 and p2 D w C 2 ; (22)
where 1 D Q 1 "2 and 2 D Q 2 "2 are detuning parameters.

Introducing now (19), (20) and (22) into (16) and (17) and replacing the ordinary
derivatives by the differential operator (21) we obtain two equations in which the
small parameter " appears. These equations should be satisfied for any value of the
small parameter, so after sorting them with respect to the powers of " we get
(i) the equations of order "1
@z1
D 0; (23)
@0
@'1
D 0; (24)
@0
(ii) the equations of order "2
@z2 @z1 1 2 i0 3 2

C we '1 '1 w2 e i0 .1C2w/ e 4i0 w '1
2
'1 D 0; (25)
@0 @1 4 8
@ @ @ @
w @'20 12 i e i0 w @'10 z1 C 12 i e i0 w @'10 z1 C w @'11 C 14 e i0 w'1 z1 .w C 2/
14 e i0 .12w/ w2 '1 z1 14 e i0 w'1 z1 .w 2/ C 14 e i0 .1C2w/ w2 '1 z1 (26)
C 12 e i0 .12w/ w '1 z1 C 12 e i0 .1C2w/ w '1 z1 D 0;
(iii) the equations of order "3
2
@z3
@0
C @z2
@1
C 32 i ˛Q z2r z1 z1 e 2i0 34 ˛Q zr e 3i0 z1 z1 e 2i0
@z1
@2
3
Q 4i0 z2r z1 e 2i0 z1 C cQ21 z1 e 2i0 C z1 12 fQ1 e i Q1 2
C 18 i ˛e
ˇ ˇ2
14 w2 e i0 '2 '1 C '2 '1 14 w2 ˇ'1 ˇ z1 z1 e 2i0

34 w2 '1 '2 e i0 .2w1/ C '2 '1 e i0 .2wC1/ 12 fQ1 e i Q1 0 i Q1 2
2 2

C 18 i w2 '12
z1 e 2i0 w C '1 z1 e C2i0 w C '12
z1 e 2i0 .w1/ C '1 z1 e 2i0 .wC1/ D 0;
(27)
h i
@'3 @'2 @ @
@0
12 i we i0 z1 e 2i0 z1 @0
@'11 z2 e 2i0 z2 @'10

C w @'21 C w @'12 C cQ22 w '1 C '1 e 2iw0 12 fQ2 e i Q2 2 12 fQ2 e i.2w0 CQ2 2 /
@ @

C 14 e i0 w .2 C w/ z2 '1 C z1 '2 14 e i0 w .w 2/ z2 '1 C z1 '2
3
48 1
i w2 e 4iw0 '1 e 2iw0 '1 14 w .w 2/ e i0 .12w/ z2 '1

C 14 w .w C 2/ e i0 .1C2w/ z2 '1 C 12 we i0 .12w/ z1 '2 1 w2

C 12 we i0 .1C2w/ z1 '2 1 C w2 D 0:
(28)
The requirement of zeroing of secular terms in (23)–(26) causes the functions

z1 .2 / ; z1 .2 / ; '1 .2 / ; '1 .2 / to depend only on the slowest timescale 2 .
Solutions of the second-order equations (25) and (26)
3i w2 i0 .2w1/ 2 2
1
z2 D G1 .1 ; 2 / C e '1 C e i0 .2wC1/ '1 C i w2 e i0 '1 '1 ;
8 .1 2w/ 4
(29)
'2 D G2 .1 ; 2 / C 14 i e i0 .2 C w/ z1 '1 C 14 i e i0 .w 2/ z1 '1
i .w 2/ e i0 .12w/ z1 '1 i .w C 2/ e i0 .1C2w/ z1 '1 (30)

C ;
8w 4 8w C 4
are then introduced into equations of the third order (27) and (28). According to the
initial conditions (18), G1 D 0 and G2 D 0.
Assuming that the system vibrates far from the internal resonance 2w 1 D 0,
the requirement that the solutions should be limited in time leads to the equations

@z1 cQ1 3 3 3i w2 w2 1 ˇ ˇ
C z1 i ˛Q zQr z1 p i ˛Q jz1 j z1 C
2 ˇ'1 ˇ z1 D 1 fQ1 e i2 Q1 ;
@2 2 2 8 4 16w2 2
(31)
@ cQ2 i w2 .w2 C2/

w @'12 C 2
14 i w2 jz1 j '1 C 16w2 4 jz1 j '1
w'1
(32)
i w2 .8w4 7w2 1/ ˇˇ ˇ
C 64w2 16
'1 ˇ '1 D 12 fQ2 e i2 Q2 :
Now real representation of the functions z1 and '1 of the following form
z1 .2 / D aQ 1 .2 / e iı1 .2 / ; '1 .2 / D aQ 2 .2 / e iı2 .2 / ; aQ i D ai " for i D 1; 2
(33)
is introduced to the above secular terms (31) and (32).

Then we go back to the original denotations according to (19) and take advantage
of the definition (21). Comparison of the real and imaginary parts of both sides of
(31) and (32) leads to four modulation equations with respect to amplitudes a1 , a2
and modified phases
1 ,
2 :
da1 1 1
D c1 a1 C f1 cos
1 ; (34)
d 2 2

d
1 3 2 3 3w2 w2 1 1
a1 D zr ˛ a1 C 1 a1 ˛ a1 C
3
a1 a22 f1 sin
1 ; (35)
d 2 8 4 16w2 2
da2 1 1
D c2 a2 C f2 cos
2 ; (36)
d 2 2w

d
2 3w w2 1 w 8w4 7w2 1 3 1
a2 D 2 a2 C a2 a1 C
2
a2 f2 sin
2 ; (37)
d 4 16w2 64w2 16 2w
where modified phases

1 ,
2 are defined as follows:
ı1 .2 / D 2 1
1 .2 / ; ı2 .2 / D 2 2
2 .2 / : (38)
The above definitions cause the systems (34)–(37) to become an autonomous one.
It describes the dynamics of the non-linear spring pendulum near simultaneously
occurring main resonances.
4 Examples
The LPT concept allows to describe the intensive energy exchange between the
degrees of freedom and external sources. The system examined in the paper is
especially sensitive to changes of the value of the parameter ˛ responsible for non-
linear characteristics of the spring. One can observe a critical value ˛ D ˛ LPT for
which the character of the vibrations dramatically changes. The value of ˛ LPT
Fig. 2 Limiting phase trajectories for longitudinal and swing vibrations; ˛ D 0.3 < ˛ LPT
Fig. 3 Amplitudes modulations; ˛ D 0.3 < ˛ LPT
Fig. 4 Time histories obtained numerically; ˛ D 0.3 < ˛ LPT
depends on all the parameters of the system. In the case of certain one degree-of-
freedom (DOF) systems ˛ LPT can be obtained analytically [2, 3]. When the number
of DOF is higher than one and couplings appear in the equations of modulation,
˛ LPT can be obtained approximately.
The results of calculations for the chosen values of parameters 1 D 0.01,
2 D 0.01, f1 D 0.0008, f2 D 0.00008 , c1 D 0, c2 D 0, w D 0.21 are presented below.
For these parameters ˛ LPT 0.654. Figures 2, 3 and 4 show some graphs concerning
the case when ˛ D 0.3 < ˛ LPT .
Fig. 5 Limiting phase trajectories for spring and swing vibrations; ˛ D 0.7 > ˛ LPT
Fig. 6 Amplitudes modulations; ˛ D 0.7 > ˛ LPT
Fig. 7 Time histories obtained numerically; ˛ D 0.7 > ˛ LPT
In Figs. 2, 3 and 4 the intensive energy exchange between the system and its
surrounding can be observed. The amplitudes are relatively small and the vibrations
are quasi-linear. The time histories presented in Fig. 4 have been received by
numerical solution to the problem (6)–(8). The amplitude modulations and time
histories (compare Figs. 3 and 4) are highly compatible.
For ˛ > ˛ LPT some non-linear effects occur. In Figs. 5, 6 and 7 amplitudes are
much larger than in the case when ˛ < ˛ LPT and the vibrations become strongly
non-linear.
Fig. 8 Limiting phase trajectories for spring and swing vibrations; ˛ D 2.0 > > ˛ LPT
Fig. 9 Amplitudes modulations; ˛ D 2.0 > > ˛ LPT
Fig. 10 Amplitudes modulations; ˛ D 2.0 > > ˛ LPT
In the case ˛ > > ˛ LPT modulations of amplitudes of both coordinates become
again more regular, as is shown in Figs. 9 and 10. However, their shape indicates
the non-linear effects. The longitudinal vibration tends to sawtooth form with
the increase of non-linearity parameter. A synchronization between the amplitude
modulations of both general coordinates in the slow timescale is observable (Fig. 9).
The oscillations of the system tend to steady state what is seen not only in the time
history but also in the phase – amplitude plane (Fig. 8).
5 Conclusions
Analytical study of the non-linear spring pendulum in planar motion has been
carried out. After transformation of the governing equations of motion to the
complex representation, the asymptotic analysis with the help of multiple timescale
(MTS) method has been applied. That approach leads to obtain a set of differential
equations of simpler form than the original ones. It is worth to note that, thanks to the
application of the MTS variant with three timescales, the non-linear terms as well
as the most important coupled terms between the generalized coordinates have been
preserved in the equations of the simplified mathematical model. The performed
investigations have been focused on the nonsteady vibrations of the forced system
near the simultaneously occurring external resonances. The modulation equations
concerning this case have been derived from the equations of first order as well as
from the requirement of vanishing the secular terms of the equations of higher order.
The solutions obtained analytically from the equations of modulation of amplitudes
and phases have been verified by comparing them with the solutions which are
received numerically from the original equations of motion. Their high accuracy
has been confirmed in all performed numerical simulations.
The main advantage of the asymptotic solutions consists in achieving qualitative
information about the dynamics of the considered system.
Analysis of the curves which represent the dynamical behaviour of the system in
the plane phase-amplitude gives evidence of very interesting features of dynamics of
the system. The shape of these curves depends strongly on the values of parameter ˛,
which is connected with the spring non-linearity. The most intensive energy
transfer between the system and its surroundings is governed by the so-called LPT.
Important non-linear dynamical transition-type phenomena are detected, monitored
and discussed, amongst others. For ˛ > ˛ LPT amplitudes are much greater than for
˛ < ˛ LPT . Moreover, it has been shown that the shape of the amplitude modulation
curves changes with the value of ˛.
Acknowledgments This paper was financially supported by the National Science Centre of
Poland under the grant MAESTRO 2, No. 2012/04/A/ST8/00738, for years 2013–2016.
References
1. Awrejcewicz, J., Krysko, V.A.: Introduction to Asymptotic Methods. Chapman and Hall, Boca
Raton (2006)
2. Manevitch, E.L., Manevitch, L.I.: Limiting phase trajectories (LPT) in 1 dof asymmetric system
with damping and 1:1 resonance. In: DSTA 10th Conference, Łódź, pp. 559–568, 2009
3. Manevitch, L.I., Musienko, A.I.: Limiting phase trajectories and energy exchange between
anharmonic oscillator and external force. Nonlinear Dyn. 58, 633–642 (2009)
4. Sado, D.: Energy Transfer in the Nonlinear Coupled Systems of Two-Degree of Freedom (in
Polish). OWPW, Warszawa (1997)
5. Sado, D., Gajos, K.: Analysis of vibrations of three-degree-of-freedom dynamical system with
double pendulum. J. Theor. Appl. Mech. 46(1), 141–156 (2008)
6. Shivamoggi, B.K.: Perturbation Methods for Differential Equations. Birkhauser, Boston (2002)
7. Starosta, R., Sypniewska-Kamińska, G., Awrejcewicz, J.: Resonances in kinematically driven
nonlinear spring pendulum. In: DSTA 11th Conference, pp. 103–108, 2011
8. Starosvetsky, Y., Gendelman, O.V.: Dynamics of a strongly nonlinear vibration absorber coupled
to a harmonically excited two-degree-of-freedom system. J. Sound Vib. 312, 234–256 (2008)
Vibration Surveillance System with Variable
Stiffness Holder for Milling Flexible Details
Krzysztof J. Kaliński, Marek Chodnicki, Michał R. Mazur,

and Marek A. Galewski
Abstract Efficient milling of the flexible details (i.e. rotor blades, thin-walled
elements) using slender ball-end tools is a difficult task due to possibility of
vibration occurrence. Because of the existence of certain conditions (small depths
of cutting, regeneration phenomena), cutting process may lose stability and self-
excited chatter vibration may appear. Frequency of the chatter vibration is close
to dominant natural frequency of the workpiece or the tool. One of the methods
of chatter vibration avoidance is matching the spindle speed to the optimal phase
shift between subsequent cutting edge passes (i.e. the Liao–Young condition). In
previous works the authors successfully implemented the idea of optimal speeds
map where optimal speed was calculated for every point of the machined surface
based on the dominant natural frequencies for local areas. During milling, spindle
speed was set according to the map. However, changing spindle speed during tool
pass may reduce surface quality in speed change point and is difficult to perform it in
some milling centres. The article presents the idea of a new workpiece holder with
adjustable stiffness. Milling process will be performed with constant spindle and
feed speed. In order to avoid vibration, stiffness of the specially designed workpiece
holder will be modified off-line. Stiffness changes modify natural frequencies of the
workpiece and thus, it is possible to modify dynamic properties of the workpiece
in such a way that arbitrary chosen, constant spindle speed will be optimal, due to
the Liao–Young condition performance. This will need calculation of the optimal
stiffness map (referred to different spindle speeds), which will be performed before
milling based on the workpiece’s modal identification results and Finite Element
Model simulations.
1 Introduction
Milling flexible details (i.e. rotor blades, thin-walled elements) using slender ball-
end tools is a frequent task for contemporary machining centres. It is also usually
demanded that the milling treatment should be “final” which means that no
K.J. Kaliński () • M. Chodnicki • M.R. Mazur • M.A. Galewski

Gdansk University of Technology, G. Narutowicza 11/12, 80-233 Gdansk, Poland
e-mail: kkalinsk@o2.pl; marchodn@pg.gda.pl; mazur.m.r@gmail.com; margalew@pg.gda.pl

176 K.J. Kaliński et al.
additional finishing operations are required for the surface quality being sufficiently
high. On the other hand, milling flexible details is also a challenging problem
because such objects can be prone to vibration especially when the workpiece is
difficult to support or cannot be supported during milling processes. In such a
case, tool–workpiece relative vibration plays a principal role because, in certain
conditions, it may lead to a loss of stability and cause a generation of self-excited
chatter vibration [8]. One of the methods of the chatter reduction depends on
the selection of the spindle speed matched to the optimal phase shift between
subsequent passes of the tool cutting edges [7]. This optimal spindle speed can be
calculated using the generalised Liao–Young condition [1]:
zn˛ f˛
D ; m D 0; 1; 2; : : : ; (1)
60 0:25 C m
where:
f˛ is the natural frequency no. ˛ of the workpiece (Hz),
n˛ is the requested spindle speed (rev/min),
z is the number of cutting edges of the milling tool.
This means that the only unknown in Eq. 1 is the dominant natural frequency f˛ .
It may be identified, for example, by the modal test performance. As the result, a
set of optimal spindle speeds is obtained. From this set, one speed is selected for
performing milling operation. The problem is that the speed, which is optimal from
the point of view of Liao–Young condition, may not be optimal from the point of
view of the other requirements, i.e. cutting process efficiency. As an example, for
f˛ D 83.3 Hz and z D 2, the set of optimal spindle speeds is n˛ 2 <10,000, 2,000,
1,111, 769, : : : > (rev/min). One can see that there is a huge gap between the first
and the second speed. For example, in case when milling centre can achieve speeds
only up to 8,000 rev/min, the highest optimal (in terms of vibration reduction) speed,
which can be selected, is 2,000 rev/min. This means that in this situation it is not
possible to make the most of milling centre potential.
Additionally, in case of complex-shaped workpieces, there may be different
dominant natural frequencies for different workpiece zones. This problem can be
solved by creating a map of optimal spindle speeds. In order to create such a
map, a Finite Element Model of a workpiece is created and tuned according to
the experimental modal tests. Then, the model is used for calculation of dominant
natural frequencies for the selected workpiece points. For these points optimal
spindle speeds are calculated. The whole procedure is described in details in [5, 6]
and successfully implemented for milling complex-shaped flexible workpieces.
However, changing spindle speed during tool pass may reduce surface quality in
speed change point and is difficult to perform in some milling centres. This happens
especially when milling centre controller does not allow speed changes without
stopping the tool or workpiece feed. The problem can be avoided by milling a whole
tool pass with only one spindle speed. However, this also means that in some zones
this speed will not be optimal.
Vibration Surveillance System with Variable Stiffness Holder for Milling. . . 177
Fig. 1 Workpiece holder with adjustable stiffness
2 Variable Stiffness Workpiece Holder
According to Eq. (1), optimal spindle speed depends on dominant natural frequency
of the workpiece. In turn, this frequency depends on the workpiece’s dynamic
properties, which may be considered as time-invariant. However, the workpiece is
mounted in a holder. Usually the holder is considered as of a stiff support. In the
presented case, a holder is one degree of freedom (DOF) structure with adjustable
stiffness (Fig. 1). Thanks to this, it is possible to modify dynamic properties of the
whole system (consisting of the holder and the workpiece) and to adjust its natural
frequency. This foresees a possibility of tuning the natural frequency in such a way
that any arbitrary spindle speed will be optimal, in accordance with Eq. (1). For
example, choosing n˛ D 10,000 rev/min implies that natural frequency should be
one of set f˛ – <83, 416, 750, : : : > (Hz). This means that the stiffness of the holder
must be suitably adjusted.
3 Holder, Workpiece and Cutting Process Models
Dynamics of cutting process can be described using a proportional model [2, 4].
In this model cutting force components depend proportionally on cutting layer
thickness and also depend on variable in time depth of cutting. Resultant cutting
force lies in orthogonal plane. Depending on the direction of action three force
components can be separated, i.e. the one acting along nominal cutting velocity
(Eq. 2), thrust force acting along change in cutting layer thickness (Eq. 3) and the
force acting along tool axis which is equal to 0 (Eq. 4):
(
kd l ap .t / ŒhDl .t / hl .t / C hl .t l / ; hDl .t / hl .t / C hl .t l / > 0;
Fyl1 .t / D
0 ; hDl .t / hl .t / C hl .t l / 0;
(2)
(
l kd l ap .t/ ŒhDl .t/ hl .t/ C hl .t l / ; hDl .t/ hl .t/ C hl .t l / > 0;
Fyl2 .t/ D
0 ; hDl .t/ hl .t/ C hl .t l / 0;
(3)
Fyl3 .t / D 0 (4)
where:
kdl is the average dynamic specific cutting pressure,
ap (t) is the depth of cutting,
hDl (t) is the desired cutting layer thickness,
4hl (.) is the dynamic change in cutting layer thickness,
l is the cutting force ratio,
l is the time-delay.
In order to idealise dynamics of non-stationary milling process of a flexibly
supported rectangular plate, a hybrid approach is adopted [2, 3]. As a result of the
milling process modelling, the hybrid system is obtained which assembles three
subsystems (Fig. 2):
– Modal subsystem, i.e. a stationary model of one-side-supported flexible plate,
which displaces itself with feed speed vf
– Structural subsystem, i.e. non-stationary discrete model of ball-end mill and
cutting process
– Connective subsystem, i.e. conventional contact point S between the tool and the
workpiece
Matrix dynamic equation of non-stationary model of the milling process in
hybrid co-ordinates has a form [2] (Eq. 5)
2 3
X il Xil
" # " # 6KC TTl .t /DP l .t /Tl .t / TTl .t /DP l .t /Wml .t / 7
M0 L 0 6 7
ŸR C ŸP C 6
6
lD1 lD1 7Ÿ D
7
0 I 0 2Z˝ 4 X
i l X i l
5
T T
Wml .t /DP l .t /Tl .t / ˝ C
2
Wml .t /DP l .t /Wml .t /
lD1 lD1
2 3 2 3
X
il X
il
6 TTl .t /F0l .t / 7 6 TTl .t /DOl .t /w .t l / 7
6 7 6 7
D6
6
lD1 7C6
7 6
lD1 7 ;
7
4 X
il
5 4 Xil
5
WTml .t /F0l .t / T
Wml .t /DOl .t /w .t l /
lD1 lD1
(5)
Fig. 2 Hybrid system of the milling process
where:

q
ŸD is the vector of hybrid co-ordinates of the hybrid system,
a
M, L, K are the matrices of inertia, damping and stiffness of the decoupled structural
subsystem,
a is the vector of modal co-ordinates of the modal subsystem,
q is the vector of generalised co-ordinates of the structural subsystem,
Tl (t) is the matrix of transformation of displacements from co-ordinate system of
the cutting tool, to co-ordinate system yl1 , yl2 , yl3 of CE no. l,
Wml (t) is the matrix of constraints between displacements in modal co-ordinates a
of the modal subsystem and displacements in co-ordinate system yl1 , yl2 , yl3 of
CE no. l,
DPl (t), DOl (t) are the matrices of proportional and delayed feedback of CE no. l,
Z is the matrix of dimensionless damping coefficients of the modal subsystem,
is the matrix of angular natural frequencies of the modal subsystem,
F0l (t) is the vector of the desired forces of CE no. l,
wl (t l ) is the vector of deflections of CE no. l for time-instant t l ,
Fig. 3 Scheme of the simplified workpiece model, the considered DOF are denoted with grey
arrows
il is the number of “active” coupling elements.

In order to identify modal model of the machined plate, the mass normalised
matrix of normal modes ‰ corresponding to angular natural frequencies of the
modal subsystem must be determined. This can be done using a simplified model
based on the Euler–Bernoulli bar finite element [2] and considering only three DOF
(Fig. 3). The appropriate mass Me and stiffness Ke matrices have forms
2 13A 6Ie 9Ae 6Ie 13Ae le Ie 3
e
C 2
6 35 5le2 70 5le 420 10le 7 2m 0 03
6 7 l
6 13Ae 6Ie 11Ae le Ie 7 4
Me D e le 6 C 2 C 7C 0 0 05; (6)
6 35 5le 210 10le 7
4 5 0 00
Ae le2 2Ie
sy m: C
105 15
2 3
12Ie 12Ie 6Ie
2 3
6 l2 le2 le 7 ks 0 0
E6 e 7
Ke D 6 12Ie 6Ie 7 C 4 0 0 0 5 (7)
le 6
4 le2 le 5
7
0 00
sy m: 4Ie
where:
e is the density of the workpiece material,
E is the Young modulus of the workpiece material,
le is the length of the plate workpiece,
Ae is the cross-section area of the plate workpiece,
Ie is the cross-section area moment of inertia of the plate workpiece,
ks is the adjustable stiffness of the holder’s support,
ml is the lumped mass of the holder’s moving part.
4 Simulations
At first, simulations are performed in order to confirm that variable stiffness of

the holder modifies normal modes of the workpiece. For the selected settings of
the holder stiffness adjusting mechanism, i.e. position of the micrometre screw
(mm), the corresponding stiffness is calculated and then the related first nat-
ural frequencies and corresponding normal modes of the system described by
matrices (Eq. 6) and (Eq. 7) are determined. Additionally, using the generalised
Liao–Young condition (Eq. 1), relevant optimal spindle speeds are calculated.
Computation is performed for the following data: workpiece material—aluminium
alloy EN AW-6101A, e D 2.7 103 kg/m3 , E D 7 1010 N/m2 , le D 0.160 m,
Ae D 2.5 104 m2 , Ie D 5.2083 1010 m4 , ml D 2 kg. The results are shown in
Table 1.
It is worthy of note that not all of the holder stiffness adjustments assure
successful result of the workpiece vibration surveillance. The optimal spindle speed
is really obtained, when the mass normalised normal mode of the workpiece:
– Has a similar form to one of the stiff support cases, i.e. relevant components of
the two compared normal modes have the same signs
– Does not concern the rigid body movement, i.e. first and second component of
the normal mode significantly differ with each other
Table 1 Holder stiffness influence on first normal mode of the workpiece and corresponding
optimal spindle speed
Micrometre
screw Holder Natural Optimal Result of
position stiffness frequency Normal mode (mass spindle speed surveil-
(mm) ks (N/m) f˛ (Hz) normalised) ‰ 1 n (rev/min) lance
Stiff support 161:4 0:0000 6:1439 52:8957 – –
20 2.3100 106 151:0 0:4002 4:8077 37:7958 18,120 Negative
21 1.9955 106 145:4 0:5070 3:9404 29:3819 17,450 Negative
22 1.7355 106 138:5 0:5787 3:1180 21:6776 16,623 Successful
23 1.5189 106 131:2 0:6189 2:5081 16:0883 15,743 Successful
24 1.3368 106 124:0 0:6412 2:0884 12:2956 14,880 Successful
25 1.1827 106 117:2 0:6544 1:7975 9:6923 14,064 Successful
26 1.0514 106 110:9 0:6627 1:5896 7:8452 13,305 Successful
27 9.3888 105 105:0 0:6682 1:4361 6:4888 12,603 Successful
28 8.4184 105 99:6 0:6722 1:3194 5:4617 11,956 Successful
29 7.5772 105 94:7 0:6750 1:2284 4:6637 11,358 Successful
30 6.8444 105 90:1 0:6772 1:1559 4:0299 10,812 Successful
31 6.2032 105 85:8 0:6789 1:0970 3:5173 10,298 Successful
32 5.6396 105 81:9 0:6802 1:0486 3:0962 9,826 Negative
35 4.3102 105 72:0 0:6829 0:9451 2:1995 8,604 Negative
40 2.8875 105 59:0 0:6853 0:8488 1:3685 7,056 Negative
Fig. 4 The selected path of the tool along the workpiece length
Fig. 5 (a) Vibration displacements, (b) corresponding amplitude spectrum for n D 10,000 rev/min
and ks D 5.845 105 N/m, optimal case, f˛ D 83 Hz
The purpose of the next step of simulations is to show that for a chosen
stiffness value the calculated spindle speed is really optimal and, on the other
hand, for a given spindle speed, this stiffness is optimal as well. Also for this
couple, i.e. optimal stiffness—optimal spindle speed, the workpiece vibration level
should be the lowest. For every simulation the root mean square (RMS) of the free
(unsupported) workpiece end displacement is observed as well as the maximum
value of vibration spectrum. The process of milling 60 mm long groove on the
workpiece surface (Fig. 4) is simulated using the hybrid model of the milling process
(Eq. 5). Some selected cases of simulation (Figs. 5, 6 and 7) as well as the summary
of the results for various cases are presented (Fig. 8).
The results of simulation (Fig. 8) confirm expectations of the holder stiffness
influence on first normal mode of the workpiece and corresponding optimal spindle
speed for a wide range of spindle speeds, i.e. between 10,000 and 14,064 rev/min.
The only one stiffness setting is optimal and gives both the lowest value of RMS
of displacement as well as the lowest maximum amplitude in spectrum. In such
cases the generalised Liao–Young condition is really fulfilled (Eq. 1). Additionally,
for some nonoptimal stiffness-spindle speed pairs, vibrations may be very high. For
and ks D 6.844 105 N/m, nonoptimal case, higher stiffness, f˛ D 72 Hz
and ks D 4.310 105 N/m, nonoptimal case, lower stiffness, f˛ D 90 Hz
a b
Max. spectrum aplitude [mm]
2,50E-04
2,00E-04
18118
1,50E-04
14064
1,00E-04 10807
10000
5,00E-05 8603
7052 Spindle speed
0,00E+00 [rev/min]
Stiffness
Fig. 8 (a) Summary of vibration displacements, (b) maximum value of corresponding amplitude
spectrum, for chosen simulated pairs of n and ks settings
example, ks D 1.02 106 N/m and n D 7,052 rev/min, which results in resonance
because workpiece’s natural frequency is 117 Hz, and frequency of excitation from
tool rotations is also 117 Hz.
5 Conclusions
Simulations showed that modifying the workpiece dynamic properties is possible

with the use of the proposed new workpiece holder. Simulations for different
pairs of holder stiffness and spindle speed show that only in case of a proper
optimal combination of these two parameters, vibrations are the lowest. These
results mean that the proposed variable stiffness holder has a great potential. In
a standard perfectly stiff holder, choosing the proper spindle speed on a basis
of the Liao–Young condition was a method of vibration reduction. However, the
determined optimal spindle speed might not be optimal from other points of view
(i.e. technological ones). The proposed variable stiffness holder allows adjusting
workpiece’s dynamic properties in such a way that an arbitrarily given spindle
speed should accomplish the generalised Liao–Young condition, in a wide range of
operational spindle speeds of the milling machine. Thus the vibration surveillance
efficiency is really evidenced.
References
1. Kalinski, K.J.: Modelling and investigation of vibration surveillance during ball end milling of
curved flexible details. In: Uhl, T. (ed.) Chosen Problems of the Modal Analysis of Mechanical
Structures, pp. 91–99. The Publication of AGH, Cracow (2006)
2. Kaliński, K.J.: A Surveillance of Dynamic Processes in Mechanical Systems (in Polish). Gdańsk
University of Technology Publishers, Gdańsk (2012)
3. Kaliński, K.J., Chodnicki, M.: Vibration surveillance during ball end milling of curved flexible
details on a basis of assessment of assurance of the model of real object. In: Cieśliński, J.T.,
Barylski, A. (eds.) Developments in mechanical engineering, vol. 3, pp. 117–124. Gdańsk
University of Technology Publishers, Gdańsk (2009)
4. Kaliński, K.J., Galewski, M.A.: Chatter vibration surveillance by the optimal-linear spindle
speed control. Mech. Syst Signal Process. 25(1), 383–399 (2011)
5. Kalinski, K.J., Mazur, M., Galewski, M.A.: The optimal spindle speed map for reduction of
chatter vibration during milling of bow thruster blade. Solid State Phenom. 198, 686–691 (2013)
6. Kalinski, K.J., Mazur, M., Galewski, M.: High speed milling vibration surveillance with the use
of the map of optimal spindle speeds. In: Proceedings of the 8th International Conference on
High Speed Machining, Metz, France, 2010
7. Liao, Y.S., Young, Y.C.: A new on-line spindle speed regulation strategy for chatter control. Int.
J. Mach. Tools Manuf. 36, 651–660 (1996)
8. Tomkow, J.: Vibrostability of Machine Tools (in Polish). The Scientific and Technical Publica-
tions, Warsaw (1997)
Diversities in the Inverse Dynamics Problem
for Underactuated Mechanical Systems Subject
to Servo-constraints
Wojciech Blajer
Abstract Underactuated mechanical systems are featured with less control inputs
than degrees of freedom. Their performance goal may then be realization of
specified in time outputs whose number coincides the number of inputs. A solution
to the inverse simulation problem (servo-constraint problem), that is, determination
of an input control strategy that forces the underactuated system to complete the
partly specified motion, is a challenging task. Since systems may be “underactuated”
in several ways and the servo-constraint realization may range from orthogonal to
tangential, diverse formulations and analysis methods of the servo-constraint prob-
lem arise. The diversity is discussed with reference to some simple case studies. The
governing equations are handled in two ways. A direct formulation in configuration
coordinates is first motivated and is then compared to a setting in which the actuated
coordinates are replaced with the outputs. The governing equations arise either as
ODEs (ordinary differential equations) or DAEs (differential-algebraic equations).
Some computational issues related to the ODE and DAE formulations are discussed,
and simulation results for the sample case studies are reported.
1 Introduction
Underactuated mechanical systems are featured with less control inputs than degrees
of freedom, represented, e.g., by cranes [1, 16], aircrafts [8, 25], manipulators
with passive and/or elastic joints [12, 22], and flexible robots [3]; see also [13,
24] for other representatives. A possible performance goal of such systems is
realization of specified in time outputs whose number is equal to the number of
inputs. The motion specifications are treated as servo-constraints on the system
[6, 11, 15], with the control forces referred to as reactions of the constraints.
An inverse dynamics problem (servo-constraint problem), that is, determination
of an input control strategy that forces an underactuated system to complete the
W. Blajer ()
Faculty of Mechanical Engineering, University of Technology and Humanities in Radom, ul.
Krasickiego 54, 26-600 Radom, Poland
e-mail: w.blajer@uthrad.pl

186 W. Blajer
partly specified motion, may in general be a challenging task. As systems may be

“underactuated” in several ways [13, 24] and the servo-constraint realization may
range from orthogonal to tangential [5], diverse formulations and analysis methods
of the servo-constraint problem arise. The solution may also be purely algebraic or
may include some internal (unspecified) dynamics [21, 23].
The diversities in the servo-constraint problem for underactuated systems are
discussed in this paper with reference to the way the servo-constraints are realized
and are illustrated with some simple case studies. A standard formulation in
configuration coordinates is first motivated and then compared to a setting in which
the actuated coordinates are replaced with the outputs. Depending on the way of
servo-constraint realization, the arising governing equations are formulated either as
ODEs (ordinary differential equations) or DAEs (differential-algebraic equations).
Solutions of the servo-constraint problem for differentially flat systems [14, 18]
(with no internal dynamics) and non-flat systems (with internal dynamics) are also
discussed. The computational issues related to the ODE and DAE formulations
are finally discussed, and some simulation results for the sample case studies are
reported.
We consider an f -degree-of-freedom underactuated mechanical system, described

by f configuration coordinates q D [q1 qf ]T and actuated by m inputs
u D [u1 um ]T , m < f. The generic dynamic equations of the system are
M .q/ qR C k .q; q/
P D g .q; q/
P C B .q/ u (1)
where M is the f f generalized mass matrix, k is the f -vector of Coriolis,

gyroscopic, and centrifugal dynamic terms, g is the f -vector of applied forces, and
B is the f m maximum rank matrix of distribution of the m control inputs u on the
f directions of q.
A solution to the inverse dynamics problem for underactuated systems is possible
only if the motion is specified by as many outputs as inputs [5, 6, 19, 21–23]. The
m outputs y D [y1 ym ]T can be expressed in terms of q as y D ˆ(q), where
ˆ D [˚ 1 ˚ m ]T are appropriately differentiable functions. Given specified in
time outputs, y D yQ .t /, m servo-constraints [6, 11, 15] on the underactuated system
arise, whose equations, at the position, velocity, and acceleration levels, are
® .q; t / D ˆ .q/ yQ .t / D 0 (2)
P q; t / D H .q/ qP yPQ .t / D 0
” .q; (3)
Diversities in the Inverse Dynamics Problem for Underactuated Mechanical. . . 187
R q;
˜ .q; P yRQ .t / D 0
P q; t / D H .q/ qR C h .q; q/ (4)
where the m f Jacobian matrix H D @ˆ/@q is of maximum rank, and h D H P qP is

the m-vector of accelerations induced by the output relationship y D ˆ(q).
As noticed in [5, 6], the servo-constraints introduced in Eq. (5) are mathe-
matically equivalent to m passive (holonomic and rheonomic) constraints on the
system, ®(q, t) D 0. By analogy to the constrained system dynamics, the generalized
actuating force gu D B u can be viewed as a generalized reaction force of the servo-
constraints. As opposed to the reactions of (ideal) passive constraints, in the classical
sense of smooth hard surfaces, joints, and supports, whose reactions are orthogonal
to constraint manifold [4], the control reaction gu may be arbitrary oriented with
respect to the servo-constraint manifold [5, 6], and in the extreme, it may be tangent
to the manifold. The diversity in realization of servo-constraints can be deduced
from the input–output relationship, obtained from Eq. (4) after using qR from Eq. (1),
i.e.,
yRQ .t / D H M1 .g k/ C H M1 B u C h (5)
An important consequence of arbitrary orientation of gu with respect to the servo-

constraint manifold is that some (or even all) of the system outputs y might not be
directly regulated by the inputs u. A measure of the control singularity is possible
deficiency in rank of the m m matrix Y D H M 1 B, which expresses the projection
of the control reactions onto the directions of the constraint gradients (represented
as rows of H). It can also be viewed, in the system configuration space Q, as the
intersection, the orthogonal subspace H (spanned by the constraint gradients), and
the actuated subspace B (spanned by the vectors represented as columns of B), Y D
H \ B [4]. According to the rank of Y (dimension of Y) three types of servo-
constraint realization can be distinguished [5, 6]:
8
< pDm orthogonal
rank .Y/ D rank H M1 B D p 0 < p < m orthogonal-tangential (6)
:
pD0 tangential
The orthogonal realization (p D m) denotes that all of the system outputs y can
directly be regulated by the inputs u. This is usually a nonideal orthogonal (skew-
orthogonal) realization in which the control reactions, being explicitly represented
in H, project also in the tangent subspace G with respect to the servo-constraint
manifold, H [ G D Q and H \ G D 0; see Fig. 1a for an illustration. The
mixed orthogonal–tangential realization (0 < p < m) and pure tangential realization
(p D 0) show that at most p (none) outputs can be actuated directly (in the orthogonal
way), whereas the realization of the remaining m p (all) outputs is performed
without direct influence of the actuating forces. The geometry of the tangential
realization of servo-constraints is illustrated in Fig. 1b.
188 W. Blajer
a b
orthogonal orthogonal
subspace subspace
gu gu
servo-constraint servo-constraint
manifold manifold
tangent tangent
subspace subspace
Fig. 1 Geometrical illustration of (a) skew-orthogonal realization of servo-constraints, (b) tan-

gential realization of servo-constraints
y~ .
. .. .. . . .. y, y
y~ ystab= y~ + α( y~ y ) + β ( y~ y )
-1
u = Y ( ystab f y) SYSTEM
..
y~ .
y, y q ,q
.
inner loop
outer loop
Fig. 2 Structure of control for orthogonal realization of servo-constraints
3 Orthogonal Realization of Servo-Constraints
For the case of (skew-) orthogonal realization of servo-constraints, conditioned upon

rank(H M 1 B) D m D max, the relationship of Eq. (5) directly leads to the output–
input inverse dynamics relationship as follows

P D Y1 yRQ fy
u .t; q; q/ (7)
where Y(q) D H M 1 B and fy .q; q/ P D H M1 .g k/Ch. The above feedforward

controller, required for reproduction of yRQ .t /, can then be enhanced by a feedback
loop after replacing yRQ with yR stab in the form (PD controller)

yR stab D yRQ C ’ yPQ yP C “ .Qy y/ (8)
where ’ and “ are the diagonal matrices of gain values, and y and yP are the actual
outputs and their time derivatives. The structure of control with feedback is seen in
Fig. 2, which is the feedforward controller structure after setting ’ D 0 and “ D 0.
The governing equations are then 2f ODEs in q and v D qP
qP D v (9)
vP D M1 .g k/ C M1 B Y1 yR stab fy
h i
The controller u D Y1 yRQ C ’ yPQ yP C “ .Qy y/ fy is equivalent in
structure to the computed torque control scheme [17] for fully actuated systems
executing a fully specified motion. Its application to underactuated systems in
partly specified motion is not straightforward, however. Firstly, it is limited to the
orthogonal realization of servo-constraints. Then, assumed det(Y) ¤ 0, in addition to
the output–input inverse dynamic model represented (in the orthogonal subspace H)
by the algebraic formula of Eq. (7), an internal dynamics (in the tangent subspace
G) exists since the system motion is only partly specified by the outputs. For the
prevailing skew-orthogonal realization of servo-constraints, the evolution in time of
the internal dynamics and the inverse dynamics control are mutually dependent. Of
critical importance for the control strategy is therefore that the internal dynamics
remains bounded, which may require a proper design of the mechanical system [5]
and/or careful definition of the imposed motion specifications.
Case study 1: Internal dynamics stability problem. Let us consider a rotational arm
(Fig. 3a) consisting of two links, q D [
1
2 ]T , connected at the passive joint A with
a spring–damper combination and supported at the active joint O with the actuating
torque, u D [ ]. The lengths, masses, and central mass moment of inertia of the
links are li , mi and JCi , i D 1, 2, and the locations of the mass centers are s1 (OC1 )
and s2 (OC2 ). The spring and damper constants are c and d, respectively, and the
vanishing torque of the torsion spring is achieved for
1 D
2 . The arm moves in the
horizontal plane perpendicular to the direction of gravity. The dynamic equations of
the system, related to Eq. (1), are defined by
b y
a E
sP P c ~
C2 yE
A
c O
y
A
2 y~P E x
C1 d
d c
O 1 R R
x T = Rs2
A C2 E A C2 E
Fig. 3 Illustration of (a) the rotational arm with one active and one passive joints, (b) the arm
initial configuration, and (c) the action of torque on link 2 at the initial rest position
190 W. Blajer
" # " #
JC1 C m1 s12 C m2 l12 m2 s2 l1 cos .
2
1 / m2 s2 l1
P22 sin .
2
1 /
MD I kD I
m2 s2 l1 cos .
2
1 / JC 2 C m2 s22 m2 s2 l1
P12 sin .
2
1 /
2 3 " #
c .
2
1 / C d
P2
P1 1
gD4 5I BD
c .
2
1 / d
P2
P1 0
(10)
The output is assumed as the vertical position of a point P on link 2, defined by

sP (distance AP), y D [yP ]. The servo-constraint equation, defined in Eq. (2), and
the Jacobian matrix H D @ˆ/@q are then

® .q; t / D l1 sin
1 C sP sin
2 yQP .t / D 0I HD l1 cos
1 sP cos
2 (11)
The matrix Y D H M 1 B, used in Eqs. (5)–(7) and (9), is

" #
JC 2 C m2 s22 l1 cos
1 m2 s2 l1 sP cos .
2
1 / cos
2
Y D ŒY D (12)
det .M/
where, by assumption, det(M) > 0. As seen, depending on the inertial and geo-
metrical properties of the system, its current configuration (coordinates
1 and
2 ), and the placement of point P (distance sP ), the numerator of the above

expression, and as such Y, can be either negative or positive or may occasionally
diminish to zero. While the latter case makes orthogonal realization of the servo-
constraint impossible, let us concentrate on the cases Y < 0 and Y > 0. Specifically,
assume for a while that both the links are homogeneous and identical, i.e., si D li /2,
JCi D mi l2i /12 (i D 1, 2), and l1 D l2 D l, m1 D m2 D m, and that the point P coincides
with the end point E of link 2, P ! E. The value of Y is then

ml 3 1 1
Y D cos
1 cos .
2
1 / cos
2 (13)
det .M/ 3 2
from which one can deduce that Y < 0 for all the configurations where
1 and
2 are
not too far from each other and especially for
1
2 . Looking at the output–input
inverse dynamics relationship of Eq. (7), for the case at hand, a negative value of Y
denotes that an assumed positive (upward) yRQ E might require a negative (clockwise)
direction of , which is against intuition and, actually, leads to instability of the
internal dynamics (inapplicability of the inverse dynamics control).
To be more specific about the physical reasons for the aforementioned instability,
let us consider that the arm is in rest position as seen in Fig. 3b,
1 D
2 D 0 and
P1 D
P2 D 0, and the task is to move the end point E in the upward direction, which
yields yRQ E .t0 / > 0 at the beginning of the maneuver. Since k, g, and h are equal
to zero at the initial state and are negligible just after the maneuver begins, which
results in fy D 0 and fy 0, respectively, the inverse control formula of Eq. (7), for
the case at hand, simplifies to D yRQ E =Y . Intuitively one can expect a positive sense
of required to initiate the upward motion of the end point E. This is not the case,
however. More strictly, the effect of a positive on link 2 is an upward reaction R in
joint A (Fig. 3c), which moved to the link mass center C2 isassociated with a couple
of clockwise (negative) torque T D R/s2 . For the geometrical and inertial settings
stated above, the acceleration of point E is then
R R s2 R R l 12 l 2R
yRE D .l2 s2 / D 2
D (14)
m2 JC 2 m 2 ml 2 m
The action of a positive has thus the “opposite” effect from what is expected—
a negative yRE is produced. The feature is related to the negative value of Y in the
output–input inverse dynamics model D yRQ E =Y , and a positive yRQ E will result in
the negative (clockwise) sense of . This “reverse” control is improper and, when
continued, leads to the collapse of the servo-constraint problem execution. A proper
solution to the problem requires therefore Y > 0.
The desired effect Y > 0 can be achieved in two ways. Firstly, as seen from
Eq. (12), one can redesign the system to modify appropriately its geometrical and
inertial characteristics; see also [21–23] where a similar approach was used. In
particular, for the case at hand, one may play with the values of m2 , JC2 and s2 ,
and Y > 0 can most effectively be achieved by decreasing s2 and increasing JC2 .
A redesign of an existing system may not always be possible, however. The other
possibility is therefore a reformulation of the imposed task yQ .t / and, in particular, a
reduction of sP value; see Eq. (12). The reference point should thus not be the end
point E (sP D l2 ) but an inner point P on link 2, and a reasonable choice is P ! C2
(sP D l2 /2). Using this setting, a rest-to-rest maneuver was modeled as
" 5 6 7 8 9 #
t t t t t
yQP .t /DyP 0 C 126 420 C540 315 C70 yPf yP 0

(15)
where yP0 D 0 m and yPf D 0.75 m are the initial (t0 D 0 s) and final (tf D 3 s) y
coordinates of P, and D tf t0 D 3 s. The system geometrical and inertial data
used in calculations were l1 D l2 D 1 m, m1 D m2 D 2 kg, si D li /2, and JCi D mi l2i /12,
and the spring and damper coefficients were c D 5 Nm/rad and d D 0.1 Nms/rad.
Selected simulation results obtained using the scheme illustrated in Fig. 2 (without
the outer loop), with the integration time step t D 0.001 s, are reported in Fig. 4.
As seen from the graphs, the internal dynamics is bounded, and the system
asymptotically approaches the final rest position after the maneuver is finished.
4 DAE Formulation of the Governing Equations
The output–input dynamics model of Eq. (7) applies to the orthogonal realization
of servo-constraints, conditioned upon det(Y) ¤ 0. A general formulation of the
servo-constraint problem should relate to all the realization types, including the
mixed orthogonal–tangent and pure tangent ones, for which det(Y) D 0. This aim
192 W. Blajer
1.0 40
yE
y [m] θ [deg] θ2
yP 20
0.5 θ1
maneuver 0

0.0
0 2 4 t [s] 6 0 2 4 t [s] 6
3.0 0.8
ω2
τ [Nm] ω [1/s]
0.0 ω1
0.0
-0.4
-3.0
0 2 4 t [s] 6 0 2 4 t [s] 6
Fig. 4 Simulation results of the rotational arm executing a rest-to-rest y-motion of point P, with
sP D l2 /2 (P ! C2 ), and the links modeled as identical homogeneous bars
can be achieved by formulating the governing equations as DAEs. In [5, 6] such a

formulation was proposed in the configuration coordinates q, resulted in 2f C m
DAEs in the same number of variables: q, v D q; P and u. The initial DAEs,
composed of f kinematic relationships qP D v, f dynamic equations (1), and m servo-
constraint equation (2), were then transformed to a numerically more tractable form
with index [2, 10] reduced by two. The reduction is consequent to the projection
of the dynamic equations into the orthogonal H and tangent G subspaces and
application of the servo-constraint equation (4) at the acceleration level. With the
use of an f (f m) matrix G which is an orthogonal complement to H, i.e.,
H G D 0 () GT HT D 0, and whose columns represent the spanning vectors of
G, the reduced index DAEs are (see [5, 6, 23] for details)
qP Dv qP Dv
GT M vP D GT .g k/ C GT B u A .q/ vP D a .q; v; u/
()
0 D H M1 .g k/ C H M1 B u C h yQR .t / 0 D ˜ .q; v; u; t /
0 D ˆ .q/ yQ .t / 0 D ® .q; t /
(16)
where A and a are of dimensions (f m) f and (f m) 1, respectively.

Another DAE formulation of the governing equations can be obtained by
applying a coordinate transformation, exploited in [21, 22] and used hereafter in
a little modified way. The configuration coordinates are first partitioned into m
actuated and f m unactuated coordinates, written symbolically as q D [qTa qTu ]T ,
and the partition is conditioned upon the m m matrix Ba , gathering the rows of B
related to qa , and is of maximal rank, det(Ba ) ¤ 0. The new set of coordinates is
then q0 D [yT qTu ]T , where, compared to q D [qTa qTu ]T , the outputs y are introduced
instead of qa . At the acceleration level, the transformation formula is

yR H qR h Ha Hu qR a h
qR 0 D D C D C D H0 .q/ qR C h0 .q; q/
P
qR u qR u 0 0 I qR u 0
(17)
:
where Ha and Hu are of dimensions m m and m (f m), H D [Ha :: Hu ], and H0
is the f f matrix of transformation, whose inverse is conditioned upon det(Ha ) ¤ 0.
By applying qR D M1 .g k C B u/ obtained from Eq. (1), the underactuated
system dynamics in the new coordinates can then be formulated in the following
resolved form:
yR D H 1 .g k/ C h C H M1 B u
" M # " # yR D fy .q; qP / C Y .q/ u
: : ()
qR u D 0 : I: M .g k/ C 0 ::
1 I M1 B u qR u D fq .q; qP / C Q .q/ u
(18)
The arising governing equations are then (f m) C (f m) C m C m C m

D 2f C m DAEs in q, v, and u (see [9, 23] for more details):
qP u D vu qP u D vu
vP u P C Q .q/ u
D fq .q; q/ vP u P u/
D b .q; q;
0 P C Y .q/ u yRQ .t /
D fy .q; q/ () 0 P u; t /
D ˜ .q; q; (19)
0 D H .q/ qP yPQ .t / 0 P t/
D ” .q; q;
0 D ˆ .q/ yQ .t / 0 D ® .q; t /
The two DAE formulations, introduced in Eqs. (16) and (19), are equivalent
and lead to same solutions of the servo-constraint problem. They have the same
dimension 2f C m and are expressed in q, v, and u and are applicable irrespective
of the servo-constraint realization type: orthogonal (p D m), orthogonal–tangent
(0 < p < m), or tangent (p D 0); see Eq. (6). They have also the same index (equal
to one for the orthogonal realization of servo-constraints) and exceeding one for the
non-orthogonal realizations [5, 6, 9, 21, 23]. Finally, the third and fourth algebraic
equations of DAEs (16) are identical to the third and fifth ones in DAEs (19).
A difference is that the f kinematic relationships qP D v in DAEs (16) are replaced
in DAEs (19) with f m kinematic relationships qP u D vu and m servo-constraint
P t / D 0 at the velocity level. The f m differential equations
equations ” .q; q;
A .q/ vP D a .q; v; u/ describing the system dynamics in G subspace are also
replaced with f m differential equations vP u D b .q; q;
P u/ that govern the system
dynamics in the unactuated coordinate directions.
194 W. Blajer
A simple numerical code was proposed in [5] to solve DAEs (16), based on
Euler backward differentiation scheme in which the variable time derivatives are
approximated with their backward differences with respect to the integration time
step t. The same can be applied to DAEs (19), resulting in
.qu /nC1 .qu / n t .vu / nC1 D0

.vu /nC1 .vu / n t b .qnC1 ; vnC1 ; unC1 / D0
˜ .qnC1 ; vnC1 ; unC1 ; tnC1 / D0 (20)
” .qnC1 ; vnC1 ; tnC1 / D0
® .qnC1 ; tnC1 / D0
[see [5, 6] for a similar numerical code related to DAEs (16)]. Given qn and vn at
time tn (un are not required), the values qn C 1 , vn C 1 and un C 1 at time tn C 1 can then
be obtained as a solution of the above nonlinear equations, and in this way the solu-
tion is advanced from tn to tn C 1 . The rough computational scheme is of acceptable
accuracy for appropriately small values of t. The accuracy can be improved by
applying more precise higher-order backward difference approximation methods or
other specialized DAE solvers [2]. The controller structure based on the solution to
DAEs (16)/(19) is seen in Fig. 5.
Case study 2: From orthogonal to tangential realization of servo-constraints. Let us
consider another two-degree-of-freedom system (Fig. 6a) that consists of a carriage
of mass m1 , moving horizontally, and a slider of mass m2 , moving along a stick
mounted on the carriage and inclined horizontally at angle ˛. The actuating force F
is applied to the carriage, u D [F ], and the specified output is a desired horizontal
position of the slider with respect to the ground, yQ .t / D ŒxQ 2 .t /. The spring and
damper coefficients are, respectively, c and d, and b is the value of s2 in the rest state
of the system. Friction in the system is neglected. Choosing q D [x1 s2 ]T , where x1
is the carriage position with respect to the inertial frame and s2 is the slider position
with respect to the carriage, the components of the dynamic equations of the system,
corresponding to Eq. (1), are
.
~
q ,q
y .
. .. .. . . y, y
y~ ystab= y + α( y~ y ) + β ( y~ y ) DAEs (16)/(19) SYSTEM
..
y~ . . u
y, y q ,q
inner loop
outer loop
Fig. 5 Structure of control based on DAE formulation

a s2 b
s2
c m2 c m2
m1 d m1
d
F F
x1 x1
~
x2(t) ~
x2(t)
Fig. 6 The spring–mass system mounted on a movable carriage, the cases of (a) skew-orthogonal
and (b) tangential realization of the servo-constraint imposed
m1 C m2 m2 cos ˛ 0 1
MD I k D 0I g D IB D
m2 cos ˛ m2 c .s2 b/ d sP2 0
(21)
The servo-constraint equations, introduced in Eqs. (2)–(4), are then defined by
' .q; t / D x1 C s2 cos ˛ xQ 2 .t / D 0I H D Œ1 cos ˛ I h D Œ0 (22)
:
and the matrices Y D H M 1 B and Q D [0 :: I ] M 1 B, introduced in Eq. (18), are
" #
m2 sin2 ˛
Y D ŒY D I Q D ŒQ D Œm2 cos ˛ (23)
m1 m2 C m22 sin2 ˛
Assuming ˛ 2 < 0, /2 >, it is evident from Eq. (23) that for ˛ D /2 one has
Y D 1/(m1 C m2 ), and an ideal orthogonal realization of the servo-constraint is
observed in which the output-inverse dynamics (motion of the two masses along
x1 ) and the internal dynamics (relative vertical motion of the slider) are decoupled,
Q D 0. For 0 < ˛ < /2 a skew-orthogonal realization arises, Y > 0 and Q ¤ 0, in
which the inverse dynamics control and the internal dynamics are mutually related.
Finally, for ˛ D 0 (Fig. 6b), there is no direct input–output relationship, Y D 0, and
the tangential realization of the servo-constraint occurs.
Test simulations were carried out for m1 D 1 kg, m2 D 2 kg, c D 5 N/m, and
b D 1 m, and different values of the inclination angle ˛ and damping coefficient d
were used. The specified output xQ 2 .t / was a rest-to-rest maneuver according to Eq.
(15) with x20 D 0.5 m and x2f D 2.5 m instead of yP0 and yPf , respectively, and the
maneuver duration was D 6 s. The integration time step for the DAE formulation
(19) was t D 0.001 s.
The results shown in Fig. 7 are for ˛ D 40ı (skew-orthogonal realization of
the servo-constraint). As seen, realization of the specified slider motion induces
196 W. Blajer
1.5 1.5
d=0 Ns/m d=2 Ns/m
F [N] F [N]
0 0
-0.75 -0.75
-1.5 -1.5
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
3 ~
x2 3 ~
x 2
[m] [m]
2 x1 2 x1
s2 s2
1 1
0 d=0 Ns/m 0 d=2 Ns/m
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
. ~x.
1 1
[m/s] ~x d=0 Ns/m [m/s] 2
d=2 Ns/m
2
0.5 .
x1
0.5
.
.
s2
x1
0 0 .
s2
-0.5 -0.5
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
Fig. 7 Simulation results for ˛ D 40ı (orthogonal realization) and for d D 0 Ns/m (left column)
and d D 2 Ns/m (right column)
substantial internal motion. When not damped, for t > , after the maneuver is
finished and the slider horizontal position is kept constant, the interaction from the
internal motion (the carriage moves to and fro, together with the slider raising up
and down) must be compensated by appropriate variations in F(t). Nonetheless, the
internal motion is bounded (stable). For d D 2 Ns/m the residual motion is damped
and the required F(t) goes to zero.
In Fig. 8, the results obtained for ˛ D 0ı (tangential realization of the servo-
constraint) are shown. As seen, the case with damping (right columns) is similar
to those obtained for ˛ D 40ı —in the present case, for t > , the residual motion
is slightly quicker damped to the final state. The case without damping is now
qualitatively different from that obtained for ˛ D 40ı in Fig. 7. The motion is
completely finished at t D . This is because the problem is differentially flat, and
no internal dynamics is left, which is explained shortly below.
1.5 1.5
d=0 Ns/m d=2 Ns/m
F [N] F [N]
0 0
-0.75 -0.75
-1.5 -1.5
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
3 3
[m] ~ [m] ~
2 x2 2 x2
1 1
s2 s2
x1 x1
0 d=0 Ns/m 0 d=2 Ns/m
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
1 1
[m/s] ~x. d=0 Ns/m [m/s] ~x. d=2 Ns/m
2 2
0.5 .
x1
0.5
.
x1
.
s .
0 2 0 s
2
-0.5 -0.5
0 2 4 6 8 t [s] 10 0 2 4 6 8 t [s] 10
Fig. 8 Simulation results for ˛ D 0ı (tangential realization) and for d D 0 Ns/m (left column) and
d D 2 Ns/m (right column)
5 Differentially Flat and Non-flat Servo-constraint Problems
The servo-constraint problems can be differentially flat [14, 18, 20] or non-flat. In
case of flatness of the inverse dynamics problems described in Eqs. (1) and (2), all
m control inputs u and 2f state that q and v D qP can be algebraically expressed in
terms of the desired m outputs yQ .t / and their time derivatives up to a certain order
r, which is by one smaller than the index of the initial DAEs. The inverse dynamics
solution is then explicit in time and unique, denoted

Q / D q yQ ; yPQ ; : : : ; yQ r2
q.t

vQ .t / D v yQ ; yPQ ; : : : ; yQ r1 (24)

Q / D u yQ ; yPQ ; : : : ; yQ r ;
u.t
and there is no internal dynamics in the system [5, 7, 9, 20–23]. The flatness-
based analytical solution of Eq. (24) is often featured by substantial complexity,
198 W. Blajer
which may render its obtainment difficult/impractical for more complex problems
of technical relevance. By contrast, the non-flat servo-constraint problems involve
some (unspecified) internal dynamics, which needs to be bounded.
For fully actuated systems in fully prescribed motion, m D f, the inverse dynamics
problem is always flat with the order r D 2. For underactuated systems in partly
specified motion, m < f, in the case of orthogonal realization of the servo-constraints
imposed, the servo-constraint problem is always non-flat (is a dynamic problem),
and of critical importance is to assure stability of the remained internal dynamics;
see Sect. 3. The servo-constraint problems for the cases of mixed orthogonal–
tangential and pure tangential realizations of servo-constraints can then be either flat
or non-flat. Examples of the differentially flat inverse dynamics problems are cranes
executing a load prescribed motion [5–7] and aircrafts in prescribed trajectory
flight [8], both characterized by mixed orthogonal–tangential realization of servo-
constraints, and the trajectory tracking problem for flexible joint manipulators [3,
12, 24] (pure tangential realization of servo-constraints). A simple illustration of the
differentially flat servo-constraint problem is given below.
Case study 3: Differential flatness. Let us reconsider the servo-constraint problem
for the variant ˛ D 0 (Fig. 6b) of the cart–mass system. As already stated, the
case results in the tangential realization of the servo-constraint defined in Eq. (22),
Y D 0, and the output yQ .t / D ŒxQ 2 .t / tracking is achieved through the spring and
damper forces (the actuating force F is not directly involved). One can then become
convinced that the problem is differentially flat for d D 0 (no damping) and is non-
flat for d > 0 (with damping). The conclusion can be deduced from the simulation
results shown in Fig. 8. For d D 0 (left column graphs), both the states and actuation
of the system, determined from the inverse dynamics, achieve the final steady state
values exactly at the instant the maneuver is finished (at t D 6 s) and must thus be
consequent only upon the specified output xQ 2 .t / and its time derivatives (no internal
dynamics remains). For d > 0 there is some internal dynamics, exemplified in the
fact that the system achieves the final steady state values at t > 6 s (after the specified
maneuver is finished). For the flat case, the inverse dynamics solution as in Eq. (24)
is the following:
m2 R m2 R
xQ 1 .t / D xQ 2 C xQ 2 bI sQ2 .t / D b xQ 2 I
c c
PQx1 .t / D PQx 2 C m2«Qx2 I PQs2 .t / D m2«Qx2 I (25)
c c
Q m1 m2 RR R
F .t / D xQ 2 C .m1 C m2 / xQ 2
k
with the order parameter r D 4. There is no such a solution for the non-flat case.
6 Summary and Conclusions
The diversity in the inverse simulation problem for underactuated mechanical

system in partly specified motion, referred here to as servo-constraint prob-
lem, concerns numerous issues. Of fundamental relevance is the way the servo-
constraints on the system are realized. By contrast to a system subject to (ideal)
passive constraints, whose reactions are orthogonal to the constraint manifold,
the orientation of actuating forces of the system (control reactions) with respect
to the servo-constraint manifold may vary from orthogonal to tangential. Only
for the orthogonal realization, conditioned upon the existence of the output–input
relationship of Eq. (7), det(Y) ¤ 0, the governing equations can be formulated as
ODEs (9). For the mixed orthogonal–tangent and pure tangent realizations of servo-
constraints, the governing equations must be formulated as DAEs. Two variants
of the DAE formulations were motivated in this paper: DAEs (16) that use the
system configuration coordinates q and DAEs (19) that are based on the coordinate
transformation q D [qTa qTu ]T ! q0 D [yT qTu ]T . An advantage of using the DAE
formulations is that they are valid for all possible types of realization of servo-
constraints.
The servo-constraint problem for the case of orthogonal realization of servo-
constraints is always non-flat. More strictly, in addition to the output–input inverse
dynamics model of Eq. (7), a residual internal dynamics in the system remains.
The internal dynamics is usually influenced by the inverse dynamics control, and,
vice versa, the enforced internal dynamics has an impact on the output–input
relationship. Stability of the internal dynamics is therefore of critical importance for
applicability of the servo-constraint problem solution, which can be achieved either
by an optimized design of the underactuated system or appropriate selection of the
output variables. The servo-constraint problems for the cases of mixed orthogonal–
tangential and pure tangential realizations of servo-constraints can then be either
non-flat or flat. In case of flatness, the servo-constraint problem becomes pure
algebraic, in which all the state and control variables can explicitly be determined
in terms of the specified in time outputs and its time derivatives.
The feedforward control of underactuated systems in partly specified motion
can effectively be designed following the schemes seen in Fig. 2 (for orthogonal
realization of servo-constraints) or in Fig. 5 (for all types of realization of servo-
constraints), where ’ D 0 and “ D 0. The design of a feedback controller, to provide
stable tracking of the specified outputs in the presence of external disturbances and
modeling uncertainties, is in general more challenging. For the case of orthogonal
realization of servo-constraints, provided that the assisted internal dynamics is
bounded, the feedback controller can be designed by applying the stabilized
output form defined in Eq. (8), which is as a direct generalization of the control
laws used for fully actuated systems. Similar feedback controllers have also been
applied to the differentially flat problems with tangent and orthogonal–tangent
realizations of servo-constraints [5–7]. Robustness of the controllers has however
been verified only through numerical tests and for “smooth” external disturbances
200 W. Blajer
during the simulated tracking of the reference trajectories. Due to the high index
of the governing DAEs (16) used there, equal to three, “higher-order” controllers,
discussed in [12], might be more appropriate. These issues are not addressed in this
contribution.
References
1. Abdel-Rahman, E.M., Nayfeh, A.H., Masoud, Z.N.: Dynamics and control of cranes: a review.
J. Vib. Contr. 9, 863–908 (2003)
2. Ascher, U.M., Petzold, L.R.: Computer Methods for Ordinary Differential Equations and
Differential-Algebraic Equations. SIAM, Philadelphia (1998)
3. Benosman, M., Le Vey, G.: Control of flexible manipulators: a survey. Robotica 22, 533–545
(2004)
4. Blajer, W.: A geometrical interpretation and uniform matrix formulation of multibody system
dynamics. Z. Angew. Math. Mech. 81, 247–259 (2001)
5. Blajer, W., Kołodziejczyk, K.: A geometric approach to solving problems of control con-
straints: theory and a DAE framework. Multibody Syst. Dyn. 11, 343–364 (2004)
6. Blajer, W., Kołodziejczyk, K.: Control of underactuated mechanical systems with servo-
constraints. Nonlinear Dyn. 50, 781–791 (2007)
7. Blajer, W., Kołodziejczyk, K.: Improved DAE formulation for inverse dynamics simulation of
cranes. Multibody Syst. Dyn. 25, 131–143 (2011)
8. Blajer, W., Graffstein, J., Krawczyk, M.: Modeling of aircraft prescribed trajectory flight as
an inverse simulation problem. In: Awrejcewicz, J. (ed.) Modeling, Simulation and Control of
Nonlinear Engineering Dynamical Systems, pp. 153–162. Springer, Netherlands (2009)
9. Blajer, W., Seifried, R., Kołodziejczyk, K.: Diversity of servo-constraint problems for underac-
tuated mechanical systems: a case study illustration. Solid State Phenom. 198, 473–482 (2013)
10. Campbell, S.L., Gear, C.W.: The index of general nonlinear DAEs. Numer. Math. 72, 173–196
(1995)
11. Chen, Y.-H.: Equations of motion of mechanical systems under servoconstraints: the Maggi
approach. Mechatronics 18, 208–217 (2008)
12. De Luca, A.: Trajectory control of flexible manipulators. In: Siciliano, B., Valavanis, K.P. (eds.)
Control Problems in Robotics and Automation, pp. 83–104. Springer, London (1998)
13. Fantoni, I., Lozano, R.: Non-linear Control for Underactuated Mechanical Systems. Springer,
London (2002)
14. Fliess, M., Lévine, J., Martin, P., Rouchon, P.: Flatness and defect of nonlinear systems:
introductory theory and examples. Int. J. Control 61, 1327–1361 (1995)
15. Kirgetov, V.I.: The motion of controlled mechanical systems with prescribed constraints
(servoconstraints). J. Appl. Math. Mech. USS 31, 433–466 (1967)
16. Lee, H.-H.: Modeling and control of a three-dimensional overhead crane. J Dyn. Syst. Meas.
Control T ASME 120, 471–476 (1998)
17. Paul, R.P.: Robot Manipulators: Mathematics, Programming, and Control. MIT, Cambridge,
MA (1981)
18. Rouchon, P.: Flatness based control of oscillators. Z. Angew. Math. Mech. 85, 411–421 (2005)
19. Sahinkaya, M.N.: Inverse dynamic analysis of multiphysics systems. Proc. Inst. Mech. Eng.
I J. Syst. Control Eng. 218, 13–26 (2004)
20. Sara-Ramirez, H., Agrawal, S.K.: Differentially Flat Systems. Marcel Dekker, New York
(2004)
21. Seifried, R.: Integrated mechanical and control design of underactuated multibody systems.
Nonlinear Dyn. 67, 1539–1557 (2012)
22. Seifried, R.: Two approaches for feedforward control and optimal design of underactuated
multibody systems. Multibody Syst. Dyn. 27, 75–93 (2012)
23. Seifried, R., Blajer, W.: Analysis of servo-constraint problems for underactuated multibody
systems. Mech. Sci. 4, 113–129 (2013)
24. Spong, M.W.: Underactuated mechanical systems. In: Siciliano, B., Valavanis, K.P. (eds.)
Control Problems in Robotics and Automation, pp. 135–150. Springer, London (1998)
25. Thompson, D., Bradley, R.: Inverse simulation as a tool for flight dynamics research: principles
and applications. Progr. Aero. Sci. 42, 174–210 (2006)
Rare Phenomena and Chaos in Switching
Power Converters
Dmitry Pikulin
Abstract This paper demonstrates the general applicability of the method of

complete bifurcation groups to provide the global analysis of nonlinear dynamics
of simple hybrid electronic systems. Our research is devoted to the investigation
of complex nonlinear phenomena, including rare attractors and chaotic modes of
operation in one of the most widely used electronic switching power converters
(SPC)—boost one. Firstly, the discrete-time model, describing the operation of the
mentioned power converter, is defined. Then the obtained model is used, providing
numerical investigation of the dynamics of SPC on the basis of the method of
complete bifurcation groups: the bifurcation map is constructed in two-parameter
plane, the complete one-parameter bifurcation diagrams for several sections of
the bifurcation map are obtained, and the most significant features of nonlinear
dynamics of boost converter are studied. The results obtained in this paper prove that
the design of reliable switching power converters is possible effectively applying the
method of complete bifurcation groups, which allows the prediction and avoidance
of occurrence of undesirable regimes (such as chaotic modes of operation, variety
of rare attractors, complex protuberances, regions of unstable periodic infinitium) in
the operation of these devices.
1 Introduction
Nowadays the existence of complex nonlinear phenomena, such as subharmonic

oscillations and chaotic attractors, exhibited during the operation of widely used
switching power converters (SPC), is commonly accepted and proved fact [2, 7, 8].
Sudden appearance of undesirable nonlinear operating regimes could damage not
only the SPC itself, but also lead to the collapse of the whole system it supports [8].
Realizing the importance of the problem a variety of different methodologies has
been proposed, providing the tools for the prediction of the first bifurcation patterns
and subsequent evolution to chaotic oscillations [1, 8]. One of the most important
D. Pikulin ()
Faculty of Electronics and Telecommunication, Riga Technical University, Riga, Latvia
e-mail: dmitrijs.pikulins@rtu.lv

204 D. Pikulin
plots used for the analysis of the dynamics of nonlinear systems with variable
parameters is the bifurcation diagram [3]. The majority of methodologies utilize
the so-called “brute-force” bifurcation diagrams, depicting only stable operating
regimes and discarding the information about existing unstable modes of operation
as unnecessary. However it has been shown that the disregard of the branches of
bifurcation diagram corresponding to the unstable orbits may lead to the loss of
very important information about the global dynamics of the system [5, 6, 9, 10].
Taking into consideration the mentioned disadvantages of commonly used
techniques, this research utilizes the relatively new approach—method of complete
bifurcation groups (MCBG)—developed in the Institute of Mechanics of Riga
Technical University [6, 9, 10], for the investigation of the global dynamics of
one of the most widely used electronic devices, boost SPC. During the analysis
of the constructed complete bifurcation diagrams, the significance of the topology
of unstable branches for the exploration of complex smooth and non-smooth
bifurcation patterns is explored.
2 The Model of the Boost Switching Power Converter
An elementary boost power converter consists of a basic RLC circuit, diode, and
switching element (see Fig. 1). The aim of this circuit is to preserve preferred output
voltage vc higher than that provided by the input source E. This can be achieved by
appropriately turning on and off the switch S, so that the circuit is repeatedly forced
by the external voltage source [7].
The position of the switching element is defined by parameters of the imple-
mented voltage feedback loop that compares the output voltage to the reference
value, amplifies the difference (see gain k in the Fig. 1), and generates the
corresponding pulse-width modulated signal.
Although SPC, being inherently hybrid systems, are modeled using switched
state-space models, discrete-time models are also found to be useful in analytical
Vref Parameter Value

-
k Switching period, T 333 (µs)
+ Inductance, L 208 (µH)
Load resistance, R 12.5 (W)
L D Capacity, C 222 (µF)
S + Input voltage, E 11-19 (V)
E C R vc
Reference voltage, Vref 25(V)
-
Feedback gain, k 0.1-0.5
Fig. 1 Simplified model of the boost SPC and its general parameters [7]
Rare Phenomena and Chaos in Switching Power Converters 205
and especially numerical investigation of nonlinear dynamics [4]. The most widely
used discrete-time model for switching converters is called iterative map, where the
state variables are periodically sampled at time instants t D nT (T is the switching
period of SPC). The map F is then the relationship that links the previous and
the current state variable samples: x(n C 1) D F(x(n)). This kind of mappings is
used as an effective tool for locating bifurcation points and constructing bifurcation
diagrams.
Tse in his book [7] proves that the operation of the boost converter, operating
in discontinuous current mode (DCM) under voltage-mode control (shown in the
Fig. 1), could be described by simple one-dimensional map in closed form for the
capacitor voltage:
2
vC .n C 1/ D ˛vC .n/ C ˇE 2 H D k vC .n/ Vref = .vC .n/ E/ (1)
where
p
D D VC .1 ˛/ .VC E/ =ˇE 2 I ˛ D 1 T =RC C T 2 =2R2 C 2 I ˇ D T 2 =2LC;
(2)
D, steady-state duty cycle; VC , steady-state output voltage; H(.), limits the range
of duty cycle between 0 and 1; Vref , reference voltage; k, small signal gain; and T,
switching period. Other notations and parameter values of the boost SPC under test
are depicted in Fig. 1.
The discrete model defined in (1) and (2) will be used during the complete
bifurcation analysis of the boost converter, as it has been shown in [7]—even
this simple one-dimensional map is capable of predicting a variety of nonlinear
phenomena in the operation of SPC with high level of accuracy.
3 The Construction of Bifurcation Map
At the beginning of the investigation the primary and secondary bifurcation param-
eters (input voltage E and small signal feedback gain k), defining the qualitative
changes in the dynamics of SPC, as well as dimensions and location of regions
with different types of periodic (or chaotic) regimes in the parameter plane, are
selected. Then the bifurcation map in the plane f(k, E):0.05 < k < 0.4; 11 < E < 19g
is obtained, providing the information about the division of the parameter plane into
regions with different periodic and chaotic operating regimes. Figure 2 depicts the
bifurcation map obtained for the parameters of the system defined in Fig. 1. Each
hatching style in the diagram represents different periodic (or chaotic) operating
regimes, specified in the legend (up to P10).
The obtained bifurcation map allows concluding that in the certain range of
parameter space the boost converter operates in the stable P1 (the only desirable
206 D. Pikulin
P3/P6
P4/P8 P6 P5 P5
19
P1
P2
P3
18
P4
P5
P6
17 P7
P8
P10 P9
16 P10
P>10
P9
/Ch
E,V
15
P8
P7
14 P1 P2
P6
13
P5
P6
12 P7
P8
P4
11
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Feedback gain k
Fig. 2 The bifurcation map of the boost switching power converter in the kE parameter plane
operating mode for SPC) that undergoes period-doubling bifurcation cascade with
subsequent transition to chaotic mode of operation. The chaotic region is not
homogenous, as there exists a great variety of periodic windows that are formed
by certain types of rare attractors (RA, periodic or chaotic regimes, being stable
in narrow range of parameter values [9]). The existence of RA is proved and the
appropriate identification is provided, during the construction of complete one-
parameter bifurcation diagrams.
The closer look on the topology of different periodic regions of the bifurcation
map allows ascertaining the fact that for the small values of the feedback gain k,
only smooth bifurcations and transitions to chaos could be observed. However, for
E < 14 V and k > 0.3 some kind of non-smooth phenomena and sudden changes in
the bifurcation structures could be expected.
The next section represents the complete bifurcation diagrams, constructed as the
cross sections of the bifurcation map in Fig. 2, selecting the small signal feedback
gain k as the primary bifurcation parameter. These diagrams provide the in-depth
information about the structure of bifurcation paths, types of RA, and chaotization
scenarios.
4 Periodic Skeleton and Complete Bifurcation Diagrams
Prior to the construction of the first bifurcation diagram, the topology and approx-
imate structure of the diagram could be estimated, obtaining the periodic skeleton
[9] for E D 16 V and k D 0.4 (other parameter values are defined in Fig. 1). The
data obtained in the process of construction of periodic skeleton are summarized in
Table 1.
Each regime in the periodic skeleton is described by the coordinates of its fixed
point and multiplier, defining the stability of the regime. As it could be seen, all
found periodic regimes (up to P6) are unstable, so in this case the converter operates
either in stable periodic regime nP with n > 6 or in chaotic mode of operation. As it
is shown in Table 1, several periodic regimes form different bifurcation groups (BG,
stable and unstable branches of bifurcation diagram, having common bifurcation
points [9])—one of the cornerstones of the MCBG. For example, P1–P2–P4 regimes
define the origin of period-doubling cascade of 1 T BG.
The information obtained in the periodic skeleton is used in the process of
numerical continuation (path-following) for the construction of complete bifurca-
tion diagrams, depicting the most significant nonlinear phenomena, observed in
boost converter and showing main paths to chaos.
Table 1 Periodic skeleton (up to P6) of the boost converter under study for E D 16 V and k D 0.4
Bifurcation group (BG) Period Fixed point position (vC,n *) Multiplier œ Stability
1T P1 25 7.2880 Unstable
P2 27.8377 10.1033 Unstable
P4 24.6659 515.7154 Unstable
3T P3 27.5612 11.8149 Unstable
25.0486 70.3031 Unstable
P6 24.6573 792.7106 Unstable
4T P4 27.3159 12.9183 Unstable
24.3970 76.7561 Unstable
5 T1 P5 24.66123 714.3610 Unstable
24.4269 118.0974 Unstable
5 T2 P5 25.1289 77.8386 Unstable
27.0880 13.6282 Unstable
6 T1 P6 24.6978 5233.2455 Unstable
25.0391 820.4291 Unstable
6 T2 P6 24.1754 581.3176 Unstable
24.7462 126.6094 Unstable
6 T3 P6 23.9490 75.5339 Unstable
26.8706 14.0650 Unstable
208 D. Pikulin
Fig. 3 The complete bifurcation diagram of the boost converter for E D 16 (V)
The obtained bifurcation diagrams (shown in Figs. 3 and 4) allow ascertaining

that for small values of parameter k, the system operates in a stable P1 regime that
loses its stability as a result of classical period-doubling, leading to the subsequent
formation of chaotic region of 1 T BG (designated as UPI-1). Thus the increase
of small signal feedback gain leads to the smooth transition to chaotic mode of
operation. However this chaotic regime is not robust, as the great variety of tip type
rare attractors, corresponding to 3–6 T3 bifurcation groups, could be observed (see,
e.g., small stable regions in Fig. 3). The topology of these RA is similar to the
structure of 1 T BG shown in Fig. 4, including the complete period-doubling route
to chaos. The overall dynamics of the system, including the route to chaos, in this
case is completely defined by smooth period-doubling and saddle-node bifurcations,
caused by the transition of the corresponding multipliers outside unit circle.
It should be noted that in all subsequent complete bifurcation diagrams dark
lines represent stable periodic regimes and light lines, the unstable ones and all
bifurcation points are accordingly marked.
5 Non-smooth Bifurcations and Submerged Isles
As it has been already mentioned for small values of input voltage E, the non-smooth
bifurcations, manifesting as sudden jumps of characteristic multipliers, determine
the dynamics of the converter. Thus, changing the value of secondary bifurcation
Fig. 4 The detailed structure of 1 T bifurcation group
parameter E allows the construction of bifurcation diagrams with more complex

topological bifurcation groups, including uncommon protuberances, submerged
isles, and non-smooth bifurcations. Figure 5 shows an example of detailed structure
of complex protuberances of 4 T bifurcation group and submerged isles of 12 T
bifurcation group for E D 11.4 V.
The topology of protuberances and the submerged islands (see [9]) in this case
is defined by complex interaction of smooth saddle-node and period-doubling
bifurcations (with classical transition to chaotic mode of operation) from the left
side, as well as non-smooth bifurcations from the right, leading to abrupt changes in
the branches of bifurcation diagrams and sudden loss of stability of definite groups
of periodic regimes. Thus the chaotization of the system as well as the resumption
of stable P8 operation is governed by simultaneous loss of stability through
smooth period-doubling cascade and rapid changes in the dynamics, defined by
the appearance of non-smooth bifurcations. It should be noted that in spite of rather
uncommon structure, the submerged isles include small regions of stable periodic
regimes—tip type rare attractors—coexisting with stable P8 regime (see Fig. 5).
6 Conclusions
In the present paper it has been demonstrated that the method of complete
bifurcation groups could be successfully applied to the investigation of complex
nonlinear dynamics of widely used electronic systems—SPC.
210 D. Pikulin
Fig. 5 The detailed structure of 4 and 12 T bifurcation groups for E D 11.4 V
It has been shown that the use of MCBG allows carrying out global bifurcation
analysis of mentioned piecewise-smooth hybrid systems, exploring new nonlinear
effects, periodic and chaotic regimes, rare attractors, protuberances, and submerged
isles. The results include the construction of bifurcation map, obtaining periodic
skeleton, and the subsequent construction of bifurcation diagrams depicting stable
and unstable periodic regimes, as well as regions of unstable periodic infinitium.
The obtained results establish that in SPC within the definite parameter range
rather specific types of non-smooth bifurcations define the global topology of
bifurcation patterns, leading to unexpected and abrupt changes in the system
dynamics.
References
1. Awrejcewicz, J.: Bifurcations and Chaos in Coupled Oscillators. World Scientific, Singapore
(1991)
2. Banerjee, S., Verghese, G.C. (eds.): Nonlinear Phenomena in Power Electronic: Attractors,
Bifurcations, Chaos and Nonlinear Control. Wiley, Hoboken (2001)
3. Di Bernardo, M., Budd, C.J., Champeneys, A.R., Kowalczyk, P.: Piecewise-Smooth Dynamical
Systems: Theory and Application. Springer, Berlin (2008)
4. Hammil, D.C., Deane, J.H.B., Jeferries, D.J.: Modeling of chaotic dc/dc converters by iterative
nonlinear mappings. IEEE Trans. Circ. Syst. Pt I 35, 25–36 (1992)
5. Pikulins, D.: Effects of non-smooth phenomena on the dynamics of DC-DC converters. Sci. J.
RTU Ser. 4 29, 119–122 (2011)
6. Schukin, I., Zakrzhevsky, M.: Application of software spring and method of complete
bifurcation groups for the bifurcation analysis of nonlinear dynamical systems. J. Vibroeng.
10, 510–518 (2008)
7. Tse, C.K.: Complex Behavior of Switching Power Converters. CRC, Boca Raton (2003)
8. Vilamitjana, E.R., Alarcon, E., El Aroudi, A.: Chaos in Switching Converters for Power
Management. Springer, New York (2012)
9. Zakrzhevsky, M.: New concepts of nonlinear dynamics: complete bifurcation groups, protu-
berances, unstable periodic infinitiums and rare attractors. J. Vibroeng. 10, 421–441 (2008)
10. Zakrzhevsky M.: Bifurcation theory of nonlinear dynamics and chaos. In: Periodic Skeletons
and Rare Attractors. Proceedings of the 2nd international symposium RA’11, pp. 26–30. RTU,
Riga (2011)
Constrained Motion of Mechanical Systems
and Tracking Control of Nonlinear Systems
Firdaus E. Udwadia and Harshavardhan Mylapilli
Abstract This paper aims to expose the interrelations and connections between
constrained motion of mechanical systems and tracking control of nonlinear
mechanical systems. The interrelations between the imposition of constraints on a
mechanical system and the trajectory requirements for tracking control are exposed
through the use of a simple example. It is shown that given a set of constraints,
d’Alembert’s principle corresponds to the problem of finding the optimal tracking
control of a mechanical system for a specific control cost function that Nature
seems to choose. Furthermore, the general equations for constrained motion of
mechanical systems that do not obey d’Alembert’s principle yield, through this
duality, the entire set of continuous controllers that permit exact tracking of the
trajectory requirements. The way Nature seems to handle the tracking control
problem of highly nonlinear systems suggests ways in which we can develop new
control methods that do not make any approximations and/or linearizations related
to the nonlinear system dynamics or its controllers. More general control costs
are used and Nature’s approach is thereby extended to general control problems.
A simple, unified methodology for modeling and control of mechanical systems
emerges. Examples drawn from diverse areas of control are provided dealing with
synchronization of multiple nonlinear gyroscopes, design of optimal Lyapunov
stable controllers for nonautonomous nonlinear systems, and energy control of
nonhomogeneous Toda chains.
F.E. Udwadia ()

Departments of Aerospace and Mechanical Engineering, Civil and Environmental Engineering,
Mathematics, Systems Architecture Engineering, and Information and Operations Management,
University of Southern California, Los Angeles, CA 90089, USA
e-mail: fudwadia@usc.edu
H. Mylapilli
Department of Aerospace and Mechanical Engineering, University of Southern California,
Los Angeles, CA 90089, USA
e-mail: mylapill@usc.edu

214 F.E. Udwadia and H. Mylapilli
1 Introduction
Sir Isaac Newton described the field of mechanics in his preface to the Principia in
the following words [12]:
In this sense rational mechanics will be the science of motions resulting from any forces
whatsoever, and the forces required to produce any motions, accurately proposed and
demonstrated.
Today, while the first part of Newton’s definition of mechanics has become our
usual understanding of this field, the second part is usually relegated primarily to
the field of control theory. Indeed, the problem that Newton famously solved was a
control problem: the determination of the forces required to be acting on the planets
so that their motions obey the observed motions described by Kepler’s first two laws.
[The third law that relates the periods of the planets to their mean radial distances,
which took Kepler a long time to ferret out, is not essential to obtain the inverse
square law of gravitation, though it is consistent with it.] To make certain that he
considered the control of mechanical systems within the purview of what he thought
of as mechanics, Newton went on in his preface to say that:
. . . I offer this work as the mathematical principles of philosophy, for the whole burden of
philosophy seems to consist in this–from the phenomena of motions to investigate the forces
of nature, and then from these forces to demonstrate the other phenomena [of motions] and
to this end the first and second books are directed.
In the above, the bracketed words have been added to the original quotation.
To illustrate the viewpoint of Newton, let us consider an elementary example, the
problem of finding the equations of motion of a spherical pendulum like the one
shown in Fig. 1.
The problem of finding the equation of motion of this simple system, which
consists of a particle of mass m constrained to move so that it is always at a fixed
distance, L, from its fixed point of support, O, in a nonuniform gravitational field,
can alternatively be looked at from the dual (the word “dual” in this paper is used
as in ordinary parlance and not in the restricted sense used in optimization theory)
standpoint of tracking control in the following way.
Fig. 1 A spherical pendulum

Constrained Motion of Mechanical Systems and Tracking Control of Nonlinear Systems 215
Consider a particle of mass m moving in a nonuniform gravitational field; it is

now required to determine the control force that needs to be applied to this particle
so that it is constrained to lie, at each instant of time t, on the sphere S2 defined by
the relation
'.x; y; z; t / WD x 2 .t / C y 2 .t / C z2 .t / L2 D 0: (1)
We will show that this control problem can be handily approached using the
theory of constrained motion of mechanical systems. Let us denote the 3 by 1 vector
(the 3-vector) q :D [x y z]T . Clearly, the equation of motion of the particle as it freely
moves in the nonuniform gravitational field in which the acceleration due to gravity
at any point is g(x, y, z, t) (see Fig. 1) is simply given by
2 3 2 3 2 3
m0 0 xR 0
R / WD 4 0 m 0 5 4 yR 5 D 4 mg.x; y; z; t / 5 WD Q ;
M q.t (2)
0 0 m zR 0
so that the acceleration of the particle at any time t can be written as the
3-vector a(q, t) D [0 g 0]T . [From here on, we shall drop the arguments of the various
quantities, unless needed for clarity.] We shall refer to Eq. (2) as the unconstrained
(or uncontrolled) equation of motion for the mechanical system. A control theorist
may prefer to call the equation a description of the “plant” whose trajectories need
to be controlled so that they satisfy the control requirement stated in (1). We observe
that from a physical viewpoint, the initial conditions, q(0) and q.0/,P can have any
arbitrary numbers for their respective components, depending, of course, on the
location and the initial velocity of the particle.
However, this particle’s motion described by Eq. (2) will not, in general, satisfy
the constraint—or alternatively put, our trajectory requirement—namely, that its
motion lies on the surface of the sphere described by Eq. (1). In order to achieve
this, an additional force will need to be applied to the particle so that its acceleration
is altered from a(q, t), and its equation of motion now becomes
M qR D Q C QC : (3)
This additional force, QC , which is a 3-vector, that needs to be applied to the

constrained system can be viewed as the force of constraint that ensures that Eq. (1)
is satisfied. It can also, from a dual perspective, be seen as the control force that
must be applied to the system described by Eq. (2), so that it satisfies the trajectory
requirement (1) that is imposed on it. The addition of a control force QC on the
right-hand side of Eq. (2) is not the only alteration needed in our update from the
unconstrained system (2) to the constrained system that now satisfies Eq. (1).
The initial conditions q(0), and q.0/P whose components could be chosen
arbitrarily in the case of system (2) can no longer be chosen arbitrarily. Instead
the components of q(t) at each instant of time (and hence also at the initial time)
Table 1 Analogous concepts in analytical dynamics and

control theory
Analytical
dynamics Control theory
Unconstrained system Uncontrolled system or plant
Constrained system Controlled system
Constraints Trajectory requirements
Constraint force Control force or control
P / must satisfy at each instant of

must satisfy relation (1); also, the components of q.t
time (and hence also at the initial time) the following relation:
P / C y.t /y.t
x.t /x.t P / C z.t /Pz.t / D 0; (4)
which is obtained by differentiating Eq. (1) with respect to time. One may want to
further differentiate Eq. (4) to obtain the relation
x.t /x.t R / C z.t /Rz.t / D xP 2 .t / yP 2 .t / zP2 .t /;

R / C y.t /y.t (5)
which can be written in matrix-vector form as
A qR D b; (6)
where A :D [x y z]T , and b D xP 2 .t / yP 2 .t / zP2 .t /. We note that for a given set
of initial conditions that satisfy Eqs. (1) and (4) at t D 0, Eq. (6) is equivalent to
Eq. (1).
This simple example thus illustrates the connections between the problem of
constrained motion and the problem of tracking control. Specifically, we find the
following analogous concepts given in Table 1 above.
As we go along, we will extend and refine this preliminary table. In what follows,
we will move back and forth between these dual concepts allowing ourselves to
be aided in our understanding of constrained motion to expose new insights into
trajectory control and vice versa.
2 General Constrained Mechanical Systems

and the Trajectory Control Problem
Our spherical pendulum problem was an illustrative “toy problem” created simply
to provide some insights into the connections that we are trying to establish.
The problem could, of course, have been made considerably more challenging by
P t / D 0,
requiring that the point of support, O, moves over a surface, say .q; q;
and/or requiring that the pendulum’s length varies in a prescribed manner so that
L.t / D f .q; q;
P t /. Though we will continue to use our toy problem to gain further
insights, we can now frame the general problem of constrained motion in analytical
dynamics as follows:
1. Consider an unconstrained (uncontrolled) nonlinear nonautonomous mechanical
system described by the equation
M .q; t / qR D Q.q; q;
P t / ; q.0/ D q0 and q.0/
P D qP 0 ; (7)
where M is a positive definite n by n matrix and q is an n-vector.

2. We require this system to satisfy the m consistent constraints (trajectory require-
ments) given by the relations
i .q; t / D 0; i D 1; 2; : : : ; h; and; (8)
P t/
i .q; q; D 0; i D h C 1; h C 2; : : : ; m: (9)
3. We need to find the constraint (control) force, QC , so that the constrained

(controlled) system described by
M .q; t / qR D Q.q; q; P t/ ;
P t / C QC .q; q; q.0/ D q0 ; and q.0/
P D qP 0 ; (10)
exactly satisfies the trajectory requirements (8) and (9).

The uncontrolled system (or “plant”) described by Eq. (7) that we shall be
dealing with is nonlinear and nonautonomous. We shall initially assume that q0
and qP 0 satisfy the trajectory requirements (8) and (9) at time t D 0. Later on, we will
relax this condition and allow for the possibility that the initial conditions q0 and
qP 0 may not lie on the constraint manifold described by (8) and (9). We define the
acceleration of the uncontrolled (unconstrained) system by
P t / D M 1 .q; t / Q.q; q;
a.q; q; P t/ : (11)
Also, assuming sufficient smoothness, we can differentiate the h equations in the

set (8) twice with respect to time (as we just did in our toy problem, see (5)), and
the (m h) equations in the set (9) once with respect to time, to obtain the relation
P t / qR D b.q; q;
A.q; q; P t/ ; (12)
where A is an m by n matrix of rank k. Each row of the matrix A corresponds to

one of the trajectory requirements in the sets (8) or (9). We permit the trajectory
requirements to be functionally dependent, though always consistent.
3 The Control Force QC
Having now laid out some of the underlying concepts relevant to the connections
between the problem of constrained motion and the problem of tracking control, let
us concentrate in this section on how one might determine the control force QC .
Before we embark on this, it might be worthwhile going back to our toy problem
and investigating if such a force QC indeed exists, so that the trajectory requirement
(1) is always satisfied, and if so, whether it can be uniquely found. That such a force
QC exists is obvious, because we know the equation of motion of a pendulum, and so
we know that a right-hand side for Eq. (3) exists so that the constraint (1) is exactly
satisfied for all time, given that the initial conditions satisfy this constraint. So there
most likely exists a control that is Lipschitz continuous, as we require in mechanics
so that the solution of (3) is unique and it concurs with practical observations of the
motions of a pendulum. Our next question is then, can QC be uniquely found?
Unfortunately, not! For the spherical pendulum, at each instant of time, we have
the following six unknowns: the three components of the 3-vector qR and the three
components of the 3-vector QC (see Eq. (3)). At each instant of time, starting with
a given state .q; q/P of the system, we have the three equations given by the set (3)
and an additional equation of constraint (1) (or alternately Eq. (6))—a total of 4
equations. The number of unknowns exceeds the number of equations by two, and
hence, at each instant of time, the problem of finding the unknowns (accelerations
and control forces) in the system is underdetermined! To get them uniquely we
would need to have two more independent equations. Moving to our dual vision of
the problem as one of trajectory control, there must then be an infinity of control
forces (controllers) QC that can exactly track the trajectory expressed by Eq. (1)!
However, the equation of motion of a spherical pendulum, which satisfies the
constraint (trajectory requirement), is unique—hence QC is unique—and its motion
pretty well agrees with what is in fact physically observed. So clearly, Nature must
then be picking the constraint force (control force) QC in such a manner so as to
satisfy some additional criterion—one which somehow yields the (additional) two
missing equations in our toy problem and yields a unique answer for the control
force! The next subsection discusses this.
3.1 D’Alembert’s Principle, Gauss’s Principle,

and the Cost Function
Flipping back to our understanding of constrained motion, we may then ask, how
does Nature pick the constraint force QC so that the motion of our spherical
pendulum matches our physical observations? This is a problem that was first
attacked by d’Alembert and later on, more generally, by Lagrange [10]. Lagrange
came up with the precise statement of what is today called d’Alembert’s principle
or prescription. D’Alembert’s prescription is as follows.
The constraint force QC is such that for all vectors v(t) ¤ 0 that satisfy the
relation A v D 0, Nature seems to require that vT QC D 0.
The nonzero vectors v that satisfy the relation Av D 0 are called virtual displace-
ments, and the quantity WC D vT QC is referred to as the total work done by the
forces of constraint under virtual displacements. And this prescription, somewhat
miraculously—for any general mechanical system—generates the correct number
of additional equations so that the constraint force QC in Eq. (10) can be uniquely
found at each instant of time!
To see how this works for our spherical pendulum, observe that the rank of our
matrix A in Eq. (6) is 1, and so the null space of this 1 by 3 matrix is 2. Thus at each
time there are two linearly independent 3-vectors v1 and v2 that satisfy the relation
Av D 0 which we can find. D’Alembert’s prescription then requires that v1 T QC D 0
and v2 T QC D 0. These two additional equations used with the four equations (the
three equations in set (3) and Eq. (6)) that we had previously yield the six equations
needed for finding the six unknowns—qR and QC —at each instant of time.
What is more astonishing—and indeed a tribute to the genius of Lagrange—is
that d’Alembert’s prescription yields the constraint force QC which when used in
Eq. (10) yields the motion, q(t), of the mechanical system that is fairly close, in
numerous situations, to what is actually observed in the physical world, hence its
enormous value in modeling physical systems.
To summarize, the equations of motion of the unconstrained system and the con-
straints cannot in general determine the constraint force QC uniquely. D’Alembert’s
principle generates additional equations (exactly the right number) to give us a
unique QC at each instant of time, which causes the constrained system to move
in a manner that is in concert with physical observations. Now it turns out that this
prescription of d’Alembert regarding the constraint force QC is exactly the same as
the following condition on the constraint (control) force QC from the dual viewpoint
[24]. This condition, called Gauss’s principle [5], is the following:
From all those control (constraint) forces QC that can exactly satisfy the
trajectory requirements (8) and (9), Nature chooses that control force QC that
minimizes the control cost
T 2
J.t / D QC .q; q; P t / D QC M 1
P t / M 1 .q; t / QC .q; q; (13)
at each instant of time. As seen from (13), J(t) is simply the square of the
weighted L2 norm of the control force, QC . So we see that d’Alembert’s prescription
in mechanics—a prescription that causes mathematical models of constrained
mechanical system to suitably predict the physically observed motions of these
systems—has a dual that says that Nature appears to be constantly solving an
optimal control problem, minimizing the cost function J(t) givenRin (13). But unlike
most control engineers today, who would prefer to minimize T0 J(t)dt, where T
is some final time over which the control is executed, Nature seems to do this
minimization at each instant of time. Also, the so-called weighting matrix that
she uses in the cost function is M 1 . This is indeed clever! For example, imagine
a multi-body system, with several masses, that is described by Eq. (7). Say we
want to control this system so that it satisfies some given trajectory requirements
given by relations (8) and (9). Realizing that the larger masses require larger forces
to be exerted on them to cause them to move, Nature attempts to satisfy these
requirements (constraints) on this multi-body system, by being in favor of applying
forces to the smaller masses—hence, the weighting by the matrix M 1 . It is only
recently that control engineers in robotics have realized, mainly by trial and error,
that this weighting matrix works well in most robotic applications and leads to
robust control designs [15].
Having underscored the duality between the problem of constrained motion and
the problem of trajectory tracking, we find then that insights from mechanics do
shed some light on control theory. But while so far we have only been talking about
the properties of the constraint force QC , we have not yet answered the question:
what is it? Can one find it explicitly, in closed form, for a general unconstrained
mechanical system?
3.2 Closed-Form Solution to the Optimal Tracking Control

Problem for Nonlinear Nonautonomous Mechanical
Systems Using the Theory of Constrained Motion
The problem of finding the constraint force QC that Nature uses has a long and
varied history. The problem was first formulated by Lagrange [10] and has been
attempted by numerous scientists and engineers [1, 4–6, 8, 16]. A simple expression
for the explicit form of the control force was obtained in 1992, and it is given by [23]
C
QC D M 1=2 AM 1=2 .Aa b/ : (14)
Here, XC denotes the Moore-Penrose inverse of the matrix X [11, 14]. The
equation of motion of the constrained system, which may be thought of as the
fundamental equation of mechanics, can thus be explicitly written in extensio, using
relation (10), as
C
M .q; t / qR D Q.q; q; P t / M 1=2 .q; t /
P t / M 1=2 .q; t / A.q; q;
P t / a.q; q;
ŒA.q; q; P t / b.q; q;
P t / : (15)
What now might be gleaned from a controls point of view from relation (15)?
P t / (see Eq. (11)) is the acceleration of the uncontrolled
First, we observe that a.q; q;
(unconstrained) system. However, to track the given trajectory described by the set
of Eqs. (8) and (9), the acceleration of the system needs to satisfy the trajectory
requirement (12). Hence the extent to which the acceleration, a, of the uncontrolled
system does not satisfy this trajectory requirement is simply
P t / WD ŒA.q; q;
e.q; q; P t / a.q; q;
P t / b.q; q;
P t / : (16)
This is in fact the error in the satisfaction of the trajectory constraint at time
t by the acceleration (at that time) of the uncontrolled system. And so Eq. (15)
says that this error signal is fed back to the system (7), just the way a modern-day
control engineer might want to do negative feedback control! Second, we observe
that Nature seems to choose a control gain matrix whose elements are, in general,
highly nonlinear functions of q; q;P and t. It is given explicitly by
C
K.q; q; P t / M 1=2 .q; t / :
P t / WD M 1=2 .q; t / A.q; q; (17)
Thus the control methodology used by Nature, so that the uncontrolled system
(7) exactly tracks the trajectory requirements stated in sets (8) and (9), can be
encapsulated by the relation
P t / K.q; q;
P t / e.q; q;
P t/ ; (18)
where K is the gain matrix and e is the error signal. Lastly, we point out that
Nature appears to use an error signal for its feedback control law that is related
to accelerations and not to displacements, nor to velocities, or to integrals of the
displacement, as is commonly done in control theory; recall PID control! She
appears to be doing acceleration feedback. The tracking controller represented by
Eq. (18) is not only optimal in that it minimizes the control cost J(t) given in
Eq. (13), but it yields exact tracking; for the set of equations (8) and (9) are the
integrals of motion of the nonlinear system described by Eq. (15) (or (18)). The
minimal control cost at each instant of time is explicitly given by
T C 2

J.t / D QC M 1 QC D AM 1=2 .Aa b/ : (19)
As mentioned before, the closed-form expression for the control force QC that
Nature uses when satisfying trajectory requirements is given by relation (14).
She gets this unique control force by minimizing the control cost J(t) given in
(13), which is simply the square of the weighted L2 norm of control force, QC .
Nature picks the weighting matrix to be the positive definite matrix M 1 (q, t). This
choice produces control forces (forces of constraint) that are in conformity with the
physically observed motions of constrained systems.
What if the control engineer wants to use a different weighting matrix in the cost
function? Namely, suppose he/she wants to minimize at each instant of time the
control cost
T 2
J.t / D QC .q; q; P t / D QC N ;
P t / N .q; t / QC .q; q; (20)
where N(q, t) is a positive definite matrix. Using our dual perspective, this may also
be thought of as a generalization of Gauss’s principle (in mechanics), wherein we
use a weighting matrix in our control cost minimization that may be different from
M 1 . It turns out that the unique control that minimizes this control cost is given
(instead of Eq. (14)) by [21]
h iC
QC DN .q; t /1=2 AC
N .Aab/ DN
1
M 1 AT A.MNM /1 AT .Aab/ ;
(21)
1 1=2
where AN D A.q; q; P t / M .q; t / N .q; t / . The last equality above follows from
the identity XC D XT (XXT )C . When N D M 1 , we get back Eq. (14), as expected.
There is one last remaining point that is worth mentioning. Recall that we had
assumed that the initial conditions of the controlled system satisfy the trajectory
requirements (8) and (9). What if the initial conditions do not lie on the so-called
manifold described by the trajectory requirements? If one is close to the trajectory
manifold, then instead of thinking of the trajectory requirements (8) and (9) as
P t / D 0, one could consider the trajectory requirement
i (q, t) D 0 and i .q; q;
as [20]
R C ˙ P C K D 0; and P D ƒ ; (22)
where and are h- and (m h)-vectors that contain the i ’s and j ’s,
respectively. The matrices ˙, K, and can be chosen so that the solutions,
and , to the equations (22) tend to zero asymptotically as t ! 1, so that
the constraints i D 0 and i D 0 are ultimately satisfied. Equations (22) lead to
trajectory requirements which can again be stated in the form of Eq. (12), and the
control force is again given explicitly by Eq. (21)! The parameters that are used
in the matrices ˙ , K, and will control the rate and nature of convergence of
the trajectories of the dynamical system towards the manifold prescribed by the
trajectory requirements, i (q, t) D 0 and i .q; q;P t / D 0.
To illustrate the nature of this control force, let us go back to our toy problem,
which we introduced in Sect. 1, of controlling a mass m in a time-varying gravity
field so that it lies on the surface ®(x, y, z, t) :D x2 (t) C y2 (t) C z2 (t) L2 D 0. The
explicit control force that causes the mass to lie on this manifold asymptotically is
given in the Appendix. We note that the control is nonlinear and no approximations
related to the nonlinear nature of the “plant” are made. No a priori assumptions
(such as a linear PD controller) are made about the controller either. Note that the
control minimizes the control cost given in Eq. (20) at each instant of time for a
user-specified weighting matrix N.
Flipping back to analytical dynamics, our closed-form equation given by (15)
for the constrained motion of the system (10) presupposes that d’Alembert’s
prescription is valid for every mechanical system. What if it isn’t? Before we take
such a contrary stance, we need to inquire whether systems for which d’Alembert’s
principle doesn’t hold do exist. Fortunately for us, it turns out that there are
many such systems, the ones most commonly exemplified are those that may have
sliding friction in them (for example, think of a mass sliding along a horizontal
surface), because now the constraint force QC does perform work under virtual
displacements [2]. Until recently, such systems have been a kind of eyesore for
analytical dynamicists, because it has been difficult to include them within the
general framework of Lagrangian mechanics. Constraint forces that do not obey
d’Alembert’s prescription are called nonideal, and often such systems are referred
to as systems with nonideal constraints. Let us now turn to such systems and find
out what insights they may provide, in our dual view, for the tracking control of
nonlinear, nonautonomous systems.
3.3 Mechanical Systems with Nonideal Constraints and the Set

of Controllers for Exact Trajectory Control
The difficulty of incorporating systems with nonideal constraints into the framework
of Lagrangian mechanics—though such systems are fairly commonplace in the
physical world—arises because of the following three main reasons:
1. We need to have the specification of constraints to be general enough so as to
encompass problems of practical utility.
2. The specification must, in order to comply with physical observations, yield the
accelerations of the constrained systems uniquely when using the math-ware of
analytical dynamics that has been developed over the last 250 years.
3. When the constraint forces do no work, we must obtain the formalisms/equations
that have thus far been obtained (e.g., Lagrange’s equations, Gibbs-Appell
equations, Poincare equations, Generalized Inverse equations, etc.).
It is for this reason that most texts and treatises on mechanics summarily dispatch
these systems beyond their boundaries, early on in their treatments of analytical
dynamics. For example, we find in Goldstein, at the beginning of his book, on page
17, the line [7]:
This [D’Alembert’s Principle] is no longer true for sliding friction and we must exclude
such systems from our Lagrangian formulation.
In Pars’s treatise on mechanics we again find on page 14, and we quote [13]:
There are in fact systems for which this [D’Alembert’s] Principle does not hold. But such
systems will not be considered in this treatise.
Put differently, the main problem is how to modify and extend d’Alembert’s
principle so that the three difficulties mentioned above can be overcome. One way of
doing this would be to extend d’Alembert’s prescription to say that at each instant
of time, the work done by the force of constraint is prescribed for the specific system
at hand. Such a principle would then state that [25]:
For any virtual displacement v.t / at time t; the work done by the force
(23)
of constraint W C WD vT QC is prescribed to be equal to vT C .q; q;
P t/ ;
P t / is prescribed by the mechanician for the given, specific

where the n-vector C .q; q;
system being modeled. The prescription of C can be done through experimentation,
and/or by analogy with other systems, or intuition, or otherwise. Since at any given
instant of time t, WC can be positive, negative, or zero; this allows the possibility
that energy can be fed into the system at the constraint, or it can be removed at
the constraint. We note that when C 0 (WC D 0) for all time t, this extension of
d’Alembert’s principle reverts to d’Alembert’s prescription.
For any sufficiently smooth C, one can find the explicit equation of motion for
such a constrained system that has nonideal constraints and satisfies exactly the
constraint requirements (8) and (9) (or alternately (12)). Dropping the arguments of
the various quantities, the equation is [25]

M qR D Q M 1=2 B C .Aa b/ C M 1=2 I B C B M 1=2 C WD Q C QC ;
(24)
where B.q; q; P t / M.q; t /1=2 , and, as before, XC denotes the Moore-

P t / D A.q; q;
Penrose inverse of the matrix X. We notice that the first two terms in the first equality
on the right-hand side of Eq. (24) are identical to those on the right-hand side
of Eq. (15), the nonideal nature of the constraint force having simply added an
additional term on the right-hand side, for any given prescribed smooth function
P t /. Furthermore, when C 0 the equation reverts to the proper equation of
C .q; q;
motion (15) when d’Alembert’s prescription is satisfied at all instants of time.
Using our dual view regarding the tracking control problem, what insights
does this equation, which originates from analytical dynamics, provide for the
constrained motion of a system with nonideal constraints? By choosing the Lips-
chitz continuous function C .q; q; P t / arbitrarily, Eq. (24) provides all the possible
Lipschitz continuous controllers [25] that can make the uncontrolled system (7)
exactly track the trajectory requirements specified by equations (8) and (9)!
The second and third members on the right-hand side in the first equality of
Eq. (24) are M-orthogonal, and so
2 2
J.t / D B C .Aa b/ C I B C B M 1=2 C : (25)
The addition of the second term on the right-hand side increases the cost from
its optimal value of kBC (Aa b)k2 (see Eq. (19)) to that now provided by (25). We
note that at those instants of time when C D 0, the control force becomes optimal
and minimizes J(t) D [QC ]T M 1 QC .
As before, more generally, when using the weighting matrix N instead of
M 1 , the explicit control that causes system (10) to exactly satisfy the trajectory
requirements (8) and (9) is given in closed form by [21]
1=2
QC D N .q; t /1=2 AC
N .Aa b/ C N
1=2
I AC
N AN M C; (26)
for arbitrary continuous functions C .q; q; P t /. When C 0, the control minimizes

the general control cost given in (20) at each instant of time. The equation of motion
of the controlled system is then
Table 2 Analogous concepts in analytical dynamics and control theory (detailed)

Analytical dynamics Control theory
Unconstrained system Uncontrolled system or plant
P t/ M .q; t / qR D Q.q; q; P t/
Constrained system Controlled system
P t / C QC .q; q;
P t/ M .q; t / qR D Q.q; q; P t / C QC .q; q;
P t/
Constraints Trajectory requirements
R C ˙ P C K D 0, i (q, t) D 0, i D 1, 2, : : : , h
P D ƒ P t / D 0, i D h C1, h C2, : : : , m.
i .q; q;
Gauss’s principle (GP) Control cost
T ZT
P t / M 1 .q; t / QC .q; q;
J.t / D QC .q; q; P t/ C T
Q .q; q; P t / M 1 .q; t / QC .q; q;
P t / dt
0
Constraint force with GP Control force or control
QC D M1/2 (AM 1/2 )C (Aa b)
Optimal at each instant of time Optimal over the interval of time [0,T]
ZT
C T
Generalized Gauss’s principle P t / N .q; t / QC .q; q;
Q .q; q; P t / dt ,
T
J.t / D QC .q; q;
P t / N .q; t / QC .q; q;
P t /, 0
where N > 0 where N > 0
Equations of motion for nonideal constraints Full set of continuous controllers that satisfy
M qR D Q N.q; t /1=2 AC
N .Aa b/ C
trajectory requirements for arbitrary
P t/
continuous C .q; q;
N 1=2 I AC
N AN M
1=2 .q;
C q; P t/
1=2
M qR D Q N 1=2 AC
N .Aa b/ C N
1=2
I AC
N AN M C WD Q C QC :
(27)
The second and third members in the first equality on the right-hand side are
N-orthogonal, and in the presence of the third member the control cost increases
and is given by
T 2 1=2 2
J.t / D QC NQC D AC
N
.Aa b/ C I AC
N AN M C : (28)
We can now expand Table 1 to expose the various analogous concepts that we
have developed (see Table 2).
4 Examples
We provide three examples in this section that utilize the connections that we have
developed between constrained motion and control of nonlinear systems. The first
deals with motion synchronization of three different highly nonlinear dynamical
systems, the second deals with the optimal design of Lyapunov stable controllers for
nonlinear systems, and the third deals with the energy control of nonlinear chain-
type systems.
Moment of inertia, Ioz = I3 Z

z j
y
(Nutation)
q Moment of inertia, Ioy = I
y
c.m. +
mg r
d(t) = d0 sin w t Y
O
y
x
Moment of inertia, Iox = I
d(t)
X j
(Precession) Ox' is parallel to Ox"
q
x"
x' Ox' is on OXY plane
Fig. 2 Symmetric gyroscope with vertical support excitation d(t) D d0 sin(!t)
4.1 Determination of Control for Synchronization

of Gyroscopes
Consider three gyroscopes mounted on a horizontal surface (that oscillates ver-

tically) with each gyroscope having different dynamical characteristics. Our aim
is to synchronize these three nonlinear systems so that two of the gyros—called
“slave gyros”—track the motion of the third gyro called the “master gyro”. While
synchronization of nonlinear systems wherein each system is a copy of the other
has been done before, here we look at the synchronization of different nonlinear
systems [3, 17, 18]. Boccaletti et al. [3] gives an excellent review of synchronization
of nonlinear systems.
A typical gyroscope (gyro) is shown in Fig. 2. We denote by m the mass of this
typical gyro, I1 D I2 is the principal equatorial moment of inertia through the center
of mass of the gyro, I :D I1 C mr2 , and I3 is the polar moment of inertia about its
symmetry axis. The point of support of the gyro is denoted by o, so that the moments
of inertia about the axes ox and oy each equal I. The quantity r denotes the distance
along the polar axis of the center of mass of the gyro from its point of support, and
d(t) D d0 sin !t is the time-varying amplitude of the vertical support motion that has
frequency ! [22].
Using the Euler angles
(nutation), (precession), and (spin) (see Fig. 2), the
Lagrangian for the system is given by
1 P2 1 2
LD I
C P 2 sin2
C I3 P C P cos
mr dP
P sin
mgr cos
: (29)
2 2
The dots in Eq. (29) refer to differentiation with respect to time, t. Since
and
are cyclic coordinates, the corresponding angular momenta p D I3 P C P cos
P 2
C p cos
are conserved. The angular velocities P and P can
and p D I sin
be eliminated by using the Routhian [7]

R
;
P ; t D L p P p ; p ;
p P p ; p ;
: (30)

The equation of motion, which is given by d @R
@R
D Fd , then reduces to
dt @
P @

p p cos
p p cos
I
R C mgr sin
mr sin
dR .t / D Fd ; (31)
I sin3
where Fd is the nonconservative force of damping, which we take here to be of

c
P b
linear-plus-cubic type, so that Fd D b e
P 3 . Were we to further assume that
p D p D p (which permits the gyro to be in the so-called “sleeping position”)
Eq. (31) can be further simplified to
.1 cos
/2

R D sin
sin !t ˛ 2 c P e
P 3 C ˇ sin
WD f
;
P ; t I P :

sin3
(32)
Under this assumption, Eq. (32) then is the differential equation that describes
the motion of the symmetric gyro, where we have denoted ˛ D p=I; c D b c=I ,
e Db e=I , ˇ D mgr/I, and D ! 2 mrd0 /I. The parameter set P D f˛, ˇ, c, e, , !g in
Eq. (32) specifies the physical characteristics of the gyro and the harmonic vertical
motion of the base on which it is supported. It may be pointed out that no assumption
on the magnitude of the vertical displacement d0 of the base has been made in
arriving at this equation. We note in passing that no singularity arises in Eq. (32)
due to the sin
term in the denominator. For various values of the parameters in the
set P, a gyro, such as the one described above, can have motions that span the entire
spectrum from regular periodic motion to period doubling and chaotic motion.
Consider three such gyros, each described by Eq. (32), that need to be synchro-
nized so that two of them follow the motion of, say, the first (master) gyro. We
shall take these three gyro systems to be different from each other, described by
the parameter sets Pi D f˛ i , ˇ i , ci , ei , i , ! i g, i Dn1, 2, 3,oand their dynamics will be

investigated for the initial condition sets ICi D
i0 ;
Pi0 ; i D 1, 2, 3, given by
n o
P1 D f10; 1; 0:5; 0:03; 35:8; 2:05g I IC1 D
10 D 0:5;
P10 D 1 I (33)
n o
P2 D f10; 1; 0:5; 0:05; 35:5; 2g I IC2 D
20 D 0:5;
P20 D 1 I and (34)
n o
P3 D f10:5; 1; 0:5; 0:04; 38:5; 2:1g I IC3 D
30 D 1;
P30 D 0:5 : (35)
The Lyapunov exponents for each of the three uncontrolled gyros that have
parameters described by the sets (33–35) are computed over a time span of 1,000 s
by using the method described in Udwadia and von Bremen [26]. The integration for
determining these exponents is performed using MATLAB’s ode45 using a relative
error tolerance of 10 9 and an absolute error tolerance of 10 13 . The Lyapunov
exponent sets, li , of the three different gyro systems are computed to be l1 f0.211,
0.896, 0g, l2 f0.216, 1.001, 0g, and l3 f0.208, 0.936, 0g, respectively.
The positive value of the largest Lyapunov exponent in each set indicates that the
motions are chaotic for each of these gyros. Furthermore, the chaotic attractors for
each system are different.
Figure 3 shows plots of (
i ;
Pi /, i D 1, 2, 3, for 50 t 100 for the three
uncontrolled gyros along with a figure (lower right corner) in which all three plots
are superposed. The integration of the equations of motion throughout the following
study is carried out using MATLAB’s ode45 with a relative error tolerance of 10 9
and an absolute error tolerance of 10 12 . The differences in the responses between
the three uncontrolled gyros,
hij .t / D
i .t /
j .t /; i ¤ j; (36)
are shown in Fig. 4. The three gyros can now be thought of as comprising one single,
uncontrolled dynamical system. The uncontrolled equation of motion of this system
is given by
2R 3 2 3

1 f1 .
1 ;
P1 ; t I P1 /
a WD 4
R2 5 D 4 f2 .
2 ;
P2 ; t I P2 / 5 ; (37)

R3 f3 .
3 ;
P3 ; t I P3 /
where the function fi (

i ,
Pi ; t; Pi ) corresponds to the ith gyro and is obtained from
Eq. (32) by using the parameter set Pi in it. We denote, as before, the uncontrolled
acceleration of this dynamical system by a. Our task is to find the control force so
that the last two gyros (i D 2, 3) track the motions of the master (i D 1) gyro. In order
to do this, we apply a tracking control force QC so that the controlled dynamical
system is described by the equations
4 4
θdot2 for Second Gyro

θdot1 for First Gyro
2 2
0 0
−2 −2
−4 −4
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
θ1 for First Gyro θ2 for Second Gyro
4 4
θdot3 for Third Gyro
θdot1, θdot2, θdot3

2 2
0 0
−2 −2
−4 −4
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
θ3 for Third Gyro θ1,θ2,θ3 for all Three Gyros

Fig. 3
i ;
Pi plots showing the dynamics of the three uncontrolled gyros for 50 t 100. The
lower right corner shows these plots superposed on one another; the first gyro is shown with a solid
line, the second with a dashed line, and the third with a dash-dot line
2 3
2R 3 f1
1 ;
P1 ; t I P1 2 C3

1 6 7 Q1
6 7
qR WD 4
R2 5 D 6 f2
2 ;
P2 ; t I P2 7 C 4 Q2C 5 : (38)
4 5

R3 f3
3 ;
P3 ; t I P3 Q3C
The trajectory requirements (constraints) for synchronization are given by
1: 1 .t / WD
1 .t /
2 .t / D 0:
2: 2 .t / WD
1 .t /
3 .t / D 0: (39)
3: The control force; Q1C ; on the first gyro is zero for all time:
The last requirement above causes the motions of the master gyro to remain
undisturbed, while the first two requirements call for the motions of the two “slave”
gyros to track those of the master (first) gyro. Notice that the gyros do not have the
1.5
1
(θ1-θ2), (θ1-θ3), (θ2-θ3)
0.5
−0.5
−1
−1.5
−2
−2.5
0 10 20 30 40 50 60
Time
Fig. 4 The differences in the responses between the three uncontrolled gyros shown for a duration
of 60 s. h12 (t) D
1 (t)
2 (t) is shown by the solid line, h13 (t) D
1 (t)
3 (t) is shown by the
dashed line, and h23 (t) D
2 (t)
3 (t) is shown by the dash-dot line
same initial conditions. So we shall use the first of the two equations of (22) with
− D −i D 1, D i D 2, i D 1, 2. The abovementioned constraint set (39) can then be
expressed as

R1
R2 D &
P1
P2 .
1
2 / ; (40)

R1
R3 D &
P1
P3 .
1
3 / ; and (41)

R1 D f1
1 ;
P1 ; t I P1 : (42)
Expressing these trajectory requirements in the form of Eq. (12), we have

2 3
1 1 0
A D 4 1 0 1 5; and; (43)
1 0 0
a
θ1, θ2, θ3 for Individual Gyros 2
−1
−2
0 2 4 6 8 10 12 14 16 18 20
Time
b
θ1, θ2, θ3 for Synchronized Gyros
−1
−2
0 2 4 6 8 10 12 14 16 18 20
Time
Fig. 5 (a) First 20 s of the response of the uncontrolled gyros with the master gyro shown with
a solid line, the second gyro shown with a dashed line, and the third gyro shown with a dash-dot
line; (b) Synchronization of the gyros by the application of control forces showing the slave gyros
following the master (solid line) as required by the constraint Eqs. (40) and (41) with − D 1 and
D 2. The motion of the master gyro remains unchanged, as required by (42)
h iT
bD &
P1
P2 .
1
2 / &
P1
P3 .
1
3 / f1
1 ;
P1 ; t I P1
(44)
The tracking control forces to be applied on this dynamical system are explicitly
obtained, at each instant of time t, by simply using relation (14). This tracking
control force minimizes the cost given in (13) at each instant of time. Figure 5a
shows the time responses for the first 20 s of the three uncontrolled gyros, and
Fig. 5b shows their controlled (synchronized) response, where the latter two gyros
(i D 2, 3) are now slaved to the master (i D 1) gyro. We observe that the error
between the responses gradually
reduces to zero as required by Eqs. (40) and (41).
P
The plots in the
i ;
i -plane, i D 1, 2, 3, superposed on one another for all
three gyros are shown in Fig. 6, indicating synchronization of the two slave gyros
with the chaotic motion of the master gyro. The plots are made using the response
of each of the gyros over a 50 s interval of time, starting at 50 s. We note that in this
figure there are three separate plots that are superimposed on top of one another. As
seen, the plots fall exactly on top of each other. The differences in the responses,
3
θdot1, θdot2, θdot3, for Three Synchronized Gyros
−1
−2
−3
−4
−1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
θ1,θ2,θ3 for all Three Gyros

Fig. 6 Superimposed plots of
i ;
Pi , i D 1, 2, 3, of the three synchronized gyros for
50 t 100. The master gyro is a chaotic system and its Lyapunov exponents [26] are l1 f0.211,
0.896, 0g. Each of the gyros executes the entire motion shown in the plot
hij (t) D
i (t)
j (t), 50 t 100, between the motions of the three synchronized
gyros are shown in Fig. 7. We notice that this error soon becomes of the same
order of magnitude as the error tolerance (10 12 ) used to numerically integrate
the equations. The exponential convergence of hij (t) towards zero as demanded by
relations (40) and (41) is obvious.
4.2 Stable Controller Design
When trying to find a controller for a nonlinear nonautonomous system, one often
postulates a controller based on experience, and/or intuition, and/or heuristics, and
then analyzes its stability. Usually, Lyapunov’s second method is used, and stability
is checked by searching for a suitable Lyapunov function V and ensuring that its
time derivative is nonpositive along the dynamical trajectory [9]. Though there are
some standard methods that one can get guidance from in the search for a suitable
Lyapunov function, when handling complex, nonlinear, and high-dimensional
dynamical systems, this can become a difficult and time-consuming process, which
may at times not be fruitful. When one is unable to find such a function, the stability
of the postulated control is left uncertain.
x 10−11
2
h12
h13
h23
1.5
0.5
hi j
−0.5
−1
50 55 60 65 70 75 80 85 90 95 100
Time
Fig. 7 h12 (t) D

1 (t)
2 (t) (solid line), h13 (t) D
1 (t)
3 (t) (dashed line), h23 (t) D
2 (t)
3 (t) (dash-dot line) for 50 t 100. Note the vertical scale, which indicates that the error in
synchronization is of the order of the numerical integration error tolerance, 10 12
Using our knowledge of the connection between constrained motion and control,
we could, so to speak, turn this problem on its head by first choosing a candidate
Lyapunov function, V—namely, a positive definite function—and placing the
constraint on the dynamical system that its derivative, VP , is nonpositive, thereby
making the candidate function a Lyapunov function for the controlled system, thus
ensuring stability. In fact, we can do this while simultaneously minimizing the
control cost.
Consider the nonlinear, nonautonomous system (plant)
M qR D Kq C Lq .3/ C H WD Q.q; q;

P t/ ; (45)
where
" #
.tC1/
m1 .tC2/ 0 k1 k1
M.t / D .tC3/ ; KD ;
0 m2 .tC2/ k1 k 1 C k2
(46)

l1 0 c1 qP 1
LD ; H D q1 q2 ;
0 l2 c2 qP 2
and the 2-vector q(3) D [q31 q32 ]T . Our aim is to find a stable controller for this
dynamical system that brings it to the fixed point qi D qP i D 0. We choose the
Lyapunov candidate (positive definite) function
1 1
P D
V .q; q/ a1 q T q C a2 qP T qP C a12 qP T q; (47)
2 2
2
with the constants a1 ; a2 > 0; a1 a2 > a12 : In order to make this function a Lyapunov
function for the controlled dynamical system, we place the constraint that
P
d V .q; q/ @V @V
VP D WD qR C qP D ˛V; (48)
dt @qP @q
which can be rewritten in the form (12) (i.e., AqR D b) with
P D a2 qP T C a12 q T D Œa2 qP 1 C a12 q1 ; a2 qP 2 C a12 q2 ; and

A.q; q/ (49)
1 1
P D ˛
b.q; q/ a1 q T q C a2 qP T qP C a12 qP T q a1 q T qP a12 qP T q:
P (50)
2 2
It is important that the constraint (12) be consistent, and when A D 0, we must

have b D 0. We now know from (14) that the control force given by
C
QC D M 1=2 AM 1=2 .Aa b/ (51)
will cause the controlled system to not only exactly satisfy relation (48) but also
simultaneously minimize the control cost J(t) D [QC ]T M 1 QC at each instant of
time. The equation of motion of the controlled system is then
M qR D Kq C Lq .3/ C H C QC (52)
where QC is given in (51). Using the numerical values m1 D 1, m2 D 2,

k1 D 100, k2 D 100, l1 D l2 D 4, and c1 D c2 D 1, in the relations shown in Eq. (46),
and the initial conditions q(0) D [1, 2]T and q.0/
P D Œ2; 3T , the uncontrolled
system given by Eq. (45) for these parameter values is unstable. We shall use the
candidate Lyapunov function V given in Eq. (47) with a1 D 1, a2 D 4, and a12 D 1,
and ˛ D 1/2. These parameter values ensure consistency of the constraint Eq. (12).
We obtain a simulation in the MATLAB environment of the controlled system
given by Eq. (52) (using QC explicitly obtained from Eq. (51)). Numerical inte-
gration of the ode’s is done using ode15s with a relative error tolerance of 10 8
and an absolute error tolerance of 10 12 . Figures 8 and 9 show the displacement
and velocity response, respectively, of the controlled system as a function of time
showing its asymptotic convergence to the fixed point q D qP D 0. This example
illustrates the power that can be invoked through the recognition of the connection
between constrained motion and the corresponding control problem.
Fig. 8 Displacement response of the controlled nonlinear system
Fig. 9 Velocity response of the controlled nonlinear system
4.3 Energy Control of a 3-DOF Toda Oscillator
Energy control of nonlinear mechanical systems has become important nowadays

and various energy harvesting schemes are being developed. We consider here the
problem of energy control of a highly nonlinear mechanical system by using the so-
called fundamental equation of mechanics to obtain the explicit nonlinear control
force required to achieve the desired energy control.
Fig. 10 A 3-DOF fixed-fixed Toda chain
We consider a 3-DOF fixed-fixed Toda chain [19] as shown in Fig. 10. The mass,
displacement, and velocity of the ith mass (i D 1, 2, 3) in the chain are described by
mi , qi , and qP i , respectively. Given any nonzero initial energy state, H0 of the chain,
our aim is to stabilize the chain at a different nonzero desired energy level, H* . And
to achieve this, control can be applied to one or more of these three masses. In the
present example, we control the energy of the chain by actuating the first mass, m1 ,
alone (see Fig. 10). We shall impose the requirement that the energy of the system
be increased to the desired value H* as a constraint on the mechanical system, and
the constraint force that will cause this constraint to be satisfied will then be the
requisite control force that would need to be applied to mass m1 . We begin with a
description of the Toda potential.
(i) Toda Potential and Spring Force: The expression for the nonlinear potential of
the Toda spring [19] is given by
a bq a
u.q/ D e aq ; a > 0; b > 0 (53)
b b
whereas its exponential spring force Fs (q) can be derived from its potential as
@u.q/
Fs .q/ D Frestori ng .q/ D D a eb q 1 : (54)
@q
A plot of the Toda spring potential and the Toda spring force is shown in
Figs. 11 and 12, respectively. For sufficiently small displacement, the spring
force is approximately linear. However, the nonlinearity of the force gains
prominence as the displacement increases. As can be inferred from Fig. 12,
a larger force is required to stretch the spring by a unit distance than is required
to compress it. Hence, the Toda chain considered possesses spring elements
that are stronger in tension than in compression. Such systems arise frequently
in structural subsystems such as the stringers in suspension bridges.
(ii) Unconstrained System: Consider the 3-DOF fixed-fixed Toda chain as shown
in Fig. 10. The total energy of the chain can be written down as
Fig. 11 Toda spring potential
Fig. 12 Toda spring force
X 3
X3
1 ai bi .qi C1 qi / ai
P D
H .q; q/ mi qP i C
2
e ai .qiC1 qi / ;
iD1
2 iD0
bi bi
(55)
where qo q4 0 describe the boundary conditions of the fixed-fixed chain.

The equations of motion of the unconstrained (uncontrolled) system can be
written down in matrix form as M qR D Q or more explicitly as
Fig. 13 Time history of displacements for the unconstrained system (top) and constrained system
(bottom)
2 32 3 2 3
m1 0 0 qR 1 a1 e b1 .q2 q1 / 1 a0 e b0 .q1 / 1
4 0 m2 0 5 4 qR 2 5 D 4 a2 e b2 .q3 q2 / 1 a1 e b1 .q2 q1 / 1 5 :

0 0 m3 qR 3 a3 e b3 .q3 / 1 a2 e b2 .q3 q2 / 1
(56)
We take, for example, the initial conditions of this Toda chain to be
q1 .0/ D 1; qP 1 .0/ D 2; q2 .0/ D 2; qP 2 .0/ D 0; q3 .0/ D 1; qP 3 .0/ D 2: (57)
Figure 10 shows the parameter values of the masses (mi , i D 1, 2, 3) used

as well as the parameter values ai , bi , i D 0, 1, 2, 3 that characterize the four
different Toda springs. Using these parameter values and the initial conditions
given in Eq. (57), the unconstrained equations of motion given in Eq. (56) can
now be numerically integrated. We note that for all the simulations presented in
this subsection, the equations of motion have been integrated using the “ode45”
scheme in the MATLAB environment with a relative integration error tolerance
of 10 10 and an absolute error tolerance of 10 13 . Figure 13 (top) shows a plot
of the displacements of the three masses from t D 0 to t D 10 time units for the
unconstrained (uncontrolled) system.
The unconstrained Toda chain is a conservative system and the energy,
being an integral of motion, remains constant throughout the duration of the
Fig. 14 Time history of energy of the 3-DOF Toda chain
simulation as can be inferred from Fig. 14. For the parameter values chosen,
the energy level of the chain is H0 D14.22 units. Our aim is to increase the
energy of the chain to a new value.
(iii) Constraints: We shall assume that we want the nonlinear Toda chain described
by Eq. (56) (with the parameter values as shown in Fig. 10) to have an energy
level H* D 100 units by controlling only mass m1 . In order to achieve this
control objective, we impose the following two different types of constraints
on the unconstrained system. The first deals with our objective to change the
energy of the system to its desired value, H* ; the second deals with the fact that
we want to achieve this by actuating just a single mass from amongst the three
masses in the chain, namely, only mass m1 (see Fig. 10).
1. Energy Control Constraint: The energy control constraint is given by
d
P H C ˇ H .q; q/
H .q; q/ P H D 0; (58)
dt
where ˇ > 0. The solution to this differential equation shows that as t ! 1,

P ! H . Notice that this constraint allows the 3-DOF Toda chain to
H .q; q/
be started from any arbitrary initial energy state H0 (see Eq. 22) so that it
reaches its desired energy state, H* , as t ! 1.
2. No Control Constraints: Since no control force is to be applied to masses m2
and m3 of the Toda chain, the second and third equations in the equation
set (56) must remain unchanged in the controlled system. Therefore, the
unconstrained equations of motion of masses m2 and m3 are themselves
the constraints that guarantee that no control is applied to either of these
two masses! Thus, in addition to the energy constraint given by Eq. (58),
the unconstrained system (Eq. 56) is also subjected to the following two
constraints.

b .q q /
m2 0 qR 2 a2 e 2 3 2 1 a1 e b1 .q2 q1 / 1
D (59)
0 m3 qR 3 a3 e b3 .q3 / 1 a2 e b2 .q3 q2 / 1
When this set of constraints (Eqs. (58) and (59)) are expressed in the
general form of Eq. (12), we obtain AqR D b or more explicitly
2 32 3 2 /
3
m1 qP1 m2 qP2 m3 qP3 qR1 qP T Q ˇ .H H
6 7 6 7 6 a e b2 .q3 q2 / 1 a e b1 .q2 q1 / 1 7
4 0 m2 0 5 4 qR2 5 D 6
4 2
1

7
5:
0 0 m3 qR3 a3 e b3 .q3 / 1 a2 e b2 .q3 q2 / 1
(60)
(iv) Explicit Control Force: With the matrices M, Q, A, and b at our disposal, the
control force QC can be calculated using Eq. (14) and is given by
2 3
o .H H / m1 qP 1
P D4
QC .q; q/ 0 5; (61)
0
where the value of ˇ D o m1 qP 12 has been chosen to avoid any singularities

in the control force, which might arise when the actuated mass m1 has zero
velocity. In the present example for illustration, the positive constant o has
been chosen to be 0.03. The control force obtained (Eq. 61) is optimal and
it minimizes the control cost given by Eq. (20) at each instant of time, with
N D M 1 . Notice from Eq. (61) that the control force acting on the first mass
appears to make it move like a self-excited oscillator!
(v) Constrained System: The equations of motion of the constrained (controlled)
Toda chain can now be written down using Eq. (10) where M and Q are given
by Eq. (56), and QC is given by Eq. (61). A plot of the displacements of the
three masses of the controlled system (using the parameters shown in Fig. 10)
is shown in Fig. 13 (bottom) from t D 0 to t D 10 time units. A plot of the time
history of the energy is depicted in Fig. 14 for constrained system. As can be
inferred from the figure, the application of the control force has resulted in an
increase of the energy of the 3-DOF Toda chain from an initial energy level of
H0 D 14.22 units to the desired energy level of H* D 100 units. Figure 15 shows
a plot of the time history of the nonlinear control force acting on the first mass
to achieve the desired transition. Once the desired energy level is attained, the
control force automatically becomes zero, and we make use of the conservative
nature of the chain to remain at the desired energy level for all future time.
It can be shown with some effort that the nonhomogeneous Toda chain that we
have considered is controllable using control on just mass m1 in the sense that the
Fig. 15 Time history of control forces acting on the 3-DOF Toda chain
system can be “moved” from any arbitrary energy state H0 ¤ 0 to any other energy
state H* ¤ 0 using the control described in Eq. (61). We don’t prove that here, since
it will take us too far afield from the central theme of this paper.
5 Conclusions and Open Problems
In this paper we have tried to establish a connection between the problem of

constrained motion and the problem of trajectory tracking of nonlinear mechanical
systems. This is its primary goal. The solution to the problem of constrained motion
when d’Alembert’s principle is valid, we see, provides the closed-form controller
that will exactly cause the uncontrolled nonlinear mechanical system to track the
trajectory requirements. No approximations and/or linearizations are made, and the
control minimizes the control cost J(t) D [QC ]T M 1 QC at each instant of time. This
cost is the norm of the control force weighted by the inverse of the mass matrix
of the mechanical system. For a general weighting matrix N, one can obtain the
explicit control force given in Eq. (21). The simplicity of the controller allows for
its application in real time for the control of complex mechanical systems.
When d’Alembert’s principle is not valid, the constrained equations of motion
provide, in closed form, the entire set of continuous controllers that would cause
the trajectory requirements to be satisfied. Furthermore, when the weighting matrix
in the control cost is a user-specified positive definite matrix N other than the
matrix M 1 (which Nature uses) we obtain the explicit exact tracking control for
the nonlinear system that minimizes this control cost at each instant of time. This is
then further extended to get the full set of controllers that can provide exact tracking.
This duality between the constrained motion problem and the tracking control
problem of nonlinear nonautonomous systems appears to be just the beginning of
our explorations into the Janus-faced field of mechanics as stated in the prescient
definition that Newton gave on the subject—the study of the motions of systems
subjected to known forces and the study of the forces required to be applied to
systems to generate known motions.
Examples provided in this paper deal with (i) the motion synchronization of
dissimilar chaotic gyroscopes, (ii) the design of optimal Lyapunov stable controllers
for nonlinear, nonautonomous systems, and (iii) the energy control of a nonho-
mogeneous Toda chain. These diverse examples exploit the connection between
control theory and constrained motion theory; the closed-form control forces that
are obtained are optimal and minimize the control cost at each instant of time;
they provide exact trajectory tracking. The examples show the ease, simplicity,
and efficacy with which control can be achieved. The full nonlinear “plant” is
considered, without the need for any linearizations/approximations.
As mentioned earlier, the developments outlined herein form just the beginnings
of a new path to our understanding of the synthesis of analytical dynamics
and control. There are numerous open questions that remain unanswered, such
as robustness of control, extensions to multi-body dynamics and the dynamics
of continua, and applications to robotics, space systems, and fluid mechanical
systems.
Appendix
We shall call the column 3-vector q D [x, y, z]T . The plant we want to control
corresponds to the unconstrained system (see Table 2) whose equation of motion
is given in Eq. (2) as
2 3 2 3 2 3
m 0 0 xR 0
R / WD 4 0 m 0 5 4 yR 5 D 4 mg.x; y; z; t / 5 WD Q
M q.t (62)
0 0 m zR 0
The trajectory requirement (constraint) is
'.x; y; z; t / WD x 2 .t / C y 2 .t / C z2 .t / L2 D q.t /T q.t / L2 D 0; (63)
and since we may not start on this manifold initially, we consider instead the
constraint
'R C c 'P C k' D 0; c > 0; k > 0 (64)

whose solution as t ! 1 is ® D 0. Thus we get asymptotic convergence to the

trajectory requirement given in Eq. (63). Differentiating the constraint (63), Eq. (64)
can be rewritten as
2 3
xR
AqR D Œx y z yR 5 D qP T qP c qP T q .k=2/ q T q L2 WD b
4 (65)
zR

so that A D [x y z] and b D qP T qP c qP T q k q T q L2 =2. The control force QC
that minimizes at each instant of time the control cost
T
J.t / D QC NQC (66)
where N is a user-specified positive definite 3 by 3 matrix is given by Eq. (21) as

23
x=m h iC ˚
QC D N 1 4 y=m 5 A.MNM /1 AT gyCqP T qCc
P qP T qC k2 q T q L2 /g
z=m
2 3
x ˚ T
mN 1
D AN 1 AT 4 y 5 gy C qP T qP C c qP T q C k
q q L2 :
. / 2
z
(67)
References
1. Appell, P.: Su rune Forme Generale des Equations de la Dynamique. C. R. Acad. Sci. Paris
129, 459–460 (1899)
2. Awrejcewicz, J., Olejnik, P.: Analysis of dynamic systems with various friction laws. Appl.
Mech. Rev. Trans. ASME 58(6), 389–411 (2005)
3. Boccaletti, S., Kurths, J., Osipov, G., Valladares, D., Zhou, C.: The synchronization of chaotic
oscillators. Phys. Rep. 366, 1–101 (2002)
4. Dirac, P.A.M.: Lecture on Quantum Mechanics. Yeshiva University Press, New York (1964)
5. Gauss, C.F.: Uber ein neues allgemeines grundgesetz der mechanik. J. Reine Angew. Mathe-
matik 4, 232–235 (1829)
6. Gibbs, J.W.: On the fundamental formulae of dynamics. Am. J. Math. 2, 49–64 (1879)
7. Goldstein, H.: Classical Mechanics. Addison-Wesley, New York (1976)
8. Jacobi, C.G.J.: Vorlesung uber Dynamik. G. Reimer, Berlin (1884)
9. Khalil, H.K.: Nonlinear Systems. Prentice Hall, New Jersey (2002)
10. Lagrange, JL.: Mechanique Analytique. Mme Ve Coureier, Paris, France (1811)
11. Moore, E.H.: On the reciprocal of the general algebraic matrix. Bull. Am. Math. Soc. 26,
394–395 (1920)
12. Newton, I.: Mathematical Principles of Natural Philosophy, Vol. 1, Motte’s translation of
F. Cajori, University of California Press, Berkeley (1962)
13. Pars, L.A.: A Treatise on Analytical Dynamics. Oxbow Press, Woodbridge (1972)
14. Penrose, R.: A generalized inverse of matrices. Proc. Cambridge Phil. Soc. 51, 406–413 (1955)
15. Peters, J., Mistry, M., Udwadia, F.E., Nakanishi, J., Schaal, S.: A Unifying framework for robot
control with redundant DOFs. Auton. Rob. 24, 1–12 (2008)
16. Poincare, H.: Su une Forme Nouvelle des Equations de la Mechanique. C. R. Acad. Sci. Paris
132, 369–371 (1901)
17. Pyras, K.: Weak and strong synchronization of chaos. Phys. Rev. E 54, 4508–4511 (1996)
18. Rulkov, N., Sushchik, M., Tsimring, L., Abarbanel, H.: Generalized synchronization of chaos
in directionally coupled chaotic systems. Phys. Rev. 51, 980–994 (1995)
19. Toda, M.: Theory of Nonlinear Lattices. Springer, New York (1989)
20. Udwadia, F.E.: A New perspective on the tracking control of nonlinear structural and
mechanical systems. Proc. R. Soc. Lond. Ser. A 459, 1783–1800 (2003)
21. Udwadia, F.E.: Optimal tracking control of nonlinear dynamical systems. Proc. R. Soc. Lond.
Ser. A 464, 2341–2363 (2008)
22. Udwadia, F.E., Han, B.: Synchronization of multiple chaotic gyroscopes using the fundamental
equation of mechanics. J. Appl. Mech. 2011, 75(2), 021011-1–021011-10 (2008)
23. Udwadia, F.E., Kalaba, R.E.: A new perspective on constrained motion. Proc. R. Soc. Lond.
Ser. A 439, 407–410 (1992)
24. Udwadia, F.E., Kalaba, R.E.: Analytical Dynamics: A New Approach. Cambridge University
Press, Cambridge (1996)
25. Udwadia, F.E., Kalaba, R.E.: What is the general form of the explicit equations of motion for
constrained mechanical systems? J. Appl. Mech. 69, 335–339 (2002)
26. Udwadia, F.E., von Bremen, H.: An efficient and stable approach for computation of Lyapunov
characteristic exponents of continuous dynamical systems. Appl. Math. Comput. 121, 219–259
(2001)
The General Conception of the Intellectual
Investigation of the Regular and Chaotic
Behavior of the Dynamical System
Hamiltonian Structure
Constantin Ruchkin
Abstract In this article the general conception of computer research of the

intellectual method of the regular and chaotic behavior of the dynamical system
of special Hamiltonian structure is presented. In accordance with the main idea of
the proposed conception, graphic images of two-dimensional and three-dimensional
Poincare’s sections of the phase space of dynamical systems by methods artificial
and computational intellect are investigated. The new approach generalizes existing
classical methods of the recognition of graphic images for badly formalized graphic
objects on Poincare’s sections such as “a closed curve” and “a two-dimensional
area.” We propose to use the generalized Hough transform for identification of
regular cases, and also methods of deformable active contours and deformable
parametric models for the recognition of chaotic cases. On the basis of the given
technique, the computer program of the intellectual research of the regular and
chaotic behavior of the mechanical systems, showing serviceability and efficiency
of the general conception, was developed.
1 Introduction
At the present time, the investigations of the nonlinear dynamical systems and
the dynamical systems with special Hamiltonian structure are closely related to
the classical methods of analytical dynamics such as KAM theory, Lyapunov’s
stability theory, and more [1, 9, 12, 16]. Currently, for the analysis of the nonlinear
dynamical systems, new computer-based theories are increasingly used: ergodic
theory, statistical forecasting methods, numerical methods of research of high
accuracy and performance, computer cognitive research methods, and methods of
computational and artificial intelligence [3, 5, 6, 10, 13, 14, 17, 19]. A new approach
gives better results if they combine analytical, numerical, and algorithmic idea
simultaneously. Recent investigations show that the new hybrid methods can be
C. Ruchkin ()
Donetsk National Technical University, FCSaT, str. Artema., 58., 83001 Donetsk, Ukraine
e-mail: c_ruchkin@mail.ru

246 C. Ruchkin
used to produce not only approximate solutions but also analytically exact solutions,
which remain as the prerogative of analytical dynamics.
In this article, in addition to previous work [17], the general conception of the
intellectual investigation of the nonlinear dynamical systems with the Hamiltonian
structure will be presented. This concept is based on the methods and algorithms for
the investigation of the regular and the chaotic behavior of nonlinear dynamical
systems and obtains analytically exact solutions. Principles of the KAM theory
about reorganizations (creation and destruction) of trajectory toruses in phase
space at the perturbation of the regular behavior of Hamiltonian systems are the
mathematical basis of the conception of the computer research by methods artificial
and computational intellect.
2 Related Research and the Analysis of the Current

State of the Problem
Fundamentals of the analytical study of the integrable dynamical systems with

special Hamiltonian structure are based on KAM theory. However, in the practical
computing, algorithmic problems related to the complexity of the behavior classifi-
cation of the nonlinear dynamical systems of the highest order are arise. Of [4, 6, 10,
11, 13–15, 17] are devoted to the solution of these problems and the development
of algorithms and intelligent methods for the study of dynamical systems. Consider
this work in more detail.
For the first time, mathematical models of pattern recognition, models of an
object, and pattern recognition training in the article of Nejmark, Yu. I. are
discussed. In their works [13, 14], authors propose to use the pattern recognition
methods to study mathematical models of the dynamical system. The possibilities
of using the pattern recognition methods to study mathematical models with a large
number of parameters are discussed. The principal point is to study models by
constructing phase and parametric portraits. This allows one to solve the problems
of predicting the states of the object described by the mathematical model in hand
and controlling the object and analyzing and studying problems that follow from the
particular content of the model. Examples of three mathematical models are given
to illustrate this problem.
In [10], it developed and implemented a methodology and data processing
algorithms for interactive study of the behavior of the nonlinear dynamical systems
based on mathematical modeling, taking into account the possible degeneracy of
the singular points, and technological constraints. The behavior of the nonlinear
dynamical system near degenerate critical points is well studied and classified.
Along with this, the classical analysis is ambiguous in those cases when a singular
point is degenerate, such as Kolmogorov. To analyze the behavior of the system in
such a case, the algorithm is based on the idea of algorithmic search dividing lines
in the state space.
The General Conception of the Intellectual Investigation of the Regular. . . 247
The main important feature of the dynamic behavior of the system is the chain-
recurrent set—the set of pseudo-periodic trajectories. It brings a lot of periodic
orbits, and its location is the initial step of many methods of investigation of
dynamical systems. The localization of these sets (except for the simplest-fixed
points) is difficult when using classical methods of numerical analysis, as they
provide some information on the asymptotic behavior of the system. To solve this
problem, it uses the idea of partitioning the phase space into a finite set of cells
and modeling the behavior of the system in accordance with the transformation of
these cells by the action of the system. Refinement of the phase portrait occurs
in sequential subdivision of cells covering. The practical implementation of such
methods faced with a rapid increase in the number of cells, which increases the
complexity of the design and implementation of workable methods. To study
and simulate the dynamical systems in [15], a method of symbolic image is
investigated. This method allows you to receive important characteristics of the
original dynamical system using the algorithms on a directed graph whose vertices
are the elements of the phase space, and the edges are defined by the behavior of the
system on the elements.
The localization of the chain-recurrent sets is achieved by allocating a symbolic
image of the graph of strongly connected component. This method depends strongly
on the method of the construction and of the recognition of the symbolic images.
The literature describes the most simple method of constructing the image of the
cell (we will call it point method), which examines the behavior of the system
through the system’s behavior on a fixed set of points [4, 15].
In the article [11], the methods of state recognitions of the dynamical systems
are considered. It is supposed that the characteristics of the dynamical systems, in
a random way, change in time and are described by some nonstationary time series.
Spectral expansion of nonstationary random process in Fourier series is applied. In
the procedure of the recognition of states of system as initial attributes, the set of the
random variables received from the description of features of spectrograms is used.
3 Basic Statement of the General Conception

of the Intellectual Investigation
The computer solution of the prediction problem of the regular and chaotic behavior
of the nonlinear dynamical systems can be obtained by several stages. At the first
stage, the process of the discretization of the system’s solutions by means of the
numerical integration is used. At the second stage, the process of graphics and
geometric modeling of the results is applicable. If the dimension of the system is
less than three, the result of this integration is conveniently represented in graphical
form on an image as a series of points forming a curve in a space and characterizing
the state of the system at any time At the third stage, the system’s behavior by the
kind of a phase trajectory is analyzed. In the case of the systems of the third and
higher orders, the result of numerical simulation of the system can be represented in
graphical form in three-dimensional space only by special images (projections and
248 C. Ruchkin
sections of Poincare). However, the computer simulation results in the fact that some
information on the nature of the system is lost and the results are incorrect. For more
information about the qualitative behavior of the trajectory space of the system we
construct sections of space curves. The general algorithm can be represented as:
1. Submission of the dynamical system in the Hamiltonian form
2. The initial conditions and parameters of integration
3. Finding numerical solutions for a limited time period
4. Construction of the phase portrait of the dynamical system
5. Calculation and visualization of the Poincare section of the phase space
6. Intelligent analysis of graphic images and finding critical points, periodic,
quasiperiodic solutions, and other regular or chaotic solutions
7. The perturbation solutions of the system and reanalysis
4 General Proposition of -£¯ Approach and the Integrable

Hamiltonian System
The system with the Hamiltonian is called integrable if there is a canonical

transformation to a new variable angle effect. The existence of n integrals of
motion J1, J2, : : : , Jn means that a 2n-dimensional phase space of an integrable
Hamiltonian path belongs to the n-dimensional set, which by analogy with the circle
in the case of a harmonic oscillator has the topology of the n-dimensional torus.
The trajectory is given by the winding. If the ratio of the frequencies along the
meridians parallels rationally ! 1 : ! 2 D m : s, where m and s are relatively prime,
then the trajectory is closed. If the frequency ratio is irrational, then the trajectory of
a dense way fills the surface n-dimensional torus. In this case the motion is ergodic.
According to the KAM theory, torus on which the winding “not irrational”
(j! 1 : ! 2 m : sj > K(") : S2,5 ) become unstable and will
be destroyed. In the last
p
cases torus will be destroyed with !1 W !2 D 5 1 =2. Also according to the
KAM theory, smaller tori are born along with the destruction of some tori, which in
the Poincare section correspond to elliptic fixed points interspersed with hyperbolic
fixed points. The process of destruction of some tori and the birth of other smaller
breeds continues the self-similar distribution of elliptic and hyperbolic fixed points
in the Poincare section (Fig. 1).
Fig. 1 Closed and unclosed trajectories on the torus

5 The Construction of Poincare’s Sections
The numerical researches of phase space, which consist of a set of phase trajectories,
will be conducted by means of Poincare’s sections. Poincare’s sections have
dimensionality on unit is less than dimensionality of researched dynamical system.
The exceptional interest of the dynamical systems of the third and fourth order
is represented. The result of these researches can be displayed graphically on the
computer monitor. Poincare’s sections (local or global sections) for such systems
will represent certain graphic images on a plane or in space accordingly. If points
of a phase flow form a curve, it is possible to speak about the regular behavior
(periodic or multiperiodic) of Hamiltonian systems. The cloud of points appearing
in the section of Poincare of a phase flow will testify to the approach of a “chaotic”
behavior of the system.
The particular interest is represented by dynamical systems for which there
is a possibility of reconstruction and investigation of global Poincare’s section,
and for them there is a possibility of the creation of all possible motion. Though
the creation of the section of Poincare of phase space happens approximately by
means of numerical integration methods on the fixed interval of time, it appears
enough what to understand in the overall picture of the behavior of Hamiltonian
system.
The received phase portraits of two-dimensional and three-dimensional sections
of Poincare of Hamiltonian systems can be researched by means of statistical
or determined pattern recognition techniques. The complexity of the application
of classical methods of recognition consist that main graphic images of two-
dimensional and three-dimensional Poincare’s sections are consisted concept “a
closed curve” and “a two-dimensional area”.
6 Recognition’s Problem of Integrable Cases
In the work [18] the global spherical Poincare sections are investigated with help
the computer program “Modeler”. This section builds a three-dimensional sphere
and shows a set of points (point cloud). In regular cases, these sets of points form
a three-dimensional “closed curve” with self-crossing or without self-crossing. A
particular case of this curve is a circle, which lies on the surface of a sphere. The
center of this circle can coincides with the center of the sphere. If the center of
the circle pass through the center of the sphere, then the radius of the sphere and the
circle are the same, otherwise the radius of the circle is smaller than the radius of the
sphere. Therefore, the problem of the detection and recognition of three-dimensional
convex closed analytic curves constructed on the Poisson sphere, is an important and
actual problem which can be solved by means of the generalized spherical Hough
transform. In the work [18] it is shown how to solve the problem of the detection
and recognition of the circles formed by the points lying on the surface of arbitrary
250 C. Ruchkin
sphere. To solve this problem, it is proposed to use a method, which generalizes

the classical Hough transform to three dimensions on a spherical and generalized
Hough transform. In the proposed algorithm we used an accumulator array increased
dimension, which coincides with the number of unknown parameters in the equation
of the family of the required curves. Computational complexity of the voting process
carried out in the Hough extended accumulator space depends on the sampling of
the parameter space. The main limitation on the use of the method—noisy source
image—must be overcome in the preliminary stage.
7 Recognition’s Problem of Chaotic Cases
So, the chaotic behavior of the system can be detect in the case when a phase
portrait of a dynamical system is include closed areas. In this paper, to solve
this problem, we use the methods of deformable active contours and deformable
parametric models [7].
The method of active contour means to building variable circuit which is
composed of n points in two-dimensional spaces. Each point of this contour
is the approximated border of the object. For each point is the energy Ei :
Ei D aEint (vi ) C bEext (vi ), where Eint (vi ) is the energy component, which depends
on the shape of the contour; Eext (vi ) is the energy component, which depends on
image properties, such as gradient; a, b are the weighting coefficients to ensure
each energy contribution to the overall equation criterion; and Ei , Eint , Eext is the
square matrix. The value at the center of each matrix corresponds to the energy in
the energy point vi (i—the vertex contour). The remaining values in the matrices
correspond to the energy in each point in the environment vi . Each vertex vi has the
potential to move to any point vi , corresponding to the minimum value of the energy
Ei , where pjk (vi ) is the point (x, y), which corresponds to points on the image in the
energy matrix. If the energy function is chosen optimally, the top loop V consistently
moves in and stays close to the borders of the object. In the cases of symmetric
dynamical systems for solving the problem of random selection of areas on the
bitmap is used the method of parametric deformable model. Deformable models are
called parametric when the shape of the detected object is determined only by the
parameters of the system. Conformity assessment model configuration shown in the
image data is also produced by the energy of the model. This model depends on the
material parameters; the sum of the internal energy, which expresses the value of
the configuration of the model limits set by the developer; and the external energy,
which measures the goodness of fit model and the data in the image. The adaptation
process of the model is similar of process of searching parameter vector and of
reaching the global maximum (minimum) of the energy model. The specific model
is characterized by the method of defining the form of the simulated object and
function that calculates the energy model. The kind of the models is defined by a
set of algebraic curves of the second, third, or higher order with certain restrictions
imposed on their possible configurations. Internal energy models set additional
limits on the desired configuration, bringing the weight penalty function to the total
energy model for unwanted deformations. The calculation of the external energy is
the basis of image features that are found in the allocated object. The external energy
model increases when the model tends to take the position of the image where the
pixels form a structure similar to a recognizable object. Energy optimization models
were made using a gradient search method local minimum.
The proposed by the author the model and the algorithm for detection chaotic
regions or algebraic curve uses the methods of deformable active contours and
deformable parametric models.
8 Intellectual Research Problem of the Motion of a Rigid

Body with a Fixed Point
The problem of the motion of a rigid body with a fixed point is considered in the
classical setting. This problem refers to the direct problems of mechanics. For a
given static, kinematic, and dynamic (structural) parameters of a rigid body with a
fixed point and the initial traffic conditions necessary to determine its trajectory in
space at any given time, to determine the type of trajectories, and to establish the
nature of the movement of the body [6, 17].
A mathematical model of the motion of a free rigid body in a mobile basis is a
system of six ordinary differential Euler equations:

J ! D J ! ! C r ; D !; (1)
where J D Diag (A, B, C) is the inertia tensor, ! D (p, q, r) is the angular velocity of
the body in the projection on the movable axle, D ( 1, 2, 3 ) is the vertical unit
vector, and r D (r1, r2, r3 ) is the vector from a fixed point to the center of the mass
of the body.
Equation (1) to determine in the phase space R12 (!, , J, r) D R 6 (!, ) R 6
(J, r) family of possible (regular and chaotic) trajectories of the dynamical system.
The initial conditions are ! 0 D !(0), 0 D (0).
The regular case is characterized by “a good behavior” of a dynamical system,
sustainable trajectory with respect to certain perturbations of the initial conditions,
and the ability to determine the position of the system at a sufficient period of time.
Chaotic motion has shown substantial dependence of solutions on the initial data,
and the trajectories are exponentially unstable. In the chaotic case, even a small
perturbation to predict the behavior of the system is possible only in a finite interval
of time, called the time horizon.
Thus, the aim of the work [17] is to develop, design, and create an interactive
computer system which can be used to find the solution of system (1), determine its
character, and visualize the motion of a rigid body with a fixed point. If system (1)
252 C. Ruchkin
can be integrated and obtains the explicit dependence of the phase variables of time,
it could be easily distinguished from the regular case of a chaotic event.
It is known [6] that the solution of (1) can be reduced to quadrature if four
integrals are found. However, for arbitrary values of the design parameters, there
are only three of the first algebraic integrals: H (!, ), the energy; G (!, ), the
kinetic; and I (), the geometric:
1
H .!; / D J ! ! r D h; G .!; / D J ! D g; I .; / D D 1:
2
(2)
The fourth integral exists only in the three most common cases, Euler–Poinsot,
the Lagrange–Poisson, and Kowalewski, and some special cases. All of these
cases are characterized by a specific set of design parameters, and the existence
of four integrals, which allow reducing the system of differential equations (1)
to a system of algebraic equations. The features of the integrable cases are the
deterministic phase space trajectories and the increasingly stable trajectory of the
body.
In the other integrable cases, the solution of a system of differential equations
(1) can be received by the Runge-Kutta method of fifth order. Determine the nature
of the family of movements and the opportunity to build the structure by analyzing
the trajectory of the phase space of the dynamical system under study. For a fixed
point of R6 (J, r), the first integrals (2) are isolated in the phase space R6 (!, ), a
compact three-dimensional manifold Q3 h,g invariant with respect to the flow. The
arrangement and relationship Q3 h,g in R3 (!) and R3 () give a fairly general idea of
changing the physical quantities of the (1).
The basis for further qualitative analysis of the orbital structure of the phase
space is the method of phase Poincaré sections and the theory of invariant curves.
The onset of the chaotic regime occurs when the Poincaré section of the phase flow
to form a cloud of points - as a two-dimensional region. If the phase flow lies on the
curve, then we can talk about the regular behavior of the system. To investigation the
topology of Q3 h,g by constructing an integrable case for local and global sections P2
Q3 h,g for the non-integrable case on the Poisson sphere S2 () D 1. Global section
of P2 is in Q3 h,g compact two-dimensional manifold. The trajectory, is filled of Q3 h,g ,
or cross the face of P2 . P2 are defined curves along which the phase flow is not
transversal to the selected section, and the track points back at the material section
for the final time interval.
On the basis of the given concept, the computer program of intellectual research
of the regular and chaotic behavior of several mechanical systems, showing
serviceability and efficiency of the given approach, was developed. An example of
solving the problem of recognition of regular and chaotic cases on 3d Poincare’s
section is shown in Fig. 2. This Poincaré section is calculated by Runge-Kutta
method for (1) and is reconstructed on the Poisson sphere using equations (2).
Fig. 2 Recognition of the regular (a) and chaotic (b) cases on 3d Poincare’s section
9 Conclusions and Future Tasks
In this article the general concept of the intellectual investigation of the nonlinear
dynamical systems with a special Hamiltonian structure will be presented. The
concept is based on the methods and algorithms for the study of the regular and
chaotic behavior of the nonlinear dynamical system. The new approaches generaliz-
ing existing classical methods of recognition for badly formalized concept “a closed
curve” and “a two-dimensional area” for recognition are considered. We propose
to use the generalized Hough transform to identification of regular cases, and also
methods of deformable active contours and deformable parametric models for the
recognition of chaotic cases. With its help we can solve the problem of detecting the
regular or chaotic behavior of dynamical systems of special Hamiltonian structure.
The result of the computer program of the intellectual research of the regular and
chaotic behavior of a mechanical system is shown.
Future investigations will address the following problems: the choice of method
of integration, the choice of the kind and form of the Poincare section, the specific
analysis of graphical images and pattern recognition methods in space, and full
automation of the research of the dynamical system.
References
1. Arnold, V.I.: Small denominators and problems of stability of motion in classical and celestial
mechanics. Russ. Math. Surv. 18, 85 (1963)
2. Ballard, D.H.: Generalizing the Hough Transform to Detect Arbitrary Shapes, p. 122.
Computer Science Department, University of Rochester, New York (1980)
3. Bolsinov, A.V., Fomenko, A.T.: Integrable Hamiltonian system. In: Geometry, Topology,
Classification, vol. 2, p. 448. Udmun-ton, Izhevsk (1999)
254 C. Ruchkin
4. Fundinger, D.: Investigating dynamics by multilevel phase space discretization. Ph.D. thesis,
Institut fur Parallele und Verteilte Systeme, Abteilung Bildverstehen, 2006. http://elib.uni-
stuttgart.de/opus/volltexte/2006/2614
5. Galgani, L., Giorgilli, A., Strelcyn, J.-M.: Chaotic motions and transition to stochasticity in
the classical problem of the heavy rigid body with a fixed point. Nuovo Cimento 61B(1), 1–20
(1981)
6. Gashenenko, I.N., Richter, P.H.: Enveloping surfaces and admissible velocities of heavy rigid
bodies. Int. J. Bifurcat. Chaos 14(8), 2525–2553 (2004)
7. Kass, M., Witkin, A., Terzopoulos, D.: Snakes: active contour models. Int. J. Comput. Vis.
1(4), 321–331 (1988)
8. Khoshelham, K.: Extending generalized Hough transform to detect 3D objects in laser range
data. IAPRS XXXVI(Part 3), 206–210 (2007)
9. Kolmogorov, A.N.: On the conservation of conditionally periodic motions under small
perturbations of the Hamiltonian. Dokl. Akad. Nauk SSSR, 98, 527–530 (1954)
10. Kononova, £.I., Troyanovskiy, V.M.: Isolation zones and special points in the state space of a
nonlinear dynamical system. Rev. Appl. Ind. Math. 16(4), 669–672 (2009)
11. Kuk, J.V.: Intellectual of recognition systems of states a dynamical objects with nonstationary
characteristics. In: Kuk, J.V., Lavrikova, E.I. (eds.) Artificial Intelligence, vol. 4, pp. 763–773.
NAS Ukraine, Donetsk (2006)
12. Moser, J.: KAM theory and problems of stability: Per./Per. In: Treshchev, D.V. (ed.) Regular
and Chaotic Dynamics, p. 448. Institute of Computer Researches, Moscow-Izhevsk (2001)
13. Neimark Yu, I., Teklina, L.G.: On possibilities of using pattern recognition methods to study
mathematical models. Pattern Recogn. Image Anal. 22(1), 144–149 (2012)
14. Nejmark, Yu. I.: New the approach to numerical research of specific dynamic systems by
pattern recognition and statistical modeling methods. In: Nejmark, Yu. I., Kotelnikov, I.V.,
Teklina, L.G. (eds.) Application-Oriented Nonlinear Dynamics, vol. 18(2), pp. 3–15s. News
High Schools (2010)
15. Osipenko, G.: Dynamical Systems, Graphs, and Algorithms. Lecture Notes in Mathematics,
vol. 1889, p. 288. Springer, Berlin (2007)
16. Poincare, A.: New Methods of Celestial Mechanics: Fav Works in 3 Volumes, p. 999. Nauka,
Moscow (1972)
17. Ruchkin, K.A.: Development of computer system for analysis of Poincare sections. In:
Artificial intelligence, vol. 1, pp. 83–87. NAS Ukraine, Donetsk (2009)
18. Ruchkin, K.A.: Application of Hough transform to the recognitions problem of regular
solutions of dynamical systems. In Artificial intelligence, 1, 112–119. NAS Ukraine, Donetsk
(2013)
19. Schwartzman, S.: Global cross sections of compact dynamical systems. Proc. Natl. Acad. Sci
48, 786–791 (1962)
Dynamic Properties of Two-Axle Freight
Wagon with UIC Double-Link Suspension
as a Non-smooth System with Dry Friction
Jan Matej, Jarosław Seńko, and Jan Awrejcewicz
Abstract The influence of chosen parameters on the lateral dynamic behavior of

the two-axle freight wagon with UIC double-link suspension is presented. This type
of suspension uses the advantages of the studied non-smooth mechanical system,
since dry friction is used to damp the system vibrations. Mathematical models
of this suspension with and without lateral bump-stop are derived owing to non-
smooth mechanics assumptions being based on the Coulomb law regarding friction
and implemented into the MBS program. Numerical simulations of dynamics of
the analyzed system are performed on a straight track followed by the methods
appropriate for predicting the dynamic stability of railway vehicles. A dynamic
reply of the vehicle to the railway track excitation, in the form of the initial
condition, is monitored and studied. The carried out analysis mainly concerns the
investigation of limit cycle dynamics exhibited by the system elements in terms
of safety of the freight wagon. In addition, an influence of the friction coefficient
coupling the interacting elements of suspension and being responsible for damping
properties of our non-smooth system is analyzed. Furthermore, an occurrence of
a low critical speed is illustrated and its physical meaning is explained, and the
usefulness of applying the lateral bump-stop in the structure of UIC double-link
suspension is justified.
1 Introduction
Many of freight wagons in Europe are equipped with the UIC link suspension in
which damping is provided only by dry friction in pivoted joints of linkages. The
structure of this suspension is quite simple, but its elements have strongly nonlinear
J. Matej () • J. Seńko

Department of Automobile and Heavy Machines, Warsaw University of Technology,
84 Narbutta Street, 02-524 Warsaw, Poland
e-mail: jmt@simr.pw.edu.pl; jsenko@wp.pl
J. Awrejcewicz
Department of Automation and Biomechanics, Lodz University of Technology,
Stefanowski Street, 90-924 Lodz, Poland

256 J. Matej et al.
characteristics. Such a type of suspension was examined by many researchers [1, 2].
Nonlinear simulation model for freight wagons with UIC double-link suspension
in two-axle freight wagon was developed, for instance, by Hoffmann in [1], but
without the lateral bump-stop limiting lateral displacements of the lower link. The
purpose of the presented article is to compare the dynamic properties of the two-axle
freight wagon equipped with standard UIC double-link suspension with and without
lateral bump-stop. The mathematical model of the UIC double-link suspension,
prepared according to non-smooth mechanics assumptions, was implemented into
the MBS program [6]. The numerical examinations were conducted on the straight
track. They included different values of coefficient of friction in joints of the UIC
double-link suspension without and with lateral bump-stop, as well as the state of
loading the car body (empty/fully loaded). In the case of the leaf spring, a piecewise
linear and progressive characteristic in vertical direction was taken into account.
The utilized MBS program enabled to use pre-calculated and tabulated wheel–rail
contact parameters generated for different lateral positions for S1002 and UIC60
real, nonlinear profiles of the wheel and rail. We recognize the wagon movement
as stable, if dynamic reply of the vehicle to the railway track excitation has the
form of decaying lateral oscillations of wheel sets, returning to the center line of the
track. At sufficiently high speed of the vehicle, the oscillations following an external
disturbance grow and lead to a limit cycle. In general, freight wagons have low
critical speeds. Fully developed limit cycle oscillation is usually termed “hunting.”
The lowest vehicle speed at which sustained oscillations appear is named the critical
speed.
2 Elements of the Freight Wagon with the UIC

Double-Link Suspension
The considered two-axle freight wagon with UIC double-link suspension consists
of the body and two wheel sets are shown in Fig. 1. In the structure of the wagon
the leaf spring 2 is located between the axle box 4 and the car body 1 (Fig. 2).
Wheel sets are guided in lateral and longitudinal direction, utilizing the links 3.
Longitudinal displacements of wheel set are blocked by guiding fork 6. Brackets 5
enable the transfer of vertical load from the car body to the axle boxes and wheels.
Each of the wheel sets can move relatively to the car body in the range of lateral and
longitudinal clearances, equal to 20 and 22,5 mm accordingly.
Upper and lower links 3, tumbling blocks 7, pins 8, C-washers 9, and links
10, shown in Fig. 3 and in Fig. 4, are the main elements of the UIC double-link
suspension. Each tumbling block rests on the pin and is secured in the lateral
direction by the C-washer. When the wheel set moves relative to the car body, the
upper and lower links roll or slide over the tumbling blocks. Rolling in the joint is
possible, because the radii of the links’ cross sections are smaller than the radii of
the half hole in the tumbling blocks.
Dynamic Properties of Two-Axle Freight Wagon with UIC Double-Link. . . 257
Fig. 1 Side view of the two-axle freight wagon adapted to transport of cars [4]
Fig. 2 Sketch of wheel set guiding with the UIC double-link suspension
Fig. 3 Elements of UIC double-link suspension without the lateral bump-stop
3 Mathematical Model of the UIC Double-Link Suspension
Authors of this article used the mathematical models of standard UIC double-
link suspension proposed by Piotrowski [5]. The main assumptions, according to
these models, were the following: contacting elements of joints are cylindrical;
the Coulomb law of dry friction is applied for the description of friction in the
258 J. Matej et al.
Fig. 4 Elements of UIC double-link suspension with the lateral bump-stop
Fig. 5 Rheological model of the UIC double-link suspension for longitudinal direction in the case
without and with the lateral bump-stop [5]
joints; the elements of the joints are assumed to be rigid; the lateral and longitudinal
displacements of the suspension are not coupled. Rheological models, used to
describe the properties of the UIC double-link suspension, have the form of elastic
elements with dry friction, composed of springs and dry friction sliders. For
longitudinal direction the model is presented in Fig. 5. For lateral direction the
model of suspension with the lateral bump-stop has the form shown in Fig. 6 [4].
When there is a lack of the bump-stop, the spring k2 and slider u0 are removed.
In case of the UIC suspension with the lateral bump-stop, the two links are active,
if the lateral displacement of the leaf spring pivots Y < u0 . When Y > u0 , only the
upper link is active and the stiffness of the suspension is equal to k C k2 .
The break force T0y and spring with stiffness of k1 , describing the elastic element
with dry friction, may be replaced by the differential equation for the force T
(Fig. 7). Then the continuity condition for the slider and spring has a form
TP =k1 C vS D &P YP (1)

Fig. 6 Rheological models of the UIC double-link suspension with lateral bump-stop for lateral
direction [4]
Fig. 7 The slider and the spring replaced by friction force T [5]
where vs is the velocity of sliding and T0 is the break force (friction force). The
non-smooth relations, basing on the Coulomb law of dry friction and describing the
characteristics of friction slider, are as follows [5]:
W T 2 ŒT0 ; CT0 ; (2)

vS 2 K TP ; D : (3)
The cone K is described by the velocity of sliding:

260 J. Matej et al.
Fig. 8 Relation between the velocity of sliding and the force of friction
8
< f0 g if jT j < T0
vs 2 RC if T D CT0 ;
:
R if T D T0
where the function [.]C is defined as

u if u0
ŒuC D : (4)
0 if u<0
When jT0 j D T0 , it is necessary to consider the differential succession of

non-smooth relation (4). The characteristics shown in Fig. 8 are non-smooth, multi-
valued, and non-differentiable. It was applied to the vehicle simulation model. In
that way the models of the suspension have been described by the differential
equations of the first order, implemented in the MBS model of the 2-axle freight
wagon in the following manner, for the lateral and longitudinal direction:
8 ˇ ˇ
ˆ ky1 YP if ˇTy ˇ < T0y ;
< C
TPy D ky1 YP if Ty D CT0y ; (5)
:̂ C
ky1 YP if Ty D T0y ;
8
< kxi XP
ˆ
C
if jTxi j < T0xi
TPxi D kxi XP if Txi D CT0xi ; .i D 1; 4/ : (6)
:̂ C
kxi XP if Txi D T0xi
In formulas (5) and (6), the lateral deflection is described by Y, while X indicates
the longitudinal deflection of the suspension. The restoring forces Py , Px of the
suspension in lateral and longitudinal directions are defined as
Table 1 Parameters of the UIC double-link suspension

for the lateral direction without the bump-stop
f T0y /Q ky /Q [1/m] ky1 /Q [1/m]
0,20 0,36518E-01 0,34534EC0 l 0,10519EC02
0,50 0,83620E-01 0,34718EC0 l 0,10155EC02
Table 2 Parameters of the UIC double-link suspension for the lateral

direction with the bump-stop
f T0y /Q ky /Q [1/m] ky1 /Q [1/m] ky2 /Q [1/m]
0,20 0,36518E-01 0,34534EC0 l 0,10519 IEC02 0,34534EC01
0,50 0,83620E-01 0,34718EC0 l 0,10155EC02 0,34658EC01
Table 3 Parameters of the UIC double-link suspension model for the longitudinal direction
f T0x1 /Q T0x2 /Q T0x3 /Q T0x4 /Q kx /Q kx1 /Q kx2 /Q kx3 /Q kx4 /Q
0,2 0,750E2 0,536E2 0,69E3 0,32E2 0,712EC1 0,428EC1 0,245EC1 0,308EC0 0,107EC1
0,183E 0,117E 0,21E 0,89E 0,661E 0,448E 0,229E 0,415E 0,126E
0,5 1 1 2 2 C1 C1 C1 C0 C1
X
iD4
P y D ky Y C T y ; P x D kx X C Txi : (7)
iD1
The parameters of the model of the UIC double-link suspension without and with
lateral bump-stop have been prepared for dimensions located in the middle of the
tolerance field, given by the technical specification of DIN 5545 Standards [4]. Four
values of the coefficient of friction between rolling/sliding elements of joints have
been assumed: f D 0,2; 0,3; 0;4, 0,5. Values of stiffness and dry friction force for
lateral/longitudinal directions were scaled with the vertical load Q of the suspension
in the following manner: k/Q, k1 /Q, k2 /Q, kxi /Q, T0y /Q, T0xi /Q, (i D 1, 4). Chosen
unit parameters are presented for f D 0,2 and f D 0,5 in Tables 1–3.
4 The MBS Model of the Two-Axle Freight Wagon

with the UIC Double-Link Suspension
A multi-body simulation model of the two-axle freight wagon (18ı of freedom) was
prepared in the MBS program [6]. The UIC double-link suspension was modeled as
nonlinear, massless element. Mathematical models of this suspension were involved
into multi-body model in the form of the differential equations (5) and (6), for
each of wheel set. These differential equations have been integrated with the wheel
set template in the MBS program. For the nominal UIC60 profile for rails and
nominal S1002 profile for wheels with nominal radii equal to 0,42 m, the one-
point contact model (based on Kalker’s rolling contact theory [3]) was assumed. The
nonlinear wheel–rail geometry and pre-tabulated contact functions according to the
262 J. Matej et al.
Table 4 Mass and inertial parameters for two-axle freight wagon with
UIC double-link suspension
Moments of inertia (kg m2 )
Part Mass (kg) Ix Iy Iz
Empty car body 13,450 282,000 318 321,000
Fully loaded car body 27,450 595,140 618,140 596,970
Wheel set 990 535 74 535
FASTSIM algorithm of Kalker [6] were used. The coefficient of friction between
wheel and rail, accepted equal to 0,4. The track model was treated as a uniform
structure with the mechanical properties described by the parameters given in [6].
The inertial parameters used in simulations for the empty and loaded freight wagon
are presented in Table 4. The other parameters are as follows: the base of the wagon
equal to 10 m, track gauge equal to 1,435 m, rail inclination equal to 1:40, and the
border displacement of the leaf spring pivoting in lateral direction u0 D 10 mm. The
distance of the wagon center of gravity from the rail level was accepted as 1,527 and
2,162 m for empty and fully loaded car body appropriately.
5 Results of Numerical Simulation
The simulations have been performed for two cases: without and with the lateral
bump-stop in UIC double-link suspension. Lateral vibrations were caused by the
initial condition introduced in the form of lateral displacement of leading wheel set,
equal to 0,005 m, relative to the track center line. As a response we can observe the
hunting motion of wheel sets. We stated that critical speed of empty and fully loaded
wagon is very small and close to 10 m/s. Freight wagons are operated usually with
the maximum velocity smaller or equal to 120 km/h. In order to judge the properties
of the suspension, further investigations were done for velocity bigger than critical
speed of the wagon. As a criterion of evaluation, the lateral displacements of leading
wheel set were accepted. In the range of velocity from 10 to 20 m/s, an advantage
of UIC double-link suspension without lateral bump-stop, with the coefficient of
friction between rolling/sliding elements equal to 0,2 was stated. But for velocity
close to 35 m/s, lateral displacements of leading wheel set of empty wagon violently
grow and undesirable wheel flange contact between wheels and rails is observed
(Fig. 9). Simultaneously, the lateral amplitudes of leading wheel set of empty wagon
with the UIC double-link suspension equipped with the lateral bump-stop achieve
the values close to 5 mm for velocity equal to 15 m/s (Fig. 10). The limit cycle
of this wheel set is presented in Fig. 11. For velocity from 20 to 40 m/s, lateral
amplitudes of leading wheel set for this case are smaller than for the compared type
of suspension without the bump-stop. In addition they are smaller than 1 mm. It
means that there is no flange contact between wheels and rails.
Fig. 9 Amplitudes of lateral displacements of leading wheel set for wagon with the suspension
without the lateral bump-stop and coefficient of friction between rolling/sliding elements of joints
equal to 0,2, in the function of wagon velocity
with the lateral bump-stop and coefficient of friction between rolling/sliding elements of joints
In the context of previous conclusion according to the UIC double-link sus-

pension with lateral bump-stop, we need to explain the cause of the large lateral
amplitude of leading wheel set, close to 0,005 m, examining the case presented in
Fig. 10 for empty wagon. Therefore an analysis of a resonance between the lateral
excitation frequency of the wheel sets fw and the yaw eigenfrequency of car body
fyaw , as well as the roll eigenfrequency froll of the car body motion, was conducted.
These frequencies can be approximated in the following manner [1]:
264 J. Matej et al.
Fig. 11 Limit cycle of leading wheel set of empty wagon with the UIC double-link suspension
equipped with the lateral bump-stop, for f D 0,2 and v D 15 m/s
p
fw D v=2 p =b0 r0 ;
fyaw D 1=2 q4a2 ky =Iz ; (8)

frol l D 1=2 4 L21 kz C L22 ky =Ix ;
where a is the half wheelbase, ky and kz denote appropriately the lateral stiffness
of suspension and the vertical stiffness of the leaf spring, Iz and Ix are the yaw and
the roll moments of inertia of the car body, v is the speed of vehicle, denotes the
wheel conicity, 2b0 and r0 are the distance between the nominal rolling circles and
the nominal rolling radius of wheel, and L1 and L2 indicate the lateral and vertical
distances between the center of car body mass and the UIC suspension, respectively.
Values of parameters in formulas (8) were assumed to be known. The value of
wheel conicity parameter was accepted as D 0,13, which corresponds to lateral
displacements of wheel set in the scope up to 6 mm. Using Klingel’s formula, we
got from the equations (8) the value of lateral excitation frequency of the wheel
set fw equal to 1,53 Hz. Applying the spectrum analysis (fast Fourier transform)
of lateral oscillations of wheel sets we obtained more probable value of fw equal
to 1,5 Hz. In the next step the hysteresis loop presented in Fig. 12 and according
to lateral direction of the UIC double-link suspension with the bump-stop was
analyzed. The following values of ky stiffness were estimated: ky1 D 113470 N/m,
ky2 D 225860 N/m, and ky3 D 560760 N/m. Next, the yaw and roll eigenfrequen-
cies of the empty car body were calculated: fyaw1 D 1,05 Hz, fyaw2 D 1,49 Hz,
fyaw3 D 2,35 Hz, froll1 D 0,93 Hz, froll 2 D 1,16 Hz, and froll 3 D 1,68 Hz. We noticed
that the values of the lateral excitation frequency of the wheel set fw D 1,5 and yaw
Fig. 12 Hysteresis loops for empty wagon equipped with the UIC double-link suspension with the
bump-stop, for velocity equal to 15 m/s and coefficient of friction between rolling/sliding elements
equal to 0,2
eigenfrequency fyaw2 D 1,49 Hz are almost identical. It means that lateral oscillations
of wheel set and car body yaw motion are in the resonance. That is the direct cause
of the high leading wheel set amplitude of examined freight wagon, equipped with
UIC double-link suspension with the lateral bump-stop. Similar analysis for empty
wagon running with velocity greater than 20 m/s and for fully loaded wagon in
the range of velocities from 10 m/s to 40 m/s showed that instability is caused
by small damping of the lateral displacements of wheel set. Hysteresis loop for
fully loaded wagon equipped with the UIC double-link suspension with the bump-
stop, for velocity equal to 15 m/s and coefficient of friction between rolling/sliding
elements equal to 0,2, is shown in Fig. 13. Similarly, but more worse situation
according to hunting motion of leading wheel sets we can observe for coefficients
of friction between rolling/sliding elements of joints greater than 0,2, especially
for f D 0,5 (Figs. 14 and 15). Also in these cases the high lateral displacements of
leading wheel sets are caused first through a resonance between the lateral excitation
frequency of the wheel sets fw and the yaw eigenfrequency fyaw of the empty and
fully loaded car body.
When the speed of wagon increases, we still can observe the limit cycles of
wheel sets caused by small damping properties of the suspension. According to
earlier assumptions, the lateral and longitudinal directions of the UIC double-link
suspension should not be coupled. In practice, wheel sets have the possibilities to
yaw and in that way the longitudinal displacement of the leaf spring occurs. In the
end, suspension works in two directions; however the longitudinal restoring forces
Px are almost four times smaller than lateral restoring forces Py (Fig. 16).
266 J. Matej et al.
Fig. 13 Hysteresis loop for fully loaded wagon equipped with the UIC double-link suspension
with the bump-stop, for velocity equal to 15 m/s and coefficient of friction between rolling/sliding
elements equal to 0,2
without the lateral bump-stop and coefficient of friction between rolling/sliding elements of joints
6 Conclusions
In the presented article the influence of chosen parameters on the lateral dynamic
behavior of the two-axle freight wagon with the UIC double-link suspension was
showed. Mathematical models of this suspension with and without lateral bump-stop
were derived owing to non-smooth mechanics assumptions being based on the
Coulomb law regarding friction.
Very small and close to 10 m/s, critical speed of empty and fully loaded freight
wagon with the UIC double-link suspension was stated.
with the lateral bump-stop and coefficient of friction between rolling/sliding elements of joints
Fig. 16 Restoring forces in the function of lateral displacement of leading wheel set for empty
wagon equipped with the UIC double-link suspension with the bump-stop, for velocity equal to
15 m/s and coefficient of friction between rolling/sliding elements equal to 0,2
An advantage of the UIC double-link suspension without lateral bump-stop, for

the coefficient of friction between rolling/sliding elements equal to 0,2, was showed
for velocity smaller than 30 m/s.
The appearance of high lateral amplitudes of wheel sets, of the empty or fully
loaded freight wagon with the UIC double-link suspension with or without lateral
268 J. Matej et al.
bump-stop, was interpreted as a resonance between the lateral excitation frequency

of the wheel sets and the yaw eigenfrequency of the car body.
It was also demonstrated, that if the speed of two-axle freight wagon increases,
we still can observe the limit cycles of wheel sets, caused by small damping
properties of the suspension, for two analyzed cases of suspension.
References
1. Hoffmann, M., True, H.: The dynamics of European two-axle railway freight wagons with UIC
standard suspension. Vehicle Syst. Dynam. 46, 225–236 (2008)
2. Jönsson P.: Modelling and laboratory investigations on freight wagon link suspensions, with
respect to vehicle-track dynamic interaction, Ph. D. thesis, Stockholm, Sweden (2004)
3. Kalker, J.J.: Three-dimensional elastic bodies in rolling contact. Kluwer Academic Publishers,
Dordrecht/Boston/London (1990)
4. Matej, J., Piotrowski, J., Seńko, J.: Analysis of ride dynamics of L02A Car Wagon unit based
on numerical simulations. Internal Report, Warsaw (2010)
5. Piotrowski, J.: Model of the UIC link suspension for freight wagons. Archive of Appl. Mech.
73, 517–532 (2003)
6. Wheel–rail element. Reference guide, ADAMS RAIL users manual. 2005 (MSC Software).
Dynamics of Mechanical Systems
with Mecanum Wheels
Klaus Zimmermann, Igor Zeidis, and Mohamed Abdelrahman
Abstract The kinematics and dynamics of a mechanical system with mecanum

wheels is studied. A mecanum wheel is a wheel with rollers attached to its
circumference. Each roller rotates about an axis that forms an angle with the plane
of the disk (for the omni-wheels, the axes of the rollers lie in the plane of the
wheel and in an ideal case are tangent to the outer circumference of the wheel).
Such a design provides additional kinematic advantages for the mecanum wheels in
comparison with the conventional wheels. Within the framework of non-holonomic
mechanics, the equations of motion are derived for the case of an arbitrary angle at
which the rollers are attached (usually, this angle is assumed to be equal to 45ı ). In
robotics, a simplified approach, in which the equations of non-holonomic kinematic
constraints are solved approximately by means of a pseudo-inverse matrix, is
frequently applied. Such an approximate approach leads to “holonomization” of
the system and allows Lagrange’s equations of the second kind to be used. In the
present paper, the equations of motion obtained on the basis of the principles of non-
holonomic mechanics are compared with the approximate equations. It is shown that
for translational motions and for the rotation of the system about its center of mass,
both these approaches lead to the same result.
1 Introduction
Mobility is an important property for people. Thus, its generation, conservation,

and, if necessary, recovery are an important goal and a great challenge for life
scientists and engineers. Some tasks encompass fields from the design of faster
and—at the same time—energy-saving transportation systems in the air to the
K. Zimmermann () • I. Zeidis • M. Abdelrahman

Department of Technical Mechanics, Technische Universität Ilmenau, Max-Planck-Ring 12,
98693 Ilmenau, Germany
e-mail: klaus.zimmermann@tu-ilmenau.de; igor.zeidis@tu-ilmenau.de;
mohamed.abdelrahman@tu-ilmenau.de

270 K. Zimmermann et al.
x
y
Fig. 1 A mecanum wheel (left) and the mechanical model (right)
recovery of upright walking through exoskeletons. For service robots and equipment
for physically disabled persons, mobile systems with a high maneuverability play
an important role. The conventional wheel moves back into the focus of interest.
Seemingly, its basic mechanical function “rolling” is easy to understand and all
mechanics are adequately described. But, there exist some new developments, which
connect the advantages of wheels and legs in a whole system (e.g., “whegs”).
Furthermore, the so-called omnidirectional wheels (e.g., “mecanum wheels”),
generating constraints different from the conventional wheel, lead to investigations,
based on the mechanics and control of non-holonomic systems. At the present
time, vehicles with mecanum wheels (Fig. 1, left) are gaining ground for different
applications. The wheels of such vehicles have rollers that are arranged along the
rim at an angle to the wheel plane [5]. As a rule, this angle is equal to 45ı . Such
wheels have additional kinematical possibilities in comparison with conventional
wheels. Due to these possibilities, a vehicle with mecanum wheels can move
forward-backward and leftward-rightward and rotate in an arbitrary way. Usually
a mecanum-wheeled vehicle has four wheels. By varying the rate and the direction
of rotation of each wheel, one can implement, for example, a translational motion of
the vehicle in any direction, as well as arbitrary turns and rotations on the spot. The
kinematics of wheeled systems, including those with mecanum wheels, are reviewed
in [2]. The issues of kinematics, dynamics, and control of systems with mecanum
wheels in a non-holonomic treatment are considered in [6, 10, 11] for a number
of particular cases. There are a great number of studies on robotics, in which the
kinematics and dynamics of robots with four mecanum wheels is approximately
treated in terms of holonomic mechanics (see, e.g., [3, 4, 7–9]). In these studies,
pseudo-inverse matrices are used to resolve the kinematic constraint relations.
In the present study, the kinematics of a vehicle with four mecanum wheels are
considered in terms of non-holonomic mechanics for an arbitrary orientation of
the rollers of the mecanum wheels. The equations obtained are compared with the
equations implied by an approximate holonomic treatment, and the types of motion
for which both treatments coincide are found.
Dynamics of Mechanical Systems with Mecanum Wheels 271
2 Non-holonomic Model of a Mechanical System

with Mecanum Wheels
Using a non-holonomic model of a mecanum wheel, we will obtain the kinematic

relations and the equation of motion for a mechanical system with such wheels.
2.1 Kinematics of a Mecanum Wheel
For a conventional wheel, the contact between the wheel and the supporting plane
is characterized by the condition that the wheel is rolling without slip. This means
that the velocity of the point by which the wheel contacts the plane at each current
instant is equal to zero. Then the projections of the velocity of the contact point onto
the direction lying in the wheel plane, as well as onto the direction perpendicular to
this plane, are equal to zero. For a mecanum wheel, there is only one direction the
projection onto which of the velocity of the point of contact of the wheel with the
supporting plane vanishes. This direction can be arbitrary, but it is fixed relative to
the wheel.
As a model of a mecanum wheel we will consider the rolling of a disk of radius
R centered at the point C on a horizontal plane. The plane of the disk is always
!

vertical. Let l denote the unit vector along the axis of attachment of the rollers, !
the unit vector lying in the wheel plane tangent to the rim at the point of contact,
and ı the angle between the plane of the wheel and the roller axis (between the
and !
vectors !

l ). The angle ı is constant. The kinematic constraint relation for a
*
mecanum wheel implies that the vector of velocity v M of the point M of contact of
the wheel with the plane points along the line perpendicular to the axis of the roller,
i.e., the projection of the velocity of the point M onto the roller axis is equal to zero
(see Fig. 1, right).
The kinematic constraint relation has the form
v !
!
l D 0: (1)
M
The velocity !
v is defined by the equation
M
!
v D !
v C * !
M C ! CM (2)
where !v is the velocity of the center C and !

! is the angular velocity of the disk.
nC o
!
!
!

Let O; e x ; e y ; e z be a fixed reference frame (inertial system) and let C
n ! ! ! o
be the origin of a movable reference frame (body fixed frame) C; E x ; E y ; E z .
!
!
!

The unit vectors E x and E y are parallel to the horizontal plane, the vector E z
!

lying in the disk plane and the vector E y being perpendicular to this plane. Let
' be the angle of rotation of the disk about the axis passing through the point C
perpendicular to the plane of the disk, and the angle formed by the disk plane
!

with a line parallel to the vector E x (the angle between the vector !
and the vector
!
e , see Fig. 1, right). The vectors * !
x ! and CM are defined by
!
!
!
! !

! D 'P E y C P E z ; CM D R E z : (3)
n Let xC , yC , oR be the coordinates of the point C in the reference frame

e ; !
O; ! !
x e y ; e z . Then
v D .xP cos CyP sin / !

! !
E x C .xP C sin CyPC cos / E y ;
C C C
* ! !

!CM D R'P E x : (4)
!
The vector l is expressed by
!
!
!

l D cos ı E x C sin ı E y : (5)
Finally, the kinematic relation (1) becomes
xP C cos . C ı/ C yPC sin . C ı/ D R'P cos ı: (6)
On the basis of the analysis of the kinematic constraints of type (6) it is shown
[1] that if a mechanical system is based on n mecanum wheels in such a way that
!

(a) n 3; (b) not all vectors l i are parallel to each other; and (c) the points of
contact of the wheels with the plane do not lie on one line, then it is always possible
to find control functions ' i (i D 1, : : : , n) that implement any prescribed motion of
the system’s center of mass.
2.2 Kinematics of a Vehicle with Four Mecanum Wheels
Consider a model of a four-wheeled vehicle with mecanum wheels (Fig. 2).

Let now C be the center of mass of the system (the bodyntogether with theo
wheels). The coordinates of the point C in the reference frame O; ! e ; !
!
x e y; e z
ˇ!ˇ ˇ!ˇ
ˇ ˇ ˇ ˇ
are xC , yC , and R; the quantities ˇCO1 ˇ D 1 and ˇCO2 ˇ D 2 are the distances
from the center of mass to the axes of the respective wheel pairs, and 2 l is the
distance between the centers of the wheels of one axis.
1
δ1
δ3
02
3 Ey ρ2
Ex δ2
l ψ
ρ1
C
Ez
2
01
ey l
δ4
0
ez ex 4
Fig. 2 A vehicle with four mecanum wheels
The corresponding kinematic relations are obtained on the basis of Eq. (6), in
which ı should be replaced by ı 1 and ı 4 for wheels 1 and 4 and by ı 2 and ı 3
for wheels 2 and 3, respectively. These relations have the form
xP C cos . ı1 / C yPC sin . ı1 / P .l cos ı1 C 2 sin ı1 / D R'P1 cos ı1 ; (7)
xP C cos . C ı2 / C yPC sin . C ı2 / C P .l cos ı2 C 2 sin ı2 / D R'P2 cos ı2 ; (8)
xP C cos . C ı3 / C yPC sin . C ı3 / P .l cos ı3 C 1 sin ı3 / D R'P3 cos ı3 ; (9)
xP C cos . ı4 / C yPC sin . ı4 / C P .l cos ı4 C 1 sin ı4 / D R'P4 cos ı4 ; (10)
where ' i is the angle of rotation of the disk about the rotation axis and ı i is the
angle between the wheel plane and the roller axis (i D 1, : : : , 4).
Equations (7)–(10) define four non-holonomic constraints. Note that if only
translational motions ( P D 0) or only rotations about the center of mass (xP C D
yPC D 0) are allowed, then the constraints become holonomic.
2.3 Dynamics of a Vehicle with Mecanum Wheels
The configuration of the mechanical system under consideration is defined by seven

parameters, which involve the coordinates q1 D xC , q2 D yC of the system’s center
of mass C, the angle q3 D between the abscissa axis of the fixed reference frame
and a straight line perpendicular to the axis of the wheel pair, and the angles of
rotation of the wheels qi C 3 D ' i ( i D 1, : : : , 4).
To describe the motion of the system we will use Lagrange’s equations with
undetermined multipliers corresponding to the non-holonomic constraints:
d @T @T
1 cos . ı1 / 2 cos . C ı2 / 3 cos . C ı3 /
dt @xP C @xC
4 cos . ı4 / D 0; (11)
d @T @T
1 sin . ı1 / 2 sin . C ı2 / 3 sin . C ı3 /
d t @yPC @yC
4 sin . ı4 / D 0; (12)
d @T @T
C 1 .l cos ı1 C 2 sin ı1 / 2 .l cos ı2 C 2 sin ı2 /
d t@P @
C 3 .l cos ı3 C 1 sin ı2 / 4 .l cos ı4 C 1 sin ı4 / D 0; (13)
d @T @T
C i R cos ıi D Mi ; i D 1; : : : ; 4: (14)
dt @'Pi @'i
Here, T is the kinetic energy of the system, Mi are the torques applied to the
respective wheels, and i are Lagrange’s undetermined multipliers (i D 1, : : : , 4).
The kinetic energy T is defined as the sum of the kinetic energies of the body
and the wheels and is expressed as follows:
1 1 1
T D m 0 xP C2 C yPC2 C J0 P 2 C m1 4 xP C2 C yPC2
2 2 2
C 4xP C yPC .1 2 / sin 4yPC P .1 2 / cos

C 4yPC P .1 2 / cos C 2 P 2 12 C 22 C 2l 2
1
C 2J2 P 2
C J1 'P12 C 'P22 C 'P32 C 'P42 : (15)
2
Here m 0 is the mass of the body, m1 is the mass of each of the wheels, J0 is the
mass moment of inertia of the body about the vertical axis passing through the center
of mass, J1 is the mass moment of inertia of each wheel about its axis of rotation,
and J2 is the moment of inertia of each wheel about the vertical axis passing through
the center of the wheel.
Consider a practically important case, for which ı i D ı (i D 1, : : : , 4). Then,
eliminating the multipliers i in Eqs. (11)–(14), we obtain
.m0 C 4m1 / xR C C 2m1 R .1 2 / sin C 2m1 P 2 .1 2 / cos

2J1
D xR C C yPC P .1 C cos 2 cos 2ı/
R cos 2 ı
2

C yRC xP C P sin 2 cos 2ı C R .1 2 / sin sin2 ı
1
C Œ.M1 C M4 / cos . ı/ C .M2 C M3 / cos . C ı/ ; (16)
R cos ı
.m 0 C 4m1 / yRC 2m1 R .1 2 / cos C 2m1 P 2 .1 2 / sin

2J1
P sin 2 cos 2ı
D R
x C C P
y C
R2 cos 2 ı

C yRC xP C P .1 cos 2 cos 2ı/ R .1 2 / cos sin2 ı
1
C Œ.M1 C M4 / sin . ı/ C .M2 C M3 / sin . C ı/ ; (17)
R cos ı

J0 C 4J2 C 2m1 .12 C 22 C 2l 2 / R
C 2m1 xR c .1 2 / sin 2m1 yRc .1 2 / cos
2J1
C R .l 2 C 0:5.12 C 22 / C .l 2 0:5.12 C 22 // cos 2ı
R2 2
cos ı
Cl.1 C 2 / sin 2ı/
C xR c .1 2 / sin sin2 ı yRc .1 2 / cos sin2 ı

CxP C P .1 2 / cos sin 2 ı C yPC P .1 2 / sin sin2 ı
1
D Œl .M2 M1 C M4 M3 / cos ı C .1 .M4 M3 /
R cos ı
C2 .M2 M1 // sin ı : (18)
As a result, for given torques Mi (i D 1, : : : , 4), we have a system of three

equations for three variables, xC , yC , and . Then the angles ' i of rotation of the
wheels can be found from Eqs. (7)–(10).
In the vehicles that are usually considered, the angle between the plane of the
mecanum wheel and the roller axis is equal to 45ı ( ı D /4), and the center of mass
of the system coincides with its geometric center (1 D 2 D ). For this case, Eqs.
(16)–(18) are simplified and become

4J1 4J1 1
m C 2 xR C C 2 yPC P D Œ.M1 C M2 C M3 C M4 / cos
R R R
C .M1 M2 M3 C M4 /sin ; (19)

4J1 4J1 1
m C 2 yRC 2 xP C P D Œ.M1 C M2 C M3 C M4 / sin
R R R
.M1 M2 M3 C M4 / cos ; (20)
4J1 l C
JC C 2 .l C /2 R D .M1 M2 C M3 M4 / ; (21)
R R
where m D m0 C 4 m1 is the total mass of the system and JC D J0 C 4J2 C 4m1

(2 C l2 ) is the moment of inertia of the entire system relative to the center of mass.
3 Approximate Model
Let vx and vy be the projections of the velocity

n ! of the center of mass of the vehicle
! ! o
onto the axes of the coordinate system C; E x ; E y ; E z that is attached to the
vehicle’s
n body and o is turned by the angle relative to the fixed reference frame
!
!
!

O; e x ; e y ; e z . Since
vx D xP C cos C yPC sin ; vy D xP C sin C yPC cos ; (22)
the kinematic constraint equations (7)–(10) become
vx cos ı1 vy sin ı1 P .l cos ı1 C 2 sin ı1 / D R'P1 cos ı1 ; (23)
vx cos ı2 C vy sin ı2 C P .l cos ı2 C 2 sin ı2 / D R'P2 cos ı2 ; (24)
vx cos ı3 C vy sin ı3 P .l cos ı3 C 1 sin ı3 / D R'P3 cos ı3 ; (25)
vx cos ı4 vy sin ı4 C P .l cos ı4 C 1 sin ı4 / D R'P4 cos ı4 : (26)
In the matrix form, Eqs. (23)–(26) are represented as follows:
v D !
J ! v D v ; v ; P T ; !
v ; ! v D .R'P ; R'P ; R'P ; R'P /T ; (27)
C W C x y W 1 2 3 4
0 1
cos ı1 sin ı1 .l cos ı1 C 2 sin ı1 /
B cos ı2 sin ı2 l cos ı2 C 2 sin ı2 C
J DB
@ cos ı3
C: (28)
sin ı3 .l cos ı3 C 1 sin ı3 / A
cos ı4 sin ı4 l cos ı4 C 1 sin ı4
Following [9], we solve the matrix equation (27) for the vector !
v using the
C
C
pseudo-inverse matrix J to obtain
! v ; J C D J T J 1 J T :
v D J C ! (29)
C W
The pseudo-inverse matrix gives an approximate

solution of the matrix equation
(27) that minimizes the Euclidean norm J !vC !v
W of the residual.
In the particular case, for which all ı i (i D 1, : : : , 4) are equal to /4 and
1 D 2 D , the solution (29) becomes
R
vx D .'P1 C 'P2 C 'P3 C 'P4 / ; (30)
4
R
vy D .'P1 'P2 'P3 C 'P4 / (31)
4
P D R
.'P1 'P2 C 'P3 'P4 / : (32)
4 .l C /
The expression (15) for the kinetic energy becomes
1 1 1
T D m v2x C v2y C JC P 2 C J1 'P12 C 'P22 C 'P32 C 'P42 : (33)
2 2 2
Substituting relations (30)–(32) into expression (33) yields

1 R2 R2 2
T D m C 2
J C C J 1 'P1 C 'P22 C 'P32 C 'P42
2 8 16. C l/

R2 R2
C m JC .'P1 'P4 C 'P2 'P3 /
8 16. C l/2
R2
JC .'P1 'P2 'P1 'P3 'P2 'P4 C 'P3 'P4 / : (34)
16. C l/2
For this kinetic energy, write out Lagrange’s equations of the second kind for 4
variables qi D ' i
d @T @T
D Mi ; i D 1; : : : ; 4: (35)
d t @'Pi @'i
Notice that this procedure is valid only for systems subject to holonomic
constraints. By differentiating Eqs. (30)–(32) with respect to time and substituting
into the resulting relations the expressions for 'Ri found from Eq. (35) we obtain the
equations for the coordinates xC and yC of the center of mass and the angle of
rotation of the vehicle:

4J1 1
mC 2 xR C C yPC P D Œ.M1 C M2 C M3 C M4 / cos
R R
C .M1 M2 M3 C M4 / sin ; (36)

4J1 1
mC 2 yRC xP C P D Œ.M1 C M2 C M3 C M4 / sin
R R
.M1 M2 M3 C M4 / cos ; (37)
4J1
JC C 2 .l C /2 R D l C .M1 M2 C M3 M4 / : (38)
R R
Fig. 3 Simulation of the vehicle motion using Mathematica® (realized by B. Adamov) (left) and
the prototype of a vehicle with four mecanum wheels (right)
4 Conclusions
Equations of motion for a vehicle with four mecanum wheels were derived
proceeding from a non-holonomic model for mecanum wheels. By comparing the
non-holonomic model with an approximate model used in robotics we established
that both models lead to the same result in the particular cases for which the vehicle
either moves translationally or rotates about its center of mass. In both these cases,
the constraints become holonomic. Based on the model considered in this study,
an experimental verification of simulation results (Fig. 3 left) was realized, using a
prototype of a mobile robot with four mecanum wheels (Fig. 3, right).
Acknowledgments The authors thank Boris Adamov from Moscow State Lomonosov University
for the permission to publish the simulation results. The authors would also like to thank Siegfried
Oberthür and Tobias Kästner for the practical realization of the prototype. This study was supported
by the Development Bank of Thuringia and the Thuringian Ministry of Economic Affairs with
funds of the European Social Fund (ESF) under grant 2011 FGR 0127.
References
1. Borisov, A.V., Kilin, A.A., Mamaev, I.S.: An omni-wheel vehicle on a plane and a sphere. Rus.
J. Nonlin. Dyn. 7(4), 785–801 (2011) (Russian)
2. Campion, G., Bastin, G., D’Andrea-Novel, B.: Structural properties and classification of
kinematic and dynamic models of wheeled mobile robots. IEEE Trans. Robot. Automat. 12(1),
47–62 (1996)
3. Dickerson, S.L., Lapin, B.D.: Control of an omni-directional robotic vehicle with mecanum
wheels. In: National Telesystems Conference Proceedings, Atlanta, USA, pp. 323–328. March
26–27, 1991
4. Han, K.-L., Choi, O.K., Lee, I., Choi, S.: Design and control of an omni-directional mobile
robot for mobile haptic interface. In: International Conference on Control, Automation and
Systems, COEX, Seoul, pp. 1290–1295. Oct. 14–17, 2008
5. Ilon, B.E.: Wheels for a course stable selfpropelling vehicle movable in any desired direction
on the ground or some other base. US Patent 3,876,255, 1975
6. Martynenko, Y.G.: Stability of steady motions of a mobile robot with roller-carrying wheels
and a displaced centre of mass. J. Appl. Math. Mech. 74(4), 436–442 (2010)
7. Muir, P.F., Neumann, C.P.: Kinematic Modeling for Feedback Control of an Omnidirectional
Wheeled Mobile Robot, Autonomous Robot Vehicles, pp. 25–31. Springer, NewYork (1990)
8. Tsai, C.-C., Tai, F.-C., Lee, Y.-R.: Motion controller design and embedded realization for
mecanum wheeled omnidirectional robots. In: Proceedings of the 8th World Congress on
Intelligent Control and Automation, Taipei, Taiwan, pp. 546–551. June 21-25, 2011.
9. Viboonchaicheep, P., Shimanda, A., Kosaka, Y.: Position rectification control for mecanum
wheeled omni-directional vehicles. In: 29th Annual Conference of IEEE Industrial Electronics
Society, pp. 854–859, 2006
10. Wampfler, G., Salecker, M., Wittenburg, J.: Kinematics, dynamics, and control of omnidirec-
tional vehicles with mecanum wheels. Mech. Base. Des. Struct. Mach. 17(2), 165–177 (1989)
11. Zimmermann, K., Zeidis, I., Behn, C.: Mechanics of Terrestrial Locomotion, With a Focus on
Nonpedal Motion Systems. Heidelberg, Springer (2009)
Reliability Analysis of the Dynamics
of a Horizontal Drill-String
T.G. Ritto and Rubens Sampaio
Abstract Nowadays directional drilling, such as horizontal drilling, is very

common. This chapter deals with the reliability analysis of a horizontal drill-
string dynamics. A stochastic probabilistic computational model is considered,
where a bar model (tension/compression) is used for the drill-string, and the system
is discretized by means of the finite element method. The frictional forces between
the column and the borehole are uncertain and relevant to the dynamic behavior
of the structure. Therefore, a stochastic model is used for the frictional coefficient:
a random field with exponential autocorrelation function. Because a probabilistic
approach is employed, the reliability is measured in terms of the probability of the
system not perform as it should, i.e., exceed stress limit, start a crack, or do not
achieve a target efficiency. For the computation of the failure probability, importance
sampling is applied to avoid too many stochastic simulations, and the stochastic
system is analyzed for different scenarios.
1 Introduction
A drill-string is a slender structure used to drill rocks in search of oil and gas.
Basically, it is composed of thin tubes (drill-pipes) and thicker tubes (drill-collars)
with a bit fixed at the drilling end. In the beginning of oil exploration there were
mainly vertical wells, but nowadays directional (horizontal in special) drilling is
very common [1, 22]. Concerning the bit-rock interaction, Perneder et al. [12]
proposed bit-rock interface laws in directional drilling, and, concerning a curved
T.G. Ritto ()

Federal University of Rio de Janeiro, Department of Mechanical Engineering,
Centro de Tecnologia, Ilha do Fundão, 21945-970, Rio de Janeiro, Brazil
e-mail: tritto@mecanica.ufrj.br
R. Sampaio
PUB-Rio, Mechanical Engineering Department, Rua Marquês de São Vicente,
225, 22451-900, Rio de Janeiro, Brazil
e-mail: rsampaio@puc-rio.br

282 T.G. Ritto and R. Sampaio
Fig. 1 Sketch of a vertical and a horizontal drilling
drill-string, Hu et al. [5] used curved-beam elements to model the drill-string

directional drilling. Figure 1 sketches a vertical and a horizontal drilling, where
BHA means Bottom Hole Assembly.
The safety of structures might be assessed if uncertainties are taken into account,
and there are many ways to take into account uncertainties in a dynamical system,
such as stochastic models and fuzzy randomness [9, 10]. This chapter describes
stochastic tools that are used to assess the reliability of the structure [2, 6, 11, 19].
Since the number of Monte Carlo simulations to assess small failure probabilities
might be big, importance sampling is applied to diminish the simulations required
for a fixed precision.
There are many works in the literature concerned with the modeling of the drill-
string dynamics in a vertical well [3, 4, 7, 13, 21], while some works have been
devoted to the stochastic analysis of the drill-string dynamics [14, 16, 18, 20]. There
are few articles in the literature dealing with the horizontal drill-string dynamics.
Ritto et al. [17] use a simplified model for the system and propose a stochastic
model for the frictional coefficient: a random field with exponential autocorrelation
function. The present chapter starts from this computational model to propose a
reliability analysis.
The drill-string is modeled using a bar model (tension/compression) and is
discretized by means of the finite element method. An oscillatory force (due to mud
motor) is imposed on the system and there is a bit-rock interaction as the column
moves forward. A failure occurs if, for instance, the stress limit is exceeded or a
crack is initiated. The present chapter considers a failure if the efficiency of the
system (defined by output/input power ratio) is below a target value. This result can
further be used for robust optimization [15].
The article is organized as follows. The deterministic and stochastic system are
presented in Sects. 2 and 3. There, the equations of the system and of the stochastic
model are presented. Then, the efficiency measure is defined in Sect. 4, where the
measure of the efficiency relates output and input power. The proposed reliability
Reliability Analysis of the Dynamics of a Horizontal Drill-String 283
Fig. 2 Sketch of the system analyzed
analysis, which takes into account the efficiency of the system, is discussed in
Sect. 5, and the numerical results are shown in Sect. 6. Finally, the concluding
remarks are made in Sect. 7.
2 Deterministic System
The present model [17] considers a bar with axial vibration only. A sketch of the
system analyzed is shown in Fig. 2. Weight, friction force, and bit-rock interaction
force are some of the forces taken into account in the modeling.
Only a part of the horizontal drill-string, called Bottom Hole Assembly (BHA),
is considered, and the equation of motion is given by
@2 u.x; t / @2 u.x; t /
A 2
EA Dfsta .x; t /Cfhar .x; t /Cfbit .Pu.x; t //Cffric .Pu.x; t // ,
@t @x 2
(1)
where u is the axial displacement, is the mass density of the material of the
column, A is the cross-sectional area, and E is the elasticity modulus. The space and
time variables are x 2 Œ0; L, t 2 Œ0; T , where L is the length of the structure and T
is the duration of the time analysis. The right-hand side of the equation presents the
forces per unit length acting on the system. fsta is related to the static force imposed
at x D 0, and fhar is the harmonic force imposed by the mud motor at x D L.
The force related to the friction field (Coulomb friction model) is given by
ffric .Pu.x; t // D .x/.A/g sgn.Pu.x; t // ; (2)
where is the friction coefficient and sgn is the sign function. And the force related
to the bit-rock interaction fbit is given by

fbit .Pu.x; t // D Œc1 exp .c2 uP .x; t // c1 ı.x L/
(3)
fbit .Pu.x; t // D 0 for uP .L; t / 0 ,
where the first expression is used if uP .L; t / > 0, c1 and c2 are the two constants of
the bit-rock interaction, and this force is applied at x D L. The discretized equations
of the deterministic system, using linear shape functions, are written as
R / C Cu.t
Mu.t P / C Ku.t / D fsta C fhar .t / C fbit .u.t
P // C ffric .u.t
P // , (4)
with the appropriate boundary and initial conditions. In the above equation, u
is the displacement vector, M is the mass matrix, and K is the stiffness matrix.
The proportional damping matrix C D ˛ M (˛ is a positive constant) is added
a posteriori to the computational model. The force vectors fsta , fhar , fbit , and
ffric are related to the static, harmonic, bit-rock interaction, and frictional forces,
respectively. The force related to the bit mass is included in the matrix M, since it
depends linearly on the acceleration. More details of the model can be found in [17].
3 Stochastic Systems
It is assumed a random frictional coefficient field because of the difficulty to model

the friction details. Thus, the random field f.x/ W x 2 Œ0; Lg is defined as a
collection of real-valued random variables from a probability space .˝; F ; P /,
where ˝ is the sample space, F is the -algebra, and P is the probability measure.
It is also assumed that is a stationary truncated Gaussian random field on Œ0; L
with exponential autocorrelation function

jx2 x1 j
R.x1 ; x2 / D 2 exp ; (5)
b
where b is the correlation length, which measures the decay of the autocorrelation
function. The stochastic field is truncated such that 0 .x/ 1.
Finally, it is assumed that the truncated Gaussian field is expanded with
Karhunen-Loève expansion [8] using standard Gaussian random variables. As long
as the probability of having .x/ … Œ0; 1 is very low, this approximation will be
good (which is the case of the present analysis):
X
N p
.x; / D .x/ C k Zk ./k .x/ ; (6)
kD1
where is the mean value of the frictional coefficient, k and k are the
kth eigenvalue and kth eigenvector of the autocorrelation function R, Zk are
independent standard Gaussian random variables, and N defines the precision of this
representation (as N increases the representation gets better). The random frictional
force (per unit length) is written as
Ffric .Pu.x; t /; / D .x; /.A/g sgn.Pu.x; t // : (7)
And the stochastic system is written as
R / C CU.t;
MU.t; P / C KU.t; / D fsta C fhar .t / C fbit .U.t;
P // C Ffric .U.t;
P /; /
(8)
where U is the random response and Ffric is the random vector related to the
frictional random field. This is a nonlinear system of equations because the bit-rock
interaction is nonlinear and because the frictional force is discontinuous (if the speed
at a point changes sign, the friction force of this point also changes sign).
4 Efficiency Measure
The efficiency of the horizontal drilling is analyzed in the following way. We estab-
lish a measure that is given by the ratio between output and input power; if this
parameter increases, the efficiency of the process also increases. The input and
output power are given by
pin .t / D fsta uP .0; t / C fhar .t /Pu.L; t / (9)
and
pout .t / D fbit .t /Pu.L; t / : (10)
And the average power is given by

Z t1
1
p in D .fsta uP .0; t / C fhar .t /Pu.L; t //dt
t1 t0 Zt0
t1 (11)
1
p out D fbit .t /Pu.L; t /dt .
t1 t0 t0
Let us define the ratio between the output and input power in the time interval
.t0 ; t1 / by y D p out =p in . When the stochastic system is considered, this ratio
becomes a random variable Y .
5 Reliability Analysis
The reliability analysis aims to evaluate the probability of failure, i.e., the probability
that the system response does not satisfy a performance criterion. The functional
relationship between the performance criterion and the random variables can be
expressed as a performance function as follows:
Z D g.X/ ; (12)
where X is a vector of random variables. In the present analysis, the elements of X

are the random variables related to the discretized random field presented in Sect. 3.
The failure surface or the limit state is defined when Z D 0, and the failure occurs
when Z < 0. The probability of failure is given by (Schueller et al., 2004)
Z
PF D fX .x/d x ; (13)
g.X/<0
in which fX is the joint probability density function of X. The result of the

integration shown in Eq. 13 is the probability of failure of the system under
analysis. In general, the integration above is not easily obtained. The standard
Monte Carlo method is a non-intrusive and well-suitable method to calculate the
failure probability, where the failure probability is calculated from the sum of all
the realizations for which a failure occurs among N runs, as follows:
b 1 X
N

PF D 1g.X/<0 ; (14)
N iD1
b
where P F is an unbiased estimator of PF , Eq. 13, and 1g.X/<0 is an indicator that
is equal to 1 when g.X/ < 0 and 0 otherwise. Eq. 14 means that the probability of
b
failure can be estimated as a ratio of the number of runs that lead to failure and the
number of the total runs performed, P F D nf =ns . By the Central Limit Theorem,
b
P F has approximately a normal distribution Normal.PF; N 1 Var.PF // for
b
large N , where Var.PF / can be estimated via the sample variance unbiased
P
estimator, S 2 D .1=.N 1// N iD1 .PFi P F /, with which we can estimate a
confidence interval. When PF is small, a large number of Monte Carlo simulations
is necessary for convergence. For instance, if PF D 103 an average of 1000
simulations is necessary for a failure to happen.
An alternative to Monte Carlo is a technique called importance sampling,
known as the fundamental variance reduction techniques. It involves choosing a
sampling distribution fX that favors important samples, improving the Monte Carlo
convergence. Equation 14 is modified as follows:
Z
PF D fX .x/d x D
g.X/<0
Z (15)
fX .x/
D f .x/d x D Ef fWX g ,
g.X/<0 fX .x/ X
where WX D fX =fX is called the likelihood ratio. The optimal sampling distribution
can be calculated, but it is necessary to know a priori what we want to estimate. The
probability of failure defined in Eq. 15 can be estimated using the standard Monte
Carlo sampling, but weighting each failure run with WX , as follows:
b 1 X
N

PF D WX : 1g.X/<0 : (16)
N iD1
6 Numerical Results
The values of the parameters used in the simulations are the following: E D 210
GPa, D 7850 km/m3 , g D 9:81 m/s2 , Di D 0:10 m (inner diameter), Do D 0:15
m (outer diameter), L=Do =400, mbit D 20 kg, ˛ D 0:2, c1 D 1:4103 N, c2 D 400,
D 0:1, D 0:1 , b D 10 and !f D 100 2=60 rad/s, t 2 Œ0; 10s,
t D 0:0001s, N D 100, fsta D 5500N, and Fo D 550N where mbit is the mass
attached at x D L to model the bit. Free-free boundary conditions are considered,
such that the column can rigidly move to the right. The system is discretized with
40 finite elements and the system of differential equations is integrated applying the
Newmark integration scheme assuming zero initial condition.
First, the deterministic responses are analyzed, to give an idea of the system
dynamics. Figure 3 shows the bit speed evolving in time. The mean speed is around
20 meters per hour, but it oscillates from 4 to 34 meters per hour. Figure 4 shows
the input power divided by the output power; this ratio also oscillates in time. For
the present case, the efficiency is about 15%. Since there are losses not considered
in the model, such as lateral and torsional vibrations, this value can be thought of as
an upper bound for the efficiency.
Now, the stochastic analysis. Figure 5 shows an approximation of the probability
density of the output/input power ratio, called Y . It presents a bimodal distribution,
where the higher peek is related to situations where the rock is so easy to drill that
the bit speed increases with time.
Figure 6 shows 1,000 Monte Carlo (MC) realizations of Y and a target efficiency
of 10%. It can be observed that few realizations are under the target value, which
means that a lot of Monte Carlo simulations will be needed to estimate the
probability that Y is under 10%. This probability is the failure probability, or the risk
of the operation.
35
30
bit speed (m/h)
25
20
15
10
0
5 6 7 8 9 10
time(s)
Fig. 3 Bit speed

0.35
0.3
0.25
0.2
pout/pin
0.15
0.1
0.05
0
4 6 8 10
time(s)
Fig. 4 Output power divided by the input power
14
12
10
8
pdfY
0
0 0.1 0.2 0.3 0.4
y
Fig. 5 Approximation of the probability density function of random variable Y
Since the present application has only one random field with constant mean
taken into account, a simple way to perform importance sampling is to use as
sampling probability density function fX the original distribution with the mean
shifted, such that failures occur more often, say around 50%. A quick test with
few Monte Carlo simulations can be done to define the size of the shift. Figure 7
shows 1,000 importance sampling (IS) realizations, where the proposed probability
density function of the random field has a mean value multiplied by 1.1 and
same covariance function. It is noted that many more points are under 10% now,
improving the convergence of the estimate. Figure 7 shows the convergence of
the failure probability for different target values, using Monte Carlo simulations.
a
0.35
target efficience
0.3
0.25
0.2
target
0.15
0.1
0.05
0
0 200 400 600 800 1000
y
b
0.35
0.3
0.25
0.2
target
0.15
0.1
0.05
0
0 200 400 600 800 1000
y
Fig. 6 Samples of the MC strategy (a) and of the IS strategy (b)
For target values of 15% or 20%, the convergence is reasonable (using 1000
simulations), but for target value of 5%, the convergence is unacceptable.
To avoid too many simulations, the failure probability is estimated using con-
veniently MC or IS, such that the convergence is reasonable for 1000 realizations.
For lower target values, the proposed IS probability density function of the random
field has mean value multiplied by 1.2, to guarantee a reasonable convergence of the
estimate.
Finally, we want to propose the construction of a map that takes into account
efficiency and probability of failure thresholds to define a safe region to operate.
Figure 8 shows this map for two different cases. An upper bound of 5 % is assumed
for the value of failure probability, and a lower bound of 5 % is assumed for the value
a
0.8
0.7
0.6
target=0.05
0.5 target=0.10
target=0.15
0.4
Pf
target=0.20
0.3
0.2
0.1
0
0 200 400 600 800 1000
b ns
x 10−3
18
16
14
12
10
Pf
8
6
4
2
0
200 400 600 800

ns
Fig. 7 Convergence of POf for different target values (a) and zoom of the lower curve (b)
of the efficiency. These limits are in dashed line in Fig. 8. It should be noticed that
when the target value increases, the failure probability also increases, meaning that
if we impose a higher target efficiency value, the risk of not performed as desired
increases.
Figure 8a shows a comparison of the system with 50RPM and 100RPM
excitation frequency at the bit. It is noted that the response is almost the same.
This is due to the fact that these excitation frequencies are far away from the natural
frequencies of the structure and, also, the computational model does not consider
the rotation of the bit. If torsional vibrations are included in the model, we would
expect, in a regular operation, an increase in the efficiency of the system.
a
0.06
0.05
0.04
Pf
0.03
0.02
0.01
100RPM
50RPM
0
0.04 0.05 0.06 0.07 0.08 0.09 0.1
target efficiency
b
0.06
0.05
0.04
Pf
0.03
0.02
0.01 c1=1400
c1=2800
0
0.04 0.05 0.06 0.07 0.08 0.09 0.1
target efficiency
Fig. 8 Map with the safe region, taking into account efficiency and probability of failure
thresholds. (a) Different frequency of excitation, (b) different value of c1 (related to the bit-rock
interaction)
Figure 8b shows the a comparison of the system with two different values of the
bit-rock interaction constant c1 D 1:4 103 N and c1 D 2:8 103 N. It is noted
that the system performs better when c1 increases. This is explained by the fact that
the bit force is significantly bigger for the same bit speed, when c1 is increased, as
shown in Fig. 9.
0
c1=1400
−500 c1=2800
Fbit (N) −1000
−1500
−2000
−2500
−3000
0 10 20 30 40
speed (m/h)
Fig. 9 Bit-rock interaction model for two values of c1
This chapter considered a simplified model to analyze the dynamics of a horizontal

drill-string. The focus of the present analysis was the reliability of the opera-
tion, which was defined by the probability of not achieving a target efficiency.
The efficiency measured was given by the mean of the input/output ratio.
To compute the probability of failure, importance sampling strategy was used
together with Monte Carlo simulations, such that a reasonable convergence of the
estimate is achieved with a small number of numerical simulations.
When the target values increase, the failure probability also increases, meaning
that if we impose a higher target value, the risk of do not perform as desired
increases. A map that takes into account efficiency and probability of failure
thresholds was proposed to obtain a safe region to operate.
Acknowledgements The authors are grateful to the financial support from the Brazilian agencies:
CAPES, CNPq, and FAPERJ.
References
1. Allouche, E., Ariaratnam, S., Lueke, J.: Horizontal directional drilling profile of an emerging
industry. J. Const. Eng. Manag. ASCE 26, 68–76 (2000)
2. Au, S.K., Beck, J.L.: Important sampling in high dimensions. Struct. Saf. 25(2), 139–163
(2003)
3. Christoforou, A.P., Yigit, A.S.: Dynamic modeling of rotating drillstrings with borehole
interactions. J. Sound Vib. 206(2), 243–260 (1997)
4. Divenyi S., Savi M.A., Wiercigroch M., Pavlovskaia E.: Drill-string vibration analysis using
non-smooth dynamics approach. Nonlinear Dynam. 70, 1017–1035 (2012)
5. Hu Y., Di Q., Zhu W., Chen Z., Wang W.: Dynamics characteristics analysis of drillstring in the
ultra-deep well with spacial curved beam finite element. J. Petroleum Sci. Eng. 82–83 166–173
(2012)
6. Jensen, H.A.: Reliability-based optimization of uncertain systems in structural dynamics.
AIAA J. 40(4), 731–738 (2002)
7. Khulief, Y.A., Al-Naser, H.: Finite element dynamic analysis of drillstrings. Finite Elem. Anal.
Des. 41, 1270–1288 (2005)
8. Loève, M.: Probability theory. Graduate Texts in Mathematics, Springer, USA (1977)
9. Moller, B., Graf, W., Beer, M.: Safety assessment of structures in view of fuzzy randomness.
Comput. Struct. 81(15), 1567–1582 (2003)
10. Moller, B., Beer, M.: Engineering computation under uncertainty-Capabilities of non-
traditional models. Comput. Struct. 86(10), 1024–1041 (2008)
11. Patelli, E., Pradlwarter, H.J., Schueller, G.I.: On multinormal integrals by Importance Sampling
for parallel system reliability. Struct. Saf. 33(1), 1–7 (2011)
12. Perneder, L., Detournay, E. Downton G.: Bit/rock interface laws in directional drilling. Int.
J. Rock Mech. Min. Sci. 51, 81–90 (2012)
13. Richard T., Germay C., Detournay, E.: A simplified model to explore the root cause of stick-slip
vibrations in drilling systems with drag bits. J. Sound Vib. 305, 43–456 (2007)
14. Ritto T.G., Soize C., Sampaio R.: Nonlinear dynamics of a drill-string with uncertain model of
the bit-rock interaction. Inter. J. Non-Linear Mech. 44(8), 865–876 (2009)
15. Ritto, T.G., Soize, C. Sampaio, R.: Robust optimization of the rate of penetration of a drill-
string using a stochastic nonlinear dynamical model. Computat. Mech. 45(5), 415–427 (2010)
16. Ritto, T.G., Sampaio, R.: Stochastic drill-string dynamics with uncertainty on the imposed
speed and on the bit-rock interaction parameters. Int. J. Uncert. Quant. 2(2), 111–124 (2012)
17. Ritto, T.G., Escalante, M.R., Sampaio, R., Rosales, M.B.: Drill-string horizontal dynamics with
uncertainty on the frictional force. J. Sound Vib. 32(1) 145–153 (2013)
18. Ritto, T.G., Sampaio, R.: Measuring the efficiency of vertical drill-strings: A vibration
perspective. Mech. Res. Comm. 52, 32–39 (2013)
19. Schueller, G.I., Pradlwarter, H.J., Koutsourelakis, P.S.: A critical appraisal of reliability
estimation procedures for high dimensions. Probabilist Eng. Mech. 19(4), 463–474 (2004)
20. Spanos, P.D., Chevallier, A.M., Politis, N.P.: Nonlinear stochastic drill-string vibrations. J. Vib.
Acoust., Trans. ASME 124(4), 512–518 (2002)
21. Tucker, R.W., Wang, C.: An integrated model for drill-string dynamics. J. Sound Vib. 224(1),
123–165 (1999)
22. Willoughby D.A.: Horizontal Directional Drilling, Utility and Pipeline applications. Civil
Engineering Series, McGraw-Hill, New York, USA (2005)
Transition in Oscillatory Behavior in Mouse
Oocyte and Mouse Embryo Through Oscillatory
Spherical Net Model of Mouse Zona Pellucida
Andjelka N. Hedrih
Abstract Zona pellucida, a 3D matrix that surrounds mammalian oocyte,

dynamically changes its elasticity during the maturation and fertilization process.
We consider fertilization as a biomechanical oscillatory phenomenon and
hypothesized that mature oocyte and embryo are in different oscillatory states.
Using an oscillatory spherical net model of mouse ZP [4], eigen circular frequencies
of mouse oocyte and mouse embryo were calculated. Frequency analysis of circular
frequencies of ZP under periodical external excitation force in the form of sperm
cell impact was done for both states. To determine the conditions for dynamical
absorption under impact of sperm cells on ZP of mouse oocyte and embryo,
numerical analyses were done.
1 Introduction
Zona pellucida (ZP), the outermost surface of the oocyte, dynamically changes its
elasticity during the maturation and fertilization process [1,6,7]. The purpose of this
structure is to protect the oocyte, to work as high selective structure for high-quality
spermatozoa, to select the “right one,” to ensure polyspermy block, and to guide the
embryo through the oviduct. The final fertilization process occurs at ZP. Considering
fertilization process as a biomechanical phenomenon we hypothesized that mature
oocyte [3,5] and embryo are in different oscillatory states. If they are in different
oscillatory states, what are oscillatory properties of mouse embryo that do not allow
penetration of other sperm cells through ZP in fertilized oocyte?
If there is only an initial perturbation by kinetic and potential energy given
to oscillatory structures, only free vibrations of ZP appear. In this case material
particles at the initial moment obtain the initial displacement measured from their
equilibrium positions and initial velocities. In order for free oscillations to appear,
it is enough that only one mass particle position is perturbed from its equilibrium
A.N. Hedrih ()

Department for Bio-medical science, State University of Novi Pazar,
Vuka Karadzica bb, 36 300 Novi Pazar, Serbia
e-mail: handjelka@hm.co.rs

296 A.N. Hedrih
position or that only one mass particle at its equilibrium position obtains initial
velocity. If we apply one or multifrequency external excitation forces to a ZP
discrete net, it oscillates in multifrequency forced regime oscillations.
The aim of our study was to compare eigen circular frequencies of mouse oocyte
and mouse embryo ZP as a property of the systems and to determine the conditions
for dynamical absorption under impact of sperm cells on ZP of mouse oocyte and
embryo through spherical surface net model of mZP [4].
2 Determination of the Eigen Circular Frequencies
Using spherical surface net model of mZP [4] and a method of discrete continuum
[2] we considered that the system of ZP oscillatory net oscillates in a free regime
after ovulation without the presence of spermatozoa. In a spherical surface net model
of mZP [4] ZP is modeled as a one-layer network that envelops the oocyte/embryo
that we supposed that is solid, elastic, and rigid. The network consists of orthogonal
chains of material particles interconnect with elastic massless springs on a specific
manner. Material particles represent the ZP glycoproteins. See Fig. 1a. To do a
frequency analysis of a part of spherical net model and to determine a particular
set of the eigen circular frequencies of mZP, we use the smallest part of the mZP
oscillatory spherical surface model (see Fig. 1b) that still preserves the molar ratio
of the mZP glycoproteins (ZP1:ZP2–ZP3 is 1:5).
We applied one or multifrequency external excitation forces to a one knot
molecule ZP discrete net (third in the chain or ninth in this chain using symmetry
Fig. 1 (a) Model of ZP spherical surface that shows a radial direction of axis of constructive
elements of the model—ZP proteins. Axis shows directions of movements of ZP proteins. Each ZP
protein is connected to the sphere with elastic springs that can oscillate in radial direction. (b) Part
of the ZP network on a part of the sphere (oocyte). Orange (ZP1), blue (ZP2), and green (ZP3)
represent ZP proteins. The network is identical in both circular and meridian direction
Transition in Oscillatory Behavior in Mouse Oocyte and Mouse Embryo. . . 297
Fig. 2 Presentation of two characteristic branches of the eleven degree polynomial function graphs
along square of circular frequencies ! 2 : (a) before and (b) after fertilization of ZP for different
material parameters (module of elasticity and masses particles)
in order of eleven mass particles in chain). Frequency equation that describes the
oscillations of the mZP glycoproteins is in the form of chain with eleven material
particles. The frequency equation is an eleven degree polynomial function along
square of possible circular frequencies ! 2 and is in the form
ˇ ˇ
f ! 2 D ˇC ! 2 Aˇ D 0: (1)
C is matrix of coefficient of elasticity, and A is matrix of coefficient of mass iner-

tia. For obtaining eigen circular frequencies we used graphical method and Mathcad
software. See Fig. 2a, b. Relative molecular masses of the three mZP glycoproteins,
MrZP1, 200,000 Da; MrZP2, 120,000 Da; and MrZP3, 83,000 Da transformed in
kg, were used. The zeros (roots) in the frequency equations are squares of eigen
circular frequencies ! 2s . For considered nonhomogenous chains there are 11 squares
of eigen circular frequencies: ! 2s , s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11.
For our numerical experiment we approximate that coefficient of elasticity is
equal for all material particles and calculated it from the experimental data of
Sun et al. [9] according to the formula

E R2 r 2
cD (2)
2R
E is Young module of elasticity, R is half diameter of the mouse oocyte/embryo,
and r is internal radius of oocyte or is half diameter of the oocyte/embryo minus
approximate thickness of mZP.
For an oocyte: Eo D 17.5kPa; 2Ro D 56.2 m, average diameter of an oocyte
from [9]; and •o D 4.8 m, approximate ZP thickness of the oocyte was used for
calculating coefficient of elasticity. Its value is co D 0.253 N/m.
For mouse embryo: Ee D 42.2kPa; 2Re D 61 m, average diameter of an embryo
from [9]; and •e D 5.34 m, approximate ZP thickness of the embryo was used for
calculating coefficient of elasticity. Its value is ce D 0.646 N/m.
298 A.N. Hedrih
For obtaining eigen circular frequencies we used graphical method and Mathcad
software tool and correction factor 106 .
Using Matcad software tool and graphical method (See Fig. 2a) the obtained
eigen circular frequencies of mouse oocyte ! os , s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 for
corresponding material parameters (module of elasticity and masses particles) are:
! 2s , s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 before fertilization for corresponding material
parameters (module of elasticity and masses particles) are: ö1 D_ 9.318 109 rad/s,
ö2 D_ 1.72 1010 _ rad/s, _ ö3 D_ 2.707 1010 _ rad/s, ö4 D_ 3.856 1010 _ rad/s,
ö5 D_ 4.379 1010 rad/s, ö6 D_ 5.011 1010 rad/s, ö7 D_ 5.522 1010 rad/s,
ö8 D_ 6.887 10 _ rad/s, ö9 D_ 6.952 10 rad/s,_ ö10 D_ 7.024 1010 rad/s,
10 10
and ö11 D_ 7.645 1010 rad/s.

Using Matcad software tool and graphical method (See Fig. 2b) the obtained
eigen circular frequencies of mouse embryo ! es , s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
for corresponding material parameters (module of elasticity and masses
particles) are: sD1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 11 for corresponding material param-
eters (module of elasticity and masses particles) are: ë1 D_ 1.484 1010 rad,
ë2 D_ 2.725 1010 _ rad/s,_ ë3 D_ 4.324 1010 _ rad/s, ë4 D_ 6.205 1010 _ rad/s,
ë5 D _ 6.977 10 rad/s, ë6 D_ 8.005 10 rad/s, ë7 D_ 8.806 1010 rad/s,
10 10
ë8 D_ 1.096 1011 _ rad/s, ë9 D_ 1.193 1011 rad/s,_ ë10 D_ 1.127 1011 rad/s,
and ë11 D_ 1.221x 1011 rad/s.
These two obtained sets of square of eigen circular frequencies of mZP ! 2s ,
s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 before and after fertilization for corresponding mate-
rial parameters (module of elasticity and masses particles) are also sets of resonant
frequency squares of corresponding one frequency external excitation square of
frequencies ˝ 2rez,s D ! 2s , s D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11.
3 Forced Oscillations
The system of ordinary differential equations of mass particles forced dynamics is

in the following matrix form:
R C C fxg D fQg D fF0 g cos ˝t

A fxg (3)
Particular solution for kth mass particle in the chains, when one frequency
external excitation is applied to the third mass particle in chain, is in the form

xk t; ˝32 D C.3/k ˝32 cos ˝ 3 t (4)
Amplitude C(3)k (23 ) of particular solution for kth mass particle in chain forced
vibration displacement under the one frequency external excitation applied to the
third mass particle in chain is in the form

2 .3/k ˝32
C.3/k ˝3 D (5)
˝32
for 23 ¤ ! 2s , s D 1, 2, 3„ 4, 5, 6, 7, 8, 9, 10, 11, and where determinant of the previous

system is in the form
ˇ ˇ
˝32 D ˇC ˝32 Aˇ ¤ 0 (6)
and must be different from zero. Then the condition 23 ¤ ! 2s , s D 1, 2, 3, 4, 5, 6, 7,

8, 9, 10, 11 have to be fulfilled.
For obtaining amplitudes C(3)k (23 ) of particular solutions, it is necessary to find
sub-determinant (3) (23 ). Corresponding sub determinants (3) (23 ) is possible to
obtain by substituting corresponding 3-rd column in determinant of the system, by
the column containing amplitude of the external excitation force F03 .
4 Conditions for Dynamical Absorption
Dynamical absorption is present in the systems with multiple degrees of freedom (in
our case chain is system with 11 degrees of freedom) when external periodical force
is applied to the system. Depending on the material properties and parameters of
the oscillatory system structure (coefficient of elasticity, mass of material particles,
external excitation frequencies, amplitudes) it is possible that under the influence
of periodical external force one or more material particles don’t oscillate in forced
mode with external excitation frequency and that other mass particles are in this
forced oscillatory mode. In the theory of oscillation [8] this phenomenon is known
as dynamical absorption.
External excitation frequencies at which dynamical absorption occurs of mass
particles on certain positions were read out from amplitude-frequency graphs for
each mass particle in chain. For determine the amplitude of forced vibrations
Ck (23 ), k D 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 for each mass particle, when third mass
particle are loaded by external one frequency excitation with frequency 3 and
external periodical force F3 D F03 cos 3 t, the required condition is:

Ck ˝32 D 0; for some of k D 1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 11 when ˝32 ¤ 0
(7)
Graphs of amplitude-frequency stationary forced regimes for forced vibration

of C3 (23 ) third, C6 (26 ) sixth, and C9 (23 ) ninth material particle in chain excited
by external excitation F3 D F03 cos 3 t force with amplitude F03 and frequency ˝ 3
applied to third mass particle in chain in mZP net model of oocyte and embryo are
given in Figs. 3, 4, and 5, respectively. When external excitation F3 D F03 cos ˝ 3 t
force with amplitude F03 and frequency 3 is applied to third mass particle in chain,
frequencies under which dynamical absorption occurs in forced oscillatory regimes
in the third material particle oscillatory displacement in the chain are:
For mouse oocyte: 1.233 1010 rad/s, 2.49 1010 rad/s, 4.605 1010 rad/s, and
5.38 1010 rad/s.
300 A.N. Hedrih
Fig. 3 Amplitude-frequency stationary forced regimes for forced oscillations of C3 (23 ) third
material particle in chain excited by external excitation F3 D F03 cos 3 t force with amplitude
F03 and frequency 3 applied to third mass particle in chain; x D 2 . for (a) oocyte, (b) embryo
Fig. 4 Amplitude-frequency stationary forced regimes for forced oscillations of C3 (23 ) sixth
F03 and frequency 3 applied to third mass particle in chain; x D 2 .for (a) oocyte, (b) embryo
Fig. 5 Amplitude-frequency stationary forced regimes for forced oscillations of C3 (23 ) ninth
F03 and frequency 3 applied to third mass particle in chain; x D 2 . for (a) oocyte, (b) embryo.
It is visible that there is no dynamical absorption
And for the embryo: 2.145 1010 rad/s, 3.95 1010 rad/s, 7.335 1010 rad/s,
8.602 1010 rad/s, 1.22 1011 rad/s, and 1.118 1011 rad/s.
We can see that in the forced regime one mode of forced oscillatory displacement
of this third mass particle does not exist at a number of external excitation
frequencies—frequencies of dynamical absorptions upon that mass particle are
loaded. Other mass particles are not directly loaded and are in forced oscillatory
regimes at these frequencies of dynamical absorption of third loaded mass particle.
This is a dynamical paradox that this mass particle is loaded by external single
frequency force and that loaded mass particle is not in the forced oscillatory
regime and that other mass particles in chain oscillate in forced regimes at these
frequencies. Amplitude of external excitation has no influence on the appearance
of dynamical absorption. Only external excitation frequency in relation with chain
material properties (system structure properties like eigen circular frequencies,
coefficients of elasticity/rigidity, and masses of material particles) influenced the
appearance of dynamical absorption.
When external excitation F3 D F03 cos 3 t force with amplitude F03 and fre-
quency 3 is applied to third mass particle in chain, frequencies under which
dynamical absorption occurs for the sixth material particle in the chain are:
For mouse oocyte: 1.672 1010 rad/s and 1.596 1010 rad/s
For the embryo: 2.683 1010 rad/s and 8.655 1010 rad/s
Under these conditions there is no dynamical absorption on the ninth material
particle in the chain either in oocyte or in embryo.
5 Conclusions
Due to limited length of the paper we did not analyze the forced oscillations of knot
molecules in ZP net; we analyzed only the phenomenon of dynamical absorption for
certain molecules in the model. We did the analysis on a part of spherical surface net
of mZP in the system with finite number of degree of freedom. In the real system of
ZP there are a lot of molecules and almost indefinite eigen circular frequencies.
Young module of elasticity of mouse ZP in mature oocyte (before fertilization)
has 2.5 times higher value compared to ZP of mouse embryo (after fertilization) [9].
Sperm cells could not penetrate the ZP of embryo. From a biological point of view,
after fertilization occurs polyspermy block, a phenomenon where new sperm cells
could not attach to the ZP and the already attached have been rejected. This process
is receptor and enzymatic mediated and includes repulsive electric charge between
ZP and sperm cells.
302 A.N. Hedrih
We tried to perceive these events from the aspect of theory of oscillations.

Sperm cells have impact on a ZP in the form of multifrequency external excitation
forces. We applied external excitation to the one knot molecule ZP discrete net
(third in the chain). According to the spherical surface net model of mouse ZP, our
numerical analysis shows that eigen circular frequencies of mouse embryo ZP have
higher values compared to eigen circular frequencies of mouse oocyte ZP. This
indicates that for potential penetration through mouse embryo, sperm cells will
require more energy. In the living system this is not possible owing to sperm energy
loss with time. After fertilization there are no sperm cells capable to respond to these
increased demands.
On amplitude-frequency graphs zeros correspond to dynamical absorptions
in forced oscillatory regime. On these frequencies certain material particle is
not oscillating and external force has no effect on it. For the same molecules,
frequencies when dynamical absorption occurs are higher for ZP of mouse embryo
then for ZP of mouse oocyte. For sixth molecule in the chain in case of embryo there
are six frequencies where dynamical absorption is possible and for the case of the
oocyte there four. All these results confirm that oocyte and embryo are in different
oscillatory states and that these oscillatory states have biological purpose. As in
biological systems this transition is an irreversible process; further investigations of
conditions under which this irreversible process is possible should be done. If small
nonlinearities are included in the system it is possible to determine the dependence
of frequencies on initial conditions as well as resonant jumps in the system.
Acknowledgments The author would like to thank Prof. Katica (Stevanović) Hedrih from
the Mathematical Institute of SASA, Belgrade, Serbia, for the help with useful consultations.
Parts of this research were supported by the Ministry of Education, Science and Technological
Development of the Republic of Serbia through Grant ON174001 and State University of Novi
Pazar.
References
1. Familiari, G., Heyn, R., Relucenti, M., Sathananthan, H.: Structural changes of the zona
pellucida during fertilization and embryo development. Frontier Biosci. 13, 6730–51 (2008)
2. Hedrih (Stevanović), K.: Discrete continuum method, symposium, recent advances in analytical
dynamics control, stability and differential geometry. In: Proceedings of the Mathematical
institute SANU Edited by Vladan Djordjević, p. 151, pp. 30–57, 2002. ISBN 86-80593-32-X.
http://www.mi.sanu.ac.yu/publications.htm
3. Hedrih, A.: Frequency analysis of knot mass particles in oscillatory spherical net model of
mouse zona pellucida. In: Lecture Session, Short Paper, Abstract book of 23rd International
Congress of Theoretical and Applied Mechanics, (IUTAM ICTAM Beijing 2012), Beijing,
China, SM01-049, p. 209, 19–24 Aug 2012. ISBN 978-988-16022-3-7
4. Hedrih, A.: Modeling oscillations of zona pellucida before and after fertilization. ENOC Young
Scientist Prize Paper. EUROMECH Newsletter 40. Europian Mech. Soc. 40, 6–14. 2011
5. Hedrih, A., Lazarevic, M., Mitrovic-Jovanovic, A.: Fertilization as a biomechanical oscillatory

phenomenon in mammals. In: Proceedings of the Fourth Serbian (29th Yu) Congress on
Theoretical and Applied Mechanics, Vrnjacka Banja, Serbia, pp. 579–584, 4–7 June 2013
6. Murayama, Y., Mizuno, J., Kamakura, H., Fueta, Y., Nakamura, H., Akaishi, K., Anzai, K.,
Watanabe, A., Inui, H., Omata, S.: Mouse zona pellucida dynamically changes its elasticity
during oocyte maturation, fertilization and early embryo development. Hum. Cell 19, 119–125
(2006)
7. Papi, M., Sylla, L., Parasassi, T., Brunelli, R., Monaci, M., Maulucci, G., Missori, M., Arcovito,
G., Ursini, F., De Spirito, M.: Evidence of elastic to plastic transition in the zona pellucida of
oocytes using atomic force spectroscopy. Appl. Phys. Lett. 94, 153902 (2009)
8. Rašković, P.D.: Teorija oscilacija (Theory of Oscillations). Beograd, Naučna knjiga (1965)
9. Sun, Y., Wan, K.T., Roberts, K.P., Bischof, J.S., Nelson, B.J.: Mechanical property characteri-
zation of mouse zona pellucida. IEEE T Nanobiosci. 2(4), 279–286 (2003)
Constrained n-Body Problems
Wojciech Szumiński and Maria Przybylska
Abstract We consider a problem of mass points interacting gravitationally whose

motion is subjected to certain holonomic constraints. The motion of points is
restricted to certain curves or surfaces. We illustrate the complicated behaviour of
trajectories of these systems using Poincaré cross sections. For some models we
prove the non-integrability analysing properties of the differential Galois group of
variational equations along certain particular solutions of considered systems. Also
some integrable cases are identified.
1 Introduction
Let us consider several point masses interacting mutually according to a certain

low. This is just the n-body problem. For the classical gravitational or electrostatic
interactions such problem with n > 2 is not integrable. Let us restrict the motion of
points to certain surfaces or curves. These holonomic constrains modify interactions
of points. In some cases these modifications lead to the non-integrability, and in
others to the integrability. The described constrained classical n-body problems can
be considered as a source of toy models for testing various methods and tools
for studying dynamics of classical systems. In fact this paper arose from such
investigations. Several simple examples show that, in fact, one can meet interesting
and difficult problems investigating this kind of systems and moreover, such, let us
say, academic investigations give unexpected results.
To describe them let us recall the anisotropic Kepler problem which appears in
quantum mechanics of solid. It was thoroughly investigated by Guztwiller [5]. The
rescaled Hamiltonian of the problem is given by
1 2 1
H D p C p22 C p32 p ; (1)
2 1 x C .y 2 C z2 /
2
W. Szumiński () • M. Przybylska

Institute of Physics University of Zielona Góra, 65-417 Zielona Góra, Poland
e-mail: uz88szuminski@gmail.com; M.Przybylska@if.uz.zgora.pl

306 W. Szumiński and M. Przybylska
Fig. 1 Motion of two masses a b

on: (a) perpendicular axes
(geometry of model 1), and
(b) plane and perpendicular
axis (geometry of model 2)
where is a positive constant. For the two-degrees-of-freedom version of this

problem the Hamiltonian reads
1 2 1
H D p1 C p22 p : (2)
2 x C y 2
2
Unexpectedly, these systems can be considered as gravitational two-body prob-

lems with constraints. To see this, let us consider two masses, one mass moving
along a line and the second mass moving along a perpendicular line; see Fig. 1a.
The Hamiltonian of the system is the following:
p12 p2 Gm1 m2
H1 D C 2 p : (3)
2m1 2m2 x2 C y2
So, by a simple rescaling we obtain Hamiltonian (2). Similarly, let one mass moves
along a line, and the other moves in a plane perpendicular to this line; see Fig. 1(b).
The Hamiltonian has the form
p12 1 Gm1 m2
H2 D C .p 3 C p32 / p (4)
2m1 2m2 2 x 2 C y 2 C z2
and again its simple rescaling gives (1).

As we can see the Hamiltonians (2) and (1) differ from the Hamiltonians of
standard planar and spatial Kepler problem only in the parameter . For ¤ 0,
contrary to the standard Kepler problem, the force is not radial. The dynamics of
the anisotropic Kepler problem is dramatically different from that of the standard
Kepler problem.
Constrained n-Body Problems 307
The chaotic behaviour of the anisotropic Kepler problems was investigated in

numerous papers (see, e.g., [2, 3, 5]) and the non-integrability of the planar problem
was proved in [4] and for planar and spatial problem in [1]. The non-integrability
proof in [1] uses the differential Galois approach and the authors state that for
62 f0; 1g there are no meromorphic integrals besides the Hamiltonian itself.
But there is no written about meromorphic functions of what variables authors
say. If one considers meromorphic functions of coordinates and momenta, then
already Hamiltonian is not a meromorphic function; thus the system trivially is
not meromorphically integrable for all values of . Thus below we formulate these
theorems in a more precise way.
Theorem 1. Hamiltonian system defined by (2) is integrable in the Liouville
sense
p with first integrals which are meromorphic in .x; y; p1 ; p2 ; r/, where r D
x 2 C y 2 , if and only if 2 f0; 1g.
In the case when D 1 this system has two additional functionally independent
additional first integrals:
x
I1 D p2 x p1 y; I2 D p2 .p1 y p2 x/ C p I
x2 C y2
thus it is super-integrable.
The spatial anisotropic Kepler problem defined by (1) has an invariant subspace
defined by z D p3 D 0. In this subspace it coincides with the previous system.
Thus, the necessary conditions of the integrability are the same as for the previous
system.
Theorem 2. Hamiltonian system defined by (1) is integrable in the Liouville sense
p first integrals which are meromorphic in .x; y; z; p1 ; p2 ; p3 ; r/, where r D
with
x 2 C .y 2 C z2 /, if and only if 2 f0; 1g.
In the case when D 1 it coincides with the three-dimensional standard Kepler
problem, and it has the following first integrals:
r
c D r p; e Dpc ;
r
p
where r D .x; y; z/, p D .p1 ; p2 ; p3 /, and r D x 2 C y 2 C z2 . Among them one
can find three functionally independent and pairwise commuting.
Hamiltonian (2) (and also (1)) because of the presence of square root r is not
single-valued and meromorphic in coordinates and momenta. Thus, formally, in
order to apply the differential Galois theory approach to such a Hamiltonian system,
we have to extend it to the corresponding Poisson system introducing r as additional
variable. However, in calculations one can work with the original Hamiltonian
system, and the only trace of this extension is the fact that we study the integrability
in the class of meromorphic functions of not only coordinates and momenta but
also of r. This extension procedure as well as its application to a certain three-body

problem was given in [7]. The similar trick is applied to all remaining Hamiltonian
systems with algebraic potentials considered in this paper.
The above examples show that it is reasonable to examine similar classes of
constrained n-body systems. In the next section we will give several examples
of such systems with a few degrees of freedoms. In a case when the considered
system reduces to a system with two degrees of freedom the Poincaré cross
sections give us quickly insight into the dynamics of the systems. However, a
challenging problem is to prove that they are non-integrable and to find values of
parameters for that they become integrable. For some presented problems we prove
their non-integrability using the so-called Morales–Ramis theory [8]. It is based
on the analysis of differential Galois group of variational equations obtained by
linearisation of equations of motion along a particular solution. The main theorem
of this theory states that if the considered system is meromorphically integrable in
the Liouville sense, then the identity component of the differential Galois group
of the variational equations is Abelian. For a precise definition of the differential
Galois group and differential Galois theory, see, e.g., [9].
2 Integrability Analysis of Several Restricted

n-Body Problems
Model 3: Two Masses on Two Inclined Straight Lines
The direct generalisation of model 1 from Fig.1a is the following. Assume that
mass m1 moves along horizontal line q2 D 0 and it has coordinates .q1 ; 0/ and
mass m2 with coordinates q2 .cos ; sin / moves along a straight line inclined to
the horizontal line: see Fig. 2. The Hamiltonian function is given by
p12 p2 Gm1 m2
H D C 2 q : (5)
2m1 2m2 q1 C q22 2q1 q2 cos
2
Fig. 2 Geometry of model 3

In the Appendix we will prove the following theorem:

Theorem 3. The system governed by Hamiltonian (5) q is integrable in the class of
functions meromorphic in .q1 ; q2 ; p1 ; p2 ; r/ where r D q12 C q22 2q1 q2 cos , ff
• either 2 f0; g and m1 ; m2 2 R, or

• 2 f=2; 3=2g and m2 D m1 .
Model 4: Masses Moving on the Parallel Lines
Let us consider a problem of n masses moving in parallel lines; see Fig. 3. As a

generalised coordinates we use the relative displacements qi D xi xi1 along axis
x, for i D 2; : : : ; n and q1 D x1 .
The Lagrange and the Hamiltonian functions do not depend on variable q1 , which
is a cyclic variable and its corresponding momentum p1 D c becomes a parameter.
Thus, we obtain the reduced system with n 1 degrees of freedom. Model of n D 2
masses is integrable. The reduced system with n D 3 masses has two degrees of
freedom and it is described by the following Hamiltonian:
!!
.cp2 /2 .p2 p3 /2 p2
H D 12 m1
C m2
C m33 p 2Gm2 m2 3 2 Cm1 p2Gm2 q 2Gm3
:
.ab/ Cq 3 a2 Cq22 b 2 C.q2 Cq3 /2
(6)
We assumed that masses m2 and m3 move along horizontal curves y D a and y D b,

respectively. Figure 4 shows the Poincaré cross sections related to (6). Clearly, the
system is generally non-integrable. However, a proof of this fact is an open question.
Fig. 4 Poincaré sections for model 4. Parameters: m1 D 1; m2 D 2; m3 D 1; G D 1; a D 3;

b D 1; c D 0; cross-plain q2 D 0; p2 > 0. (a) E D 2:25; (b) E D 2:1
Model 5: Two Masses Moving on an Ellipse and a Straight Line

Parallel to the Main Axis of the Ellipse
In Fig. 5 the geometry of the system is shown. Now we assume that the mass m1
moves on the ellipse with coordinates .cos ; sin /, where D c=.1 C e cos /;
and mass m2 moves along a straight line parallel to the main axis of ellipse with
coordinates .x; a/. The Hamiltonian function is given by
0 1
1B 2 p2 .1 C e cos / 4 C
H D B px C r
2Gm1 m2 C
@
2 m2 c 2 m1 1 C e 2 C 2e cos 2 2 A :
c cos c sin
1Ce cos x C 1Ce cos a
(7)
Figure 6 shows the Poincaré cross sections. They present that for certain fixed
values of parameters, the system is not integrable. In fact we can prove the following
theorem:
Theorem 4. If a D 0 and .m1 m2 /m1 m2 ¤ 0 then the system governed
by Hamiltonian (7) is not completely integrable with first integrals which are
Fig. 6 Poincaré sections for model 5. Parameters: m1 D 1; m2 D 2; G D 1; a D 3,

c D 2; e D 0:5; cross-plain x D 0; px > 0. (a) E D 1; (b) E D 0:5
a b
Fig. 7 Motion of two masses on: (a) two confocal ellipses (geometry of the model 6), and (b) two
concentric ellipses with parallel main axes (geometry of the model 7)
meromorphic in .x; ; p1 ; p2 ; r/, where

s
2 2
c cos c sin
rD x C :
1 C e cos 1 C e cos
This theorem is in particular true for the circle when e D 0 and c D .
Model 6: Two Mass Points Moving in Two Conics
In Fig. 7a the geometry of the system is presented. In this case, masses m1

and m2 move along two confocal ellipses with coordinates 1 .cos 1 ; sin 1 / and
2 .cos 2 ; sin 2 /, where
c1 c2
1 D ; 2 D ;
1 C e1 cos 1 1 C e2 cos 2
Fig. 8 Poincaré sections for model 6. Parameters: m1 D 2; m2 D 2; G D 1; c1 D 1;

c2 D 2; e1 D 12 ; e2 D 32 ; cross-plain 1 D 0; p1 > 0. (a) E D 4:4; (b) E D 1:8
Fig. 9 Poincaré sections for model 6. Parameters: m1 D 2; m2 D 1; G D 1; c1 D 3;

c2 D 1; e1 D 35 ; e2 D 0; cross-plain 1 D 0; p1 > 0. (a) E D 0:7; (b) E D 0:5
and interact gravitationally. Hamiltonian function takes the form
.1 C e1 cos 1 /4 p12 .1 C e2 cos 2 /4 p22 Gm1 m2

H D C ;
2c1 m1 2e1 cos 1 C e1 C 1
2 2
2c2 m2 2e2 cos 2 C e2 C 1
2 2 B
s
2 2
c1 cos 1 c2 cos 2 c1 sin 1 c2 sin 2
BD C :
1 C e1 cos 1 1 C e2 cos 2 1 C e1 cos 1 1 C e2 cos 2
(8)
To present the dynamics of considered system we make several Poincaré cross
sections; see Figs. 8–9.
Fig. 10 Poincaré sections related to model 7. Parameters: m1 D 1; G D 1; a1 D 0:8; a2 D

1:1; b1 D 1; b2 D aa2 b1 1 D 1:4; m2 D ma12a1 D 0:73; cross-plain 1 D 0; p1 > 0 (a) E D 0:82;
(b) E D 0:8
Model 7: Two Masses Moving in Concentric Ellipses

with Parallel Main Axes
The geometry of the system is shown in Fig. 7b. In this case, masses m1 and m2
move in two ellipses which have common centres and parallel main axes. Using the
standard trigonometric parametrisations of points on ellipses .ai cos i ; bi sin i /
for i D 1; 2, we can derive the Hamiltonian:

1 p12 p22
H D C
2 a12 m1 cos 12 C a22 m1 sin 12 b12 m2 cos 22 C b22 m2 sin 22
! (9)
2Gm1 m2
p ;
.a2 cos 1 b2 cos 2 / 2 C .a1 sin 1 b1 sin 2 / 2
where a1 ; a2 and b1 ; b2 are major and minor semi-axes of both ellipses. The Poincaré
cross sections are shown in Fig. 10.
Model 8: n-Masses Moving in the Circles
Let us consider the motion of n-masses moving on the concentric circles which
interact gravitationally. As a generalised coordinate we use the relative angles
i ;
see Fig. 11. Similarly to the fourth model the Hamiltonian function has one cyclic
variable
1 and its corresponding momentum p1 D c is a first integral of the system.
Thus, we get the reduced system with n 1 degrees of freedom. Case of two masses
is of course integrable, but the model of n D 3 has much more complex dynamics.
Fig. 12 Poincaré sections for model 8. Parameters: m1 D 1; m2 D 2; m3 D 1; G D 1;

1 D 2; 2 D 3; 3 D 1; c D 0; cross-plain
2 D 0; p2 > 0. (a) E D 3:5; (b) E D 2:8
To present this complexity we make several Poincaré sections; see Fig. 12. The
Hamiltonian of this reduced system has the form
0
1 B .c p2 / 2 .p2 p3 / 2 p2 2Gm1 m2
H D @ 2
C 2
C 23 q
2 1 m1 2 m2 3 m3 1 C 22 21 2 cos
2
2
1 (10)
2Gm2 m3 2Gm1 m3 C
q q A:
22 C 32 22 3 cos
3 12 C 32 21 3 cos .
2 C
3 /
Acknowledgements The authors are very grateful to Andrzej J. Maciejewski for many
helpful comments and suggestions. This research has been supported by grant No. DEC-
2011/02/A/ST1/00208 of the National Science Centre of Poland.
Appendix: Proof of Theorem 3
Hamilton equations for Hamiltonian (5) have the form
p1 p2 Gm1 m2 .˛q2 q1 / Gm1 m2 .˛q1 q2 /

qP 1 D ; qP 2 D ; pP1 D 2 ; pP2 D 2 ;
m1 m2 q1 C q22 2˛q1 q2 3=2 q1 C q22 2˛q1 q2 3=2
(11)
where ˛ WD cos . In order to simplify the calculations, we make the following

non-canonical transformation:
2 p 3
ˇ 2 22 .1 C2 /
2 3 p 1 p p 0 0 2 3
q1 6 ˇ2 2 2 ˇ2 21 22 / 7 x1
6 7
6 q2 7 6 0 1 0 0p 7 6 x2 7
6 7D6 p 76 7;
4p1 5 6 2 2 2
2 74
y1 5
6 0 0 p1 2
ˇ
p 2 1 2 7
p2 4 2 ˇ 2 2
2 2 ˇ 2 . C / 5
1 2 y2
0 0 0 1
q
1 D m1 C m2 ; 2 D m2 m1 ; ˇ D .m1 m2 /2 C 4m1 m2 ˛ 2 :
(12)
System (11) after this transformation takes the form
Gx1 .1 ˇ/ .2 ˇ/ 3

xP 1 D y1 ; yP1 D r ;
ˇ 2 22 2ˇx22 .ˇC1 /.ˇ2 /
2 2x2 x1 .ˇ C 1 / 2 2 C 1 2
C x1
2 3=2
1 2
q
2y2 G.ˇ2 /2 x1 .21 22 /.ˇ 2 22 /.1 C2 /x2 .ˇC1 /.ˇ2 /
xP 2 D ; yP2 D s ! :
1 C 2 4 2x2 x1 .ˇC1 /
ˇ 2 22 2ˇx22 .ˇC1 /.ˇ2 / 2 3=2
2
1 22C 1
2
Cx 1
(13)
˚
It has invariant manifold N D .x1 ; x2 ; y1 ; y2 / 2 C4 j x1 D y1 D 0 and its restric-
tion to N is
2y2 G .1 C 2 / Œ.1 2 / .ˇ 2 / 3=2

xP 2 D ; yP2 D p p : (14)
1 C 2 8 2x22 ˇ 3 .ˇ C 1 /
Let the particular solution of (13) be defined by (14), and Z D ŒX1 ; X2 ; Y1 ; Y2 T

denotes the variations of Œx1 ; x2 ; y1 ; y2 T . Then, the variational equations along this
particular solution have the form ZP D AZ , where
2 3
0 0 1 0
6 0 0 0 2 7
6 1 C2 7
AD6
6 G. ˇ/.1 2 /3=2 .2 ˇ/3
p 1 0 0 0
7:
7
4 4 2ˇ 3=2 x23 .ˇC1 /3=2 .ˇ
p 2/ p
3=2
5
G.ˇC31 /.ˇ2 /.1 2 /2 ˇC2 1 C2 G.ˇ2 /3=2 .1 2 /3=2 .1 C2 /
p
5=2 3 .ˇC /3=2
p p 0 0
16 2ˇ x2 1 4 2ˇ 3=2 x23 ˇC1
Equations for X1 and Y1 form a subsystem of normal variational equations and can
be rewritten as a one second-order differential equation for variable X X1
!
G .1 2 / 3=2 .ˇ 1 / .ˇ 2 / 3=2
XR C p X D 0: (15)
4 2ˇ 3=2 x23 .ˇ C 1 / 3=2
We transform this equation using the following change of independent variable:

p q ˇ.ˇC1 /
4 2E .1 2 /.ˇ2 /
t ! z D 2 x2 .t /; (16)
G 1 22
where E is a level of Hamiltonian transformed by means of (12) and restricted

to N . Then normal variational equation (15) takes the form
1 1 ˇ C 1 ˇ 1
X 00 C pX 0 C qX D 0; pD C ; qD C ;
2z 2.z 1/ 2.ˇ C 1 /z2 2.ˇ C 1 /.z 1/z
(17)
where 0 d
dz
. We recognise that this equation is a Riemann P equation (see, e.g.,
[6, 8])

d2 X 1 a a0 1 c c0 dX aa0 cc 0 bb 0 aa0 cc 0
C C C C C X D 0;
dz2 z z1 dz z 2 .z 1/ 2 z.z 1/
(18)
with exponents
q q
aD 1
4 3C 1C 16ˇ
ˇC1 ; a0 D 1
4 3 1C 16ˇ
ˇC1 ; b D c 0 D 0; b 0 D 1; c D 12 :
(19)
The differences of exponents are given by

s
1 0 17 C 1 1
Daa D ; D b b 0 D 1; D c c0 D ; (20)
2 C1 2
where D ˇ=1 . The Riemann P equation is solvable iff one of the four numbers
C C , C C , C , and C is an odd integer or or
and or and or belong (in an arbitrary order) to the so-called Schwarz

table [6, 8]. Conditions ˙ C C D 2p C 1, where p 2 Z, give the following
expression for
3 C 5p 2p 2
D ;
1 5p C 2p 2
that takes only two non-negative values 0 and 1. Similarly, conditions C D

2p C 1 and C D 2p 1, where p 2 Z, give
p 2p 2 1 C 3p 2p 2
D ; D ;
2 p C 2p 2 1 3p C 2p 2
respectively, that only take two non-negative values 0 and 1.

Since two differences of exponents are equal to 1/2 and 1, only the first case in
the Schwarz table is admissible that leads to the condition D 1=2 C p, where
p 2 Z. It gives
p C p2
D ;
4 C p C p 2
and this expression takes only two non-negative values 0 and 1. Value D 1 gives
˛ D cos D ˙1, and that implies 2 f0; g. Parameter vanishes only when
ˇ D 0 that gives m2 D m1 and simultaneously ˛ D cos D 0. These are the
only cases when the identity component of differential Galois group of Riemann P
equation (18) with exponents (19) is solvable that is necessary for its Abelianity and
the integrability of the system.
References
1. Arribas, M., Elipe, A., Riaguas, A.: Non-integrability of anisotropic quasi-homogeneous

Hamiltonian systems. Mech. Res. Comm. 30(3), 209–216 (2003)
2. Casasayas J., Llibre J.: Qualitative Analysis of the Anisotropic Kepler Problem. Memoirs of
Amer. Math. Soc. 52(312), 1–115 (1984)
3. Devaney R. L.: Blowing up Singularities in Classical Mechanical Systems. Amer. Math.
Monthly. 89, 535–552 (1982)
4. Gutzwiller M.C.: Bernoulli sequences and trajectories in the anisotropic Kepler problem.
J. Math. Phys. 18(4), 806–823 (1977)
5. Gutzwiller M.C.: Chaos in Classical and Quantum Mechanics. Springer, New York (1990)
6. Kimura T.: On Riemann’s equations which are solvable by quadratures. Funkcial. Ekvac. 12,
269–281 (1970)
7. Maciejewski A.J., Przybylska M., Simpson L., Szumiński W.: Non-integrability of the dumbbell
and point mass problem. Celestial Mech. Dynam. Astronom. 117, 315–330 (2013)
8. Morales Ruiz J.J.: Differential Galois Theory and Non-Integrability of Hamiltonian Systems.
Birkhäuser, Basel (1999)
9. Van der Put M, Singer M.F.: Galois Theory of Linear Differential Equations. Springer, Berlin
(2003)
Hunting French Ducks in Population Dynamics
Ferdinand Verhulst
Abstract Equations with periodic coefficients for singularly perturbed growth

can be analysed by using fast and slow timescales in the framework of Fenichel
geometric singular perturbation theory and its extensions. The analysis is restricted
to one-dimensional time-periodic ordinary differential equations and shows the
presence of slow manifolds, canards and the dynamical exchanges between several
slow manifolds. There exist permanent (or periodic) canards and periodic solutions
containing canards.
1 Introduction
In this note we consider systems with slow-fast motion in a singularly perturbed

setting; the slow motion is characterised by the exponential closeness of solutions
to slow manifolds. In the case that the solution moves along a stable slow invariant
manifold and at some point the slow manifold becomes unstable, we have the
possibility of ‘exponential sticking’ or canard (French duck) behaviour. In this case,
the solution continues for an O.1/ time along the slow invariant manifold that has
become unstable and jumps after that away, for instance, to the neighbourhood of
another invariant set. Following Pontrjagin (see [10]), one also calls this ‘delay of
stability loss’.
This delay or sticking process is closely connected to the so-called canard
phenomenon for differential equations that can be described as follows: Canard
solutions are bounded solutions of a singularly perturbed system that, starting near
an normally hyperbolic attracting slow manifold, cross a singularity of the system of
differential equations and follow for an O.1/ time a normally hyperbolic repelling
slow manifold.
F. Verhulst ()
Mathematisch Instituut, University of Utrecht, PO Box 80.010,
3508 TA Utrecht, The Netherlands
e-mail: f.verhulst@uu.nl

320 F. Verhulst
The canard behaviour will depend on the dimension of the problem and the nature
of the singularity. An example of canard behaviour was found by the Strassbourg
group working in non-standard analysis for a perturbed van der Pol equation; see [2]
for details and references. In this example, the singularity crossed is a fold point. The
analysis of this problem is quite technical.
Canards arising at transcritical bifurcations have been described in [6] and [9].
The purpose of the present note is to study such phenomena in growth equations that
can be handled explicitly, both analytically and numerically; this may increase our
understanding. In Sects. 2.4 and 4, we consider simple illustrations of exchange of
stability between slow manifolds. The equations in Sect. 5 model growth phenom-
ena with daily or seasonal fluctuations as appear in models of mathematical biology
and economics. They are a natural extension of the logistic model introduced in [19].
We shall also meet permanent canards, solutions that are attracted to a slow
manifold, travel across a singularity resulting in instability of the slow manifold, but
are remaining near the slow manifold without leaving the unstable part.
The numerics which we used to obtain illustrations is based on the package
CONTENT; see [8].
2 Fenichel Theory
In the period 1950–1960, many results were obtained for the asymptotic analysis
of singularly perturbed ordinary differential equations; see for instance [15].
Later, these results were supplemented by essential qualitative insight based on
developments of invariant manifold theory. This slow manifold analysis, as it is
usually called, is often named after one of its main inventors, N. Fenichel. The
canard aspects are an extension of the theory. We will summarise some important
aspects.
2.1 The Slow Manifold: Fenichel’s Results
Approximation theorems like Tikhonov’s theorem ([15]) are concerned with the
attraction, at least for some time, to the regular expansion that corresponds with a
stable critical point of a boundary layer equation. The theory is quite general and
deals with non-autonomous equations.
In many problems, it is possible to associate with the regular expansion a
manifold in phase or solution space and to consider the attraction properties of
the flow near this manifold. This raises the question of whether these manifolds
really exist or whether they are just a phantom phenomenon. Such questions were
addressed and answered in a number of papers by Fenichel (1971–1979) and other
authors; for references and extensive introductions the reader is referred to the
survey papers [3, 9] and [5]; see also [16].
Hunting French Ducks in Population Dynamics 321
Consider for t 0 the .n C m/-dimensional non-autonomous system:
xP D f .x; y; t / C " ; x 2 D Rn ;
"yP D g.x; y; t / C " ; y 2 G Rm :
A corresponding theory can be developed for the system:
x 0 D "f .x; y; "t / C "2 ; y 0 D g.x; y; "t / C " ;
where the prime denotes differentiation with respect to D t =".

We will call y the fast variable and x the slow variable. The zero set of g.x; y; t /
is given by y D .x; t /, which represents a first-order approximation M0 of the
slow manifold M" . The flow on M" is to a first approximation described by xP D
f .x; .x; t /; t /.
In the Fenichel theory we assume
ReSp gy .x; .x; t // ¤ 0; x 2 D:
That is, all the eigenvalues of the linearised flow near M0 , derived from the equation
for y, have nonzero real parts.
A manifold is called hyperbolic if the local linearisation is structurally stable
(real parts of eigenvalues all nonzero), and it is normally hyperbolic if in addition
the expansion or contraction near the manifold in the transversal direction is larger
than in the tangential direction (the slow drift along the slow manifold).
Note that this perspective on dynamics allows for interesting phenomena. One
might approach M" for instance by a stable branch, stay for some time near M" , and
then leave again a neighbourhood of the slow manifold by an unstable branch. This
produces solutions indicated as ‘pulse-like’, ‘multibump solutions’, etc. This type
of exchanges of the flow near M" is what one often looks for in geometric singular
perturbation theory.
2.2 Existence of the Slow Manifold
The question of whether the slow manifold M" , approximated by y D .x; t /,

persists for " > 0 was answered by Fenichel. The main result is as follows. If M0 is
a compact manifold that is normally hyperbolic, it persists for " > 0 (i.e., there exists
for sufficiently small, positive " a smooth manifold M" close to M0 ). Corresponding
with the signs of the real parts of the eigenvalues, there exist stable and unstable
manifolds of M" , smooth continuations of the corresponding manifolds of M0 , on
which the flow is fast.
There are some differences between the cases where M0 has a boundary or not.
For details, see [3, 4] and the original papers by Fenichel.
322 F. Verhulst
2.3 The Compactness Property
Note that the assumption of compactness of D and G is essential for the uniqueness
of the slow manifold. In many examples and applications, M0 , the approximation
of the slow manifold obtained from the fast equation, is not bounded. This can
be remedied, admittedly in an artificial way, by applying a suitable cutoff of the
vector field far away from the domain of interest. In this way, compact domains arise
that coincide locally with D and G. However, this may cause some problems with
the uniqueness of the slow manifold. Consider for instance the following example
from [16]:
Example. Consider the system
(
xP D 1; x.0/ D x0 > 0;
(1)
"yP D xy2 ; y.0/ D y0 0:
Putting " D 0 produces y D 0, which corresponds with M0 . We can obtain a

compact domain for x by putting x0 x L with x0 and L positive constants
independent of ". However, the limiting behaviour of the solutions depends on the
initial condition and L. Integration of the equations yields

1 1 1
y.x/ D y0 exp . / :
" t C x0 x0
As t tends to infinity, the solution for y.t / tends to

1
y0 exp ;
"x0
so the solutions are, after an initial fast transition, all exponentially close to y D 0.
However, there are an infinite number of slow manifolds dependent on x0 and L, all
tunnelling into an exponentially small neighbourhood of M0 given by y D 0; see
Fig. 1.
It is easy to modify the example to keep the variables on a compact domain; the
slow manifold will then be unique. Consider for instance:
Example.
(
xP D cos t; x.0/ D x0 > 1;
(2)
"yP D xy2 ; y.0/ D y0 0:
We have x.t / D x0 C sin t and .x0 1/2 x 2 .t / .x0 C 1/2 . It is easy to see that
we have the estimate:
Fig. 1 Slow manifold behaviour of system (1) with initial conditions x0 D 1; 2; 3, y0 D 0:5; " D
0:1. The limiting behaviour depends on the initial condition and the time interval
1" t
1" t
y0 e .x0 1/2 y.t / y0 e .x0 C1/2 :
The solutions decrease towards y D 0.

One might wonder about the practical use of exponential closeness as such
solutions are difficult to distinguish numerically. The phenomenon is important
and of practical use when there is a change of stability, a bifurcation of the slow
manifold. As we will see in a number of examples, exponentially close orbits may
trigger different canard phenomena.
2.4 Permanent or Periodic Canards
Consider an example discussed in [16] (example 8.13):
"yP D x.t /y.1 y/; y.0/ D y0 ; 0 y0 1:
The continuous function x.t / may be explicitly given or can be derived from a
coupled oscillator equation; " is a small positive parameter. The slow manifolds are
y D 0 and y D 1; one expects that if x.t / changes sign periodically, the solutions
324 F. Verhulst
will move very fast and periodically from one slow manifold to the other. This turns
out not to be the case.
Solving the initial value problem we find
1
Rt
y0 e " 0 x.s/ds
y.t / D 1
Rt :
1 y0 C y0 e " 0 x.s/ds
The solutions y D 0 and y D 1 are the (exact) slow manifolds of the dynamical
system. If x.t / > 0; y D 0 is stable, y D 1 unstable; if x.t / < 0; y D 1 is stable,
y D 0 unstable.
Assume that x.t / is T -periodic with alternating positive and negative values. We
can express x.t / as
Z T Z T
1
x.t / D a C f .t /; a D x.t /dt; f .t /dt D 0:
T 0 0
The solution can then be written as

a
y0 e " t E.t / 1
Rt
y.t / D a" t
; E.t / D e " 0 f .s/ds
: (3)
1 y0 C y0 e E.t /
The function E.t / is positive and bounded. We conclude:

1. The condition for periodicity y.0/ D y.T / applied to eq. (3) produces the
requirement a D 0. If a D 0, all solutions with 0 < y0 < 1 are T -periodic.
2. If a > 0, limt!1 y.t / D 0; although x.t / takes alternating positive and negative
values, the solution stays near y D 0; we have a permanent canard. See Fig. 2
where x.t / D a C sin t . Increasing n does not change the picture qualitatively.
Fig. 2 Permanent canards resulting from the dynamics of "yP D .a C sin nt /y.1 y/,
y0 D 0:6; " D 0:01. Left a D 0:5; n D 1, right a D 0:5; n D 1. Although the stability of
the slow manifolds changes periodically, the solutions tend after transient behaviour to y D 0 and
y D 1, respectively, to remain in their neighbourhood
Fig. 3 Spike solutions from the equation "yP D sin.2t /y.1 y/; y0 D 0:6
3. If a < 0, limt!1 y.t / D 1; the slow manifold y D 1 shows permanent canard

behaviour after a transient excursion to y D 0. See Fig. 2. Increasing n does not
change the picture qualitatively.
4. If a D 0, we have even more interesting canard behaviour with the possibility
of spikes for y.t /; see Fig. 3 where we have near-homoclinic behaviour in the
sense that the spikes return periodically; between the spikes the solutions are
exponentially close to the slow manifold y D 0. In the case of Fig. 3 we have
1
E.t / D e 2" .cos 2t1/ :
Thepspikes arise whenever cos 2t 1 D O."/; the spike width in this case is
O. "/.
Increasing n for x.t / D sin nt , the spikes diminish in height; this can also be
directly concluded from eq. (3).
In [16] it is indicated that if x.t / is derived from a chaotic oscillator, we may have
chaotic jumping between the slow manifolds.
326 F. Verhulst
3 Periodic Solutions
In a slow-fast system of the form
xP D f .x; y; t /; "yP D g.x; y; t /;
with a normally hyperbolic slow manifold (so without critical points), periodic
solutions can be found by localising to the slow manifold and applying standard
theory for periodic solutions. The idea is based on the rigorous theory of slow
manifolds. This is not the main subject of this note but see for instance [17].
Transcritical bifurcations play a part in [6] and [12]. In the last reference a prey-
predator system is analysed with a Holling II interaction term. This produces the
intersection of two slow manifolds with a periodic solution involving a canard when
the intersection is crossed. In the case of a constant carrying capacity, this model
is also considered in [18]. As we shall see, a modified logistic equation can be
considered as a standard form for this behaviour; an additional advantage of the
one-dimensional equations is that often they can be analysed in more analytic detail.
Although we can solve a number of equations explicitly, it is convenient to have
general theorems. From [11] we have the one-dimensional equation:
xP D f .x; t /; (4)
with f .x; t / defined for x 2 R and t 2 Œ0; 1/, continuous and satisfying the
uniqueness condition; f .x; t / is T -periodic in t .
• The solutions x.t / of the eq. (4) are either periodic or monotonic in the sense that
x.0/ < x.T / < x.2T / < : : : < x.nT / < : : : (or x.0/ > x.T / > : : : etc.); see
Theorem 9.1 in [11].
• If an isolated periodic solution of eq. (4) is Lyapunov-stable, then it is asymptot-
ically stable; see Theorem 9.2 in [11].
4 A Modified Logistic Equation
Consider the non-autonomous equation for y 0:
"yP D x.t /y y 2 ; (5)
in which the growth rate x.t / is sufficiently smooth and can take positive and
negative values. Values of t for which x.t / vanishes correspond with a transcritical
bifurcation. The equation can model daily or seasonal changes of the growth rate.
We solve the equation for general continuous x.t / and y.0/ D y0 > 0. Putting
Z t
˚.t / D x.s/ds;
0
we find for the solution of eq. (5):

1
e " ˚.t/
y.t / D Rt 1 : (6)
1
y0
C 1" 0 e " ˚.s/ ds
If x.t / is T -periodic, we can write
x.t / D a C f .t /
and
Z t
˚.t / D at C F .t /; F .t / D f .s/ds; (7)
0
with a a real constant and f .t / a zero average T -periodic function; F .t / is bounded

and F .T / D 0. In the cases that x.t / is quasi- or almost-periodic, we can write
similar expressions; see for an example Fig. 4.
Fig. 4 Solutions from eq. (5) with almost-periodic growth rate "yP D .0:5 C sin 2t C sin t /y
y 2 ; y0 D 0:4; " D 0:01; see Sect. 4
328 F. Verhulst
From now on, we assume that x.t / is continuous and T -periodic and ı is a small
positive constant, but independent of "; we have the following cases:
1. If the growth rate is negative, a C f .t / ı < 0 for 0 t T , we state
that limt!1 y.t / D 0. This can be deduced from the equation with a simple
estimate from "yP ı < 0 for all time, but also from the explicit solution (6).
Clearly a < 0 and the proof is simple: multiplying with exp.at ="/ produces
a bounded numerator and a monotonically increasing denominator. There is no
periodic solution with y0 > 0.
2. Less trivial is the case a D 0. In this case ˚.t / is bounded. The solution y.t /,
given by expression (6), decreases to zero as t ! 1 as the integral takes over
a positive function. Again, there is no periodic solution with y0 > 0. In Fig. 5
we choose x.t / D sin t for an illustration; y D 0 is associated with a permanent
canard.
3. a > 0 with a C min0tT f .t / < ı < 0. In this case a periodic solution exists.
Several proofs are possible, but a simple one runs as follows:
Assuming the periodicity condition y.0/ D y.T / D y0 .> 0/, we obtain from
the solution (6) the expression
aT
e " 1
y0 D RT 1
:
1 " ˚.s/
" 0 e ds
Fig. 5 Solution approaching the periodic solution in y D 0 of the modified logistic equation (5)
with x.t / D sin t; x.0/ D 0; y.0/ D 1; " D 0:01. To the right of the y-axis, y D 0 corresponds
with an unstable slow manifold
Fig. 6 Canard-like periodic solution of the modified logistic equation (5) with x.t / D 0:5 C
sin t; " D 0:01. To the right of the y-axis, y D 0 is an unstable slow manifold; x D 0 corresponds
with a transcritical bifurcation
With the right-hand side being positive, we have constructed a positive solution
for y0 and so a positive periodic solution.
For more explicit choices of x.t / we can show that y0 can both be expo-
nentially small and can be O.1/. See Fig. 6. The periodic solution follows
closely the slow manifold y D 0, produces a canard and increases quickly at
a positive value of x.t /; this value depends on a (in this case 0:5) and the shape
of x.t / as it depends on the time interval of exponential attraction during the
stable slow manifold phase. For instance, taking for simplicity the non-periodic
transition x.t / D a C bt , a; b > 0, we have ˚.t / D at C 12 bt 2 . It is
easy to show that at the time of transition of the y-axis, t D a=b, we have
y.a=b/ D O.exp:.a2 =.2b"/. For t > a=b the solution remains exponentially
close to y D 0 (canard) until the solution in an O."/-neighbourhood of t D 2a=b
jumps off to the slow manifold y D x.t /. In the case of the example displayed
in Fig. 6, this jump off (end of the canard) is delayed somewhat by the shape of
˚.t / D 0:5t C 1 cos t in this example.
The transcritical bifurcation takes place also periodically by the descent of
x.t / following the slow manifold y D x.t /. Starting in an O."/-neighbourhood
of the periodic solution on the slow manifold at x D y D 1 and choosing for the
transition the simple example x.t / D 1 bt; b > 0, we find
1
1 e "2b p
y. / D R 1
D O. "/:
b 1 1 2
1 C 1" 0b e " .t 2 bt / ds
330 F. Verhulst
Assuming that we can non-trivially linearise the slow manifold and the solutions
on it near the transcritical bifurcation, this estimate is typical.
Solutions near the slow manifold y D 0 correspond with near-extinction,
followed by a sudden explosive increase of the y-variable. The stability question
is more delicate in this case; we will consider this in Sect. 5.
4. If a < 0, we find a negative expression for y0 and no periodic solution, but a
permanent canard associated with y D 0.
5. If the growth rate is periodic but always positive, a C f .t / ı > 0, we find a
periodic solution in an "-neighbourhood of the slow manifold y D x.t /. This is
an example of the theory mentioned earlier; see [17].
5 The Periodic P.F. Verhulst Model
Consider the classical logistic equation of [19], but now with periodically varying
growth rate r.t / and carrying capacity K.t /, period T . In standard notation for the
population size N.t / with positive growth rate r.t /, the equation is

N
"NP D r.t /N 1 ; N.0/ D N0 > 0: (8)
K.t /
We have K.t / > m > 0. Without the fast growth perspective, the equation was
studied in [1, 14] and [13].
It is a natural assumption, at least for limited intervals of time, that r.t / can take
negative values. For such cases we modify the logistic equation to
N2
"NP D r.t /N ; N.0/ D N0 > 0 (9)
R.t /
with R.t / > 0 and T -periodic. Without this modification, a negative growth rate
would be accompanied by a positive nonlinear term; there is no rationale for this.
Equations (8) and (9) describe the dynamics if interaction with other populations
is negligible.
5.1 Positive Growth Rate
The slow manifolds that exist for eq. (8) are approximated by the limit cases
N.t / D 0 and N.t / D K.t /. Consider the stability of the slow manifolds. The
‘eigenvalues’ are respectively r.t / and r.t / so that the slow manifold N.t / D K.t /
is stable if r.t / > ı > 0. Replacing x.t / by r.t / we can repeat the calculation of
Sect. 4 to obtain
1
e " ˚.t/
N.t / D R 1
: (10)
1 t r.s/
1
N0
C " 0 K.s/ e
" ˚.s/ ds
As before, we put r.t / D a C f .t / with a a constant and f .t / a T -periodic

function with average zero. From eq. (10) we find with the periodicity condition
N0 D N.T / > 0:
aT
e " 1
N0 D R r.s/ 1 ˚.s/
: (11)
1 T
" 0 K.s/
e" ds
As a > 0, a periodic solution exists. We conclude: Assume that the growth rate
is T -periodic with r.t / ı with ı > 0, 0 t T , so we have stability of the slow
manifold N.t / D K.t /; N0 from eq. (11) is positive. For the periodic solution, we
have N.t / D K.t / C O."/ for all time. This is not immediately clear from the exact
solution (10).
5.2 Positive and/or Negative Growth Rate
The solution of eq. (9) is easy to find:

1
e " ˚.t/
N.t / D R 1
: (12)
1 t 1
1
N0
C " 0 R.s/
e " ˚.s/ ds
For intervals of time when r.t / ı > 0, a slow manifold is given by N.t / D
r.t /R.t /. On such intervals, this slow manifold is stable. When r.t / ¤ 0, N.t / D 0
is a slow manifold.
For a periodic solution to exist we apply the periodicity condition N.T / D N0 .
This produces
aT
e " 1
N0 D R 1
: (13)
1 T 1
" 0 R.s/
e " ˚.s/ ds
As r.t / changes sign during a period, we have the possibility of periodic transitions
between the approximate slow manifold N.t / D r.t /R.t / for r.t / > 0 and a
neighbourhood of the slow manifold N.t / D 0 if r.t / < 0 We draw a number
of conclusions:
1. Assume stability of the trivial solution (r.t / ı < 0 for 0 t T ).
As expected, it follows from solution (12) that N.t / ! 0 for t ! 1, so the
population will become extinct. No periodic solution exists.
332 F. Verhulst
Fig. 7 The case of eq. (9) with sign changing r.t / D 0:5 C sin t , R.t / D 2 C cos t , N0 D 0:4.
The first interval where r.t / > 0 is =6 < t < 5=6.=6 D 0:52 : : : ; 5=6 D 2:62 : : :/.
At t D 5=6, the trivial solution becomes stable; at t D 13=6 D 6:80 : : :, the trivial solution
becomes unstable again. If " D 0:1, a canard develops after r.t / becomes negative (at t D 2:62 : : :)
that puts the stable slow manifold N D r.t /R.t / that exists for positive values of r.t / outside of
reach. If " D 0:01, a canard develops earlier, at t D 0:52 : : :, the transient maximum is larger, a
permanent canard develops near t D 2:62 : : :
RT
2. Assume that r.t / changes sign with a < 0 (or 0 r.t /dt < 0). In the periodicity
condition (13), the numerator is negative and the denominator is positive, so no
periodic solution exists. All solutions of eq. (9) are monotonic. This can also be
seen as follows. The integral in the denominator of solution (12) is positive, so
we have the estimate:
1
N.t / N0 e " .atCF .t/ :
In the case a < 0, this is a strong estimate of exponential decay, but how do we
conciliate this with the stability of the slow manifold N D r.t /R.t / for intervals
of time when r.t / > 0? In Fig. 7 we have r.t / D 0:5 C sin t so we start with
negative values of r.t / for 0 t < =6 D 0:52 : : :; r.t / is positive for =6 <
t < 5=6 D 2:62 : : :. After t D 13=6 D 6:80 : : :, a canard develops that keeps
the solution close to the trivial solution N.t / D 0. The solution cannot make the
jump transition back to the slow manifold N D r.t /R.t /. Taking " D 0:01, a
Fig. 8 Three solutions of eq. (9) approaching periodic solutions in the cases r.t / D a C sin t ,
R.t / D 2 C cos t with a D 0:5 (largest amplitude), a D 0:3; 0:1; N.0/ D 0:4; " D 0:1. Canards
develop at the transitions from positive to negative values of r.t /
canard develops earlier, between t D =6 D 0:52 and 1; but near t D 1, the
solution grows again after which a permanent canard emerges near t D 2:62.
RT
3. Assume that r.t / changes sign with a > 0 (or 0 r.t /dt > 0). In the periodicity
condition (13), the numerator is positive, so a unique periodic solution exists.
In Fig. 8 we present three solutions approaching periodic solutions if r.t / D
a C sin t with a D 0:5; 0:3; 0:1; R.t / D 2 C cos t . Canards develop at the sign
transitions of r.t /. For instance, if a D 0:5, the first transition is at t D 7=6 D
3:67 : : :, but the canard delays the transition until a little above t D 4. If a > 0 is
diminished, the corresponding canard increases in size, as expected. In all cases,
each cycle faces the possibility of near-extinction.
4. Assume that we have the boundary case when r.t / changes sign with a D
RT
0 or 0 r.t /dt D 0. In this case the periodicity condition (13) cannot be
satisfied. In solution (12), the numerator varies periodically; the integral in
the denominator produces a positive, constant contribution at each cycle. The
maximum population density will decrease algebraically with time; see Fig. 9.
334 F. Verhulst
Fig. 9 Solution of eq. (9) in the case r.t / D sin t , R.t / D 2 C cos t , N.0/ D 0:4; " D 0:1. The
solution decreases as predicted; for smaller values of " the decrease is faster
6 Conclusion
It is remarkable that we can analyse the dynamics of a number of one-dimensional

growth models in some analytic detail. The analysis may serve as an inspiration for
problems with more complicated exchange of stability of slow manifolds.
Our models can be extended to involve one or more predators and to involve
spatial diffusion. This will be a subject of a forthcoming paper.
Acknowledgements Taoufik Bakri kindly introduced me to the use of CONTENT. Comments

by Odo Diekmann on the interpretation of the time-varying P.F. Verhulst model are gratefully
acknowledged.
References
1. Bernadete, D.M., Noonburg, V.W., Pollina, B.: Qualitative tools for studying periodic solutions
and bifurcations as applied to the periodically harvested logistic equation. MAA Monthly 115,
202–219 (2008)
2. Eckhaus, W.: Relaxation oscillations including a standard chase on French ducks. Asymptotic
Analysis II. Lect. Notes Math.985, 449–494 (1983)
3. Jones, C.K.R.T.: Geometric singular perturbation theory. Dynamical Systems, Lect. Notes
Math. Montecatini Terme 1609, 44–118 (1994)
4. Kaper, T.J.: An introduction to geometric methods and dynamical systems theory for singular
perturbation problems. Proc. Symp. Appl. Math. 56 (1999)
5. Kaper, T.J., Jones, C.K.R.T.: A primer on the exchange lemma for fast-slow systems. IMA
Vols. Math. Appl. 122 65–88 (2000)
6. Krupa, M., Szmolyan, P.: Extending slow manifolds near transcritical and pitchfork singulari-
ties. Nonlinearity 14, 1473–1491 (2001a)
7. Krupa, M., Szmolyan, P.: Relaxation oscillation and canard explosion. J. Diff. Equat.174,
312–368 (2001b)
8. Kuznetsov, Y.A., Levitin, V.V.: A multiplatform environment for analysing dynamical systems.
Dynamical Systems Laboratory, Centrum voor Wiskunde en Informatica, Amsterdam, (1997)
(http://www.math.uu.nl/people/kuznet/)
9. Kuznetsov, Y.A., Muratori, S., Rinaldi, S.: Homoclinic bifurcations in slow-fast second order
systems. Nonlinear Anal. Theory, Methods Appl. 25, 747–762 (1995)
10. Neihstadt, A.I.: Asymptotic investigation of the loss of stability by an equilibrium as a pair of
imaginary eigenvalues slowly cross the imaginary axis. Usp. Mat. Nauk 40, 190–191 (1985)
11. Pliss, V.A. Nonlocal problems of the theory of oscillations. Academic Press, NewYork (1966)
12. Rinaldi, S., Muratori, S., Kuznetsov, Y.A.: Multiple attractors, catastrophes and chaos in
seasonally perturbed predator-prey communities. Bull. Math. Biol. 55, 15–35 (1993)
13. Rogovchenko, S.P., Rogovchenko, Y.V.: Effect of periodic environmental fluctuations on the
Pearl-Verhulst model. Chaos, Sol. Fract. 39, 1169–1181 (2009)
14. Sonneveld, P., van Kan, J.: On a conjecture about the periodic solution of the logistic equation.
J. Math. Biol. 8, 285–289 (1979)
15. Tikhonov, A.N.: Systems of differential equations containing a small parameter multiplying
the derivative. Mat. Sb. 31(73), 575–586 (1952)
16. Verhulst, F.: Methods and Applications of Singular Perturbations. Springer, NewYork (2005)
17. Verhulst, F.: Periodic solutions and slow manifolds. Int. J. Bifur. Chaos. 17(8) 2533–2540
(2007)
18. Verhulst, F., Bakri, T.: The dynamics of slow manifolds. J. Indones. Math. Soc. (MIHMI) 13,
73–90 (2007)
19. Verhulst, P.F.: Notice sur la loi que la population poursuit dans son accroissement. Correspon-
dance Mathématique et Physique 10, 113–121 (1838)
Model of Nonlinear Fractal Oscillator
in Nanosystem
Valeriy S. Abramov
Abstract Based on the theory of fractional calculus and the concept of fractal
the model of nonlinear fractal oscillator is proposed. Hamilton operators in the
representation of occupation numbers for different structural states are obtained.
The features of the behavior of energy spectrum, deformation, and stress fields
for different structural states of the fractal dislocation in a model nanosystem are
investigated.
1 Introduction
Serge Haroche and David Wineland, 2012 Nobel Laureates in Physics, proposed
experimental methods that made it real to measure individual quantum systems
and govern them [8, 10]. The experimental studies of the features of the statistical
properties of individual quantum systems in neutron spin measurements [9], with
the observation of Bose-Einstein condensation [13], showed the presence of corre-
lations in the measured values. Near singular points (Dirac points) Dirac fermions
in molecular graphene show quantum and statistical features of behavior [11].
Actuality of fundamental research of individual quantum systems is related to the
possible use of them in quantum information technology. Using the principles of
quantum optics [15], the recording and subsequent reading of quantum information
(the coded in the polarization states of photons) is carried out on the quantum states
of single atoms or collective quantum states of atomic ensemble. It is noted the main
difficulty in the processing of quantum information: the results of photodetection are
random and reveal the presence of quantum chaos [17], and there is the phenomenon
of entanglement states of photons [7]. Physical properties of these quantum systems
(nanosystems) are essentially nonlinear. The methods of nonlinear dynamics have
been applied to the theoretical description of the chaos in structural mechanics [6],
the analysis of nonlinear chaotic models [16], and rare attractors and nonlinear
oscillators [18]. Fractal dislocation is one of the nonclassical structural objects in
V.S. Abramov ()

National Academy of Sciences of Ukraine, Donetsk, Ukraine
e-mail: vsabramov@mail.ru

338 V.S. Abramov
nanostructured materials [1–5]. In order to describe possible correlation effects and

statistical properties of the deformation field of fractal dislocation of mixed states a
two-point model was proposed [3].
This model is based on the theory of fractional calculus [14] and the concept of
a fractal [12]. Accounting for the effect of bifurcation of solutions of nonlinear
equations [2] leads to the appearance of the four branches of the lattice nodes
displacement function in model nanosystem. The aim of this paper is to describe
the nonlinear fractal oscillator in a model of fractal dislocation. The investigation
of features of behavior correlation and statistical properties of mixed states of the
fractal dislocation in the two-point model.
2 Fractal Oscillator in a Model of Fractal Dislocation
At the construction of model of fractal dislocation in the [1–5] was used the
b 2 from [4] for the energy spectrum of fractal dislocation
Hamilton operator H
b 2 D"1b
H n1 C"2b n03 ;
n2 C"3b b aC
n1 Db1ba1 ; b aC
n2 Db2ba2 ; n03 Db
b aC 3 ;
a3b b aC 3b
n3 Db a3 :
(1)
Here b
n1 ;b
n2 ;b
n3 are the operators of occupation numbers states of the disloca-
tion with nondimensional own energies "1 D "2 , "3 . Relations between the new
aC
b aC
1 ;b a3 and old b1 ;b
2 ;b c C ;b
c operators are defined by expressions
aC
b b c C C t31b
1 Dt11 1 Ct21b c; baC b
2 Dt12 1 Ct22b c C Ct32b
c; b c; b1 DDz˛ :
c C Ct33b
a3 Dt23b
(2)
b are defined by the relations
In expressions (2) the elements tij of the matrix T
t11 D k 0 ; t12 D k; t13 D 0; t21 D k cn .u˛ ; k/ ; t22 D k 0 cn .u˛ ; k/ ;

t23 D sn .u˛ ; k/ ; t31 D k sn .u˛ ; k/ ; t32 D k 0 sn .u˛ ; k/ ; t33 D cn .u˛ ; k/ :
(3)
Here k and u˛ D F(®˛ , k) are the module and the argument of the Jacobi elliptic
functions sn(u˛ , k), cn(u˛ , k); (k0 )2 D 1 k2 ; F is an incomplete elliptic integral of
the first kind; ®˛ is the polar angle. Using (2), we find the commutation relations for
the new operators
h i
b
a1 ;b aC Dbn0 bn1 D 2kk 0 s21b
b ˛2 C k 2 b0b
b ˛3 ;
1
h 1i
b
b ˛2 D b1 ;bz D ˛Iz1˛ ; bn01 D b aC
a1b1 ;
h i
b 0
n2 D b a2b C
a2 ; b C
a2 D b
a2 ;b 0
n2 b n2 D 2kk 0 s21b b ˛2 .k 0 / 2 b0b
b ˛3 ; b2 D Dz1˛ ;
Model of Nonlinear Fractal Oscillator in Nanosystem 339
p h i
s21 D .cn .u˛ ; k/ Csn. u˛ ; k // = 2; b aC
a3 ;b Db n3 Db0b
n03 b b ˛3 ; b0 D12n30 ;
h i3 h i
n30 Dsn2 .u˛ ; k/ ; b
b ˛3 D b2 ;bz D .1˛/ Iz˛ D b c;bcC ;
(4)
where bz is the coordinate operator. Structure operators fractional partial derivative

(integral) of the Riemann-Liouville D˛z (I ˛z ) on dimensionless coordinate z with the
index order ˛ is defined as
Z z Z z
Dz˛ ˆD@z ˆ ./ jzj ˛ d =# .1˛/ ; Iz˛ ˆD ˆ ./ jzj ˛1 d =# .˛/ ;
z0 z0
(5)
where @z is operator ordinary partial derivative on z and is gamma function.

Indices ˛, 1 ˛ have the meaning of fractal dimensions along the axis Oz. Taking
into account (2) and (4) we obtain the relations
h i 2
h i 2
n o 2
2b
cb z2 C b2 ;b
c C Db z b2 ; 2b
c Cb z2 b2 ;b
cDb z b2 ; b c C Db
c;b c C Cb
cb c Cb z2 b 2 :
cDb
(6)
n o
Acting by the operator b c C to the function ˆ˛c , we obtain the equation of a
c;b
fractal oscillator
2

bz2 b2 ˆ˛c D b z2 Dz1˛ Dz1˛ ˆ˛c D b cbcC C b
c Cbc ˆ˛c D .2n˛c C 1/ ˆ˛c :
(7)
Here n˛c , ˆ˛c are the eigenvalues (generally fractional values may depend on
z, ˛) and eigenfunctions of the fractal oscillator. Acting by the operator b
b ˛3 to the
function ˆ˛c , we obtain the equation
h i
b
b ˛3 ˆ˛c D b c C ˆ˛c D .1 ˛/ Iz˛ ˆ˛c D %˛c ˆ˛c :
c;b (8)
Function %˛c in the general case may depend on ˛ and z. Acting by the operator
D˛z on the left of (8), we obtain the nonlinear equation
h i
Dz˛ .%˛c ˆ˛c / DDz˛ .1˛/ Iz˛ ˆ˛c D .1˛/ ˆ˛c DDz˛ b c C ˆ˛c DDz˛b
c;b b ˛3 ˆ˛c :
(9)
To find the eigenvalues n˛c and eigenfunctions ˆ˛c , equations (7) and (9) must
be solved together. These equations are fundamental to describe the nonlinear
fractional oscillator. Note that if ˛ D 0 from (7) and (8) follow the equation
for the well-known linear quantum oscillator with positive integer eigenvalues
340 V.S. Abramov
nc D 0, 1, 2,..; eigenfunctions ˚ c ; boson commutation relations for the operators

b
c;bc C ; and parameter %˛c D 1
h i
d2
bz2 Dz1 Dz1 ˆc D b
z 2 2 ˆc D b c C Cb
cb c Cb
c ˆc D .2nc C1/ ˆc ; b c C ˆc Dˆc :
c;b
dz
(10)
On the basis of the commutation relations (4) we obtain the Hamiltonians

in the representation of occupation numbers for other structural states of fractal
dislocation:
b 1 D "2 .b
H n2 / C "3b
n1 C b n3 ; Hb 2 D "2 .bn1 C b n2 / C "3b n03 D Hb 1 C "3 b0b b ˛3 ;
b 3 D "1b
H n1 C "2b 0
n2 C "3bn3 D Hb 1 "2 b
A˛2 ; b A˛2 D 2kk s21b 0
b ˛2 C .k 0 /2 b0bb ˛3 ;
b
H 4 D "1b 0
n1 C "2bn2 C "3b b b b 0
n3 D H 1 C "2 A˛1 ; A˛1 D 2kk s21 b ˛2 C b 2 b
k b0 b ˛3 ;
b 5 D "1b
H n1 C "2bn02 C "3b b 1 C 2"2 kk 0 s21b
n03 D H b ˛2 C "3 C .k 0 /2 "2 b0b b ˛3 ;

b 6 D "1b
H 0
n1 C "2bn2 C "3b 0
n3 D Hb 1 2"2 kk 0 s21b
b ˛2 C "3 k 2 "2 b0b b ˛3 ;
b
H 7 D "1b 0
n1 C "2b 0
n2 C "3b b b
n3 D H 1 C "2 b0 b ˛3 ;
b 8 D "1b
H n01 C "2b
n02 C "3b b 1 C ."2 C "3 / b0b
n03 D H b ˛3 :
(11)
Dimensionless displacement u points of fractal dislocation (deformation field)

are connected with a parameter ˛ (stress field) by model relations (Hooke’s law):
u D ˛ =0 D F .'; k/ ; u˛ D u u0 ; (12)
where 0 is the normalization parameter and u0 is the constant (critical)

displacement.
3 The Two-Point Model for Mixed States Nanosystem
In this two-point model [3] based on the Hamiltonian H b 2 (1) the deformation field
of the fractal linear dislocation [1, 2] in a model sample of finite nanosize with vol-
umetric discrete lattice N1 N2 N3 is investigated. Deviations of the lattice nodes
from the state of equilibrium in a separate plane N1 N2 for two different points of
z1 (j0 ) and z2 (j0 ) are described by non-Hermitian displacement operators b u .z1 / and
b
u .z2 /, corresponding to the rectangular matrix with dimensions N1 N2 , j 2 [1, N3 ].
The original rectangular matrix displacements b u .z1 / and bu .z2 / with elements
unm (z1 ) D u"1 (z1 ), unm (z2 ) D u"1 (z2 ) in bulk lattice N1 N2 N3 D 30 40 67 were
obtained by the method of iterations on an index m for the first and second branches
of the dimensionless complex displacement functions u(z) D u"1 (z) and u(z) D u"2 (z)
by the formulas in [3], respectively,
u.z/ D u"1 .z/ D .g1 g2 C g4 / =2; u.z/ D u"2 .z/ D .g1 g2 g4 / =2: (13)
Functions g1 , g2 , g3 by analogy with [3] are modeled by expressions

g1 .u; ˛/ D .1 ˛/ 1 2sn2 u u0 ; k // = .p0
p1 n p2 mp p3 j / ;
2˛ 3˛1=2 ˛
g2 .z; ˛/ D 2 3 jz zc j # ˛ C 1=3 # ˛ C 2=3 = # ˛ C 1=2 ;

g3 .z; ˛/ D 33˛1=2 2 jz zc j 2˛ # ˛ C 1=3 # ˛ C 2=3 =;
h i
g4 D .g1 C g2 /2 g3 1=2 :
(14)
Here p0 , p1 , p2 , p3 are some governing parameters. The initial parameters

were the following: ˛ D 0.5; k D 0.5; u0 D 29.537; p0 D 0.01; p1 D 0.00075;
p2 D p3 D 0; zc D 2.813. In the calculations it should be : z1 D 0.053 C hz (j0 C 33);
z2 D 6.653 hz (j0 C 33); j0 2 [33; 33]; hz D 0.1, which corresponds to the forward
and backward waves of displacements unm (z1 ), unm (z2 ); n D 1; 30; m D 1; 40. The
value of j0 plays the role of dimensionless current discrete time. For j0 D 0 we have
z1 D z2 D 3.353. The choice of calculation scheme for the forward and backward
0 0
waves, an example of the behavior of real parts u"1 D Reu"1 and u"2 D Reu"2 for
the first and second branches of the direct wave of complex displacement functions
from (12) are shown in Fig. 1. The values of functions jz1 zc j, jz2 zc j at the
Fig. 1 Dependencies of jz1 zc j (curve 1), jz2 zc j (curve 2) on index j0 at zc D 2.813 for
the forward and backward waves (a); behavior Reu"1 (z1 ) (d), its projections (e, f); projections
Reu"2 (z1 ) (b, c) on the lattice index n, m for the forward wave at z1 D 0.053
342 V.S. Abramov
minimum (equal to 0.04) are achieved at j0 D 5, j0 D 5, respectively. The values

of these functions at the intersection point (j0 D 0) are equally 0.54 (Fig. 1a). The
choice of the governing parameters p0 , p1 , p2 , p3 determines the state of a discrete
rectangular sublattice N1 N2 with fractal dislocation, localized within this region
nOm parallel to the axis Om at the moment time j0 D 33 for the forward wave
0
(Fig. 1b–f). The core of the linear dislocation for the first branch u"1 is localized
at n D 14 (Fig. 1e). Deviation nodes of the core along the axis Om are pronounced
stochastic behavior (Fig. 1f). At n < 14 there is a pronounced stochastic behavior of
the whole plane nOm, and at n > 14 these deviations are practically zero. The core
0
of the linear dislocation for the second branch u"2 is localized at n D 17 (Fig. 1b). In
this case, on the background of the stochastic deviations nodes of the core along the
axis Om, a pronounced peak (at m D 23) is observed (Fig. 1c). The appearance of
this peak indicates about the presence of nonlinear fractional oscillator with special
properties. At n > 15 there is a pronounced stochastic behavior of the whole plane
nOm, and at n < 15 these deviations are practically zero. The results in Fig. 1 make
it possible to interpret the linear fractal dislocation as a set of coupled nonlinear
fractal oscillators.
For the description of mixed states the effective composite operators of displace-
ments are introduced:
b
u1 D b uC .z1 / ;
12b b
u3 D b uC .z2 / ;
12b b
u5 D b T12 ;
u .z1 /b b
u7 D b T12 ; (15)
u .z2 /b
u2 D b
b 21b
u .z1 / ; b
u4 D b
21b
u .z2 / ; b uC .z1 /b
u6 D b T21 ; b uC .z2 /b
u8 D b T21 : (16)
For the states p D 1, 2,...8, respectively. Here the symbols “ C ” and “T” mean
the operation of Hermitian conjugation and transposition. The square matrices
with sizes N1 N1 for p D 1, 3, 5, 7 (squeezed states) and N2 N2 for p D 2, 4, 6, 8
(stretched states) correspond to the introduced operators b up , so that bu5 D b uC1 ,
C C C T T
b
u7 D b u3 , b
u6 D b u2 , b
u8 D b u4 . The density state operators b
12 ; b
12 ; b
21 ; b
21 are
represented by
T T
12 D b
b N1b
N 2 =N 1 N2 ; T12 D b
b N 2b
N1 =N 1 N2 ; b T12 ;
21 D b T21 D b
b 12 ; (17)
where b N1 , b
N 2 are row vectors of dimensions 1 N1 , 1 N2 , with elements equal
to one. The rectangular matrices b 12 , b
21 have dimensions N1 N2 , N2 N1 . For
T
the operators in (17) the normalization conditions are fulfilled b b b D 1;
N1 12 N 2
T
b 21b
N 2b N1 D 1. Having performed an averaging over the index nodes n, m by
calculating trace Sp of square matrices (15), (16), the averaged functions up , jup j,
tg' p for p D 1, 2,...8 are obtained:
ˇ ˇ
up D u0p C i u00p D ˇup ˇ exp i'p ;
up D Spb uC
up D Spbp; tg'p D u00p =u0p ;
(18)
0
where up D Reup , u00p D Imup , the symbol “ ” means the operation of complex con-
jugation, and jup j, ®p are a module, a phase of the complex averaged functions up .
Here the averaging across indexes j, j0 is not made. Then we find the correlation
function of the first order Kpq for p, q 2 [1, 8]
ˇ ˇ ˇ ˇ ˇ ˇ ˇ ˇ
Kpq D Spq Hpq D Kpq 0 00
C iKpq D ˇKpq ˇ exp i
pq ; ˇHpq ˇ D ˇup ˇ ˇuq ˇ ;
ˇ ˇ C
Spq D Spb 0
D Spq
S pq C iS 00
D ˇSpq ˇ exp i pq ; bS pq Db uC
upb b
S pq ¤ b
pq q ; S pq ;
ˇ ˇ
Hpq D Spbup Spb C 0 00 ˇ ˇ
uq Dup uq DHpq CiHpq D Hpq exp i ıpq ; ıpq D'p 'q :
(19)
Having done the normalization of the above functions, we obtain the distribution
0
function of mixed states of Bose-Einstein type fpp (ground state) and f pp (excited
0
state), Fermi-Dirac type Fpp (ground state) and F pp (excited state), and inversion
parameters dpp in the form
0 0
fpp fpp D 1; fpp D Spp =Hpp ; fpp D Kpp =Hpp ; (20)
0 0
Fpp C Fpp D 1; Fpp D Hpp =Spp ; Fpp D Kpp =Spp ; dpp D 1 2Fpp :
(21)
Distribution functions of the Fermi-Dirac type carry information about the defor-
mation field. The distribution functions of Bose-Einstein type carry information
about the stress field. The behavior of the distribution functions of the Fermi-Dirac
type, Bose-Einstein type, and inversion parameters on the dimensionless time j0
for squeezed and stretched states is shown in Figs. 2 and 3. From these figuresa
significant difference in the behavior of these functions of the dimensionless time j0
for the first u"1 and second u"2 branches of the complex displacement function u is
seen.
Note that the behavior of these functions for the first branch u"1 , as a function
of the dimensionless coordinates z, discussed in detail in [3]. For the second branch
u"2 is characterized by the presence of peaks up to the dependencies f11 , f22 on j0 at
j0 D 17. For a squeezed state (Fig. 2b, d), this peak is formed from the ground state
with F11 D 0.228 (d11 D 0.544), and for a stretched state (Fig. 3b, d), this peak is
formed from an inverted state with d22 D 0.402 at F22 D 0.701.
The appearance of such peaks indicates the presence of nonlinear fractal
oscillators with special properties. These results can be interpreted by analogy in
terms of transient effects such as induction, avalanche, self-induced transparency,
and echo, known from quantum optics [15]. At p ¤ q from (19) it follows that the
function Kpq is complex. For some values p, q this function has a sense of cross-
correlated function (for a pair of different points z1 , z2 ).
344 V.S. Abramov
Fig. 2 Dependencies of the distribution functions of the Fermi-Dirac type (a, b) and Bose-Einstein
type (c, d) on j0 for squeezed states (p, p) D (1, 1) of branches u"1 (a, c), u"2 (b, d)
4 The Characteristic Functions for Effective Pair Operators

of Displacements
To investigate the correlation between the effective pair operators of displacements

C
S pq , b
b S pq , there is a need for second-order correlation functions Gpq
C
Gpq DVpq Wpq DSp bpq ; Vb pq DbS pq b bC b
pq DV pq ;
; Vpq
C
V
ˇ ˇ
S pq ; V
Wpq D Spb S pq Spb S pq DSpq Spq
D ˇSpq ˇ 2 ; h0pq Chpq D1; h0pq DWpq =Vpq ;
hpq DGpq =Vpq ; Dpq D 1 2h0pq ; 0
Cpq D 1= h0pq ; 0
Cpq Cpq D 1:
(22)
Dependencies from (23) of the inversion parameters D57 (Fig. 4a, b) and D68
0 0
(Fig. 5a, b), the distribution functions of Bose-Einstein type C57 (Fig. 4c, d) and C68
(Fig. 5c, d) on the dimensionless time j0 for the squeezed and stretched states (p, q)
show significant differences in the behavior deformation and stress fields for the two
branches of the complex displacement functions u"1 , u"2 .
Fig. 3 Dependencies of the inversion parameters (a, b) and the distribution functions of Bose-
Einstein type (c, d) on j0 for stretched states (p, p) D (2, 2) of branches u"1 (a, c), u"2 (b, d)
Thus functions h57 , h68 are positive and can be interpreted as the distribution
functions of Fermi-Dirac type for effective pair operators of displacements. Using
(19) we find a representation for
ˇ ˇ 2 ˇ ˇ ˇ ˇ ˇ ˇ ˇ ˇ
ˇSpq ˇ D ˇKpq jj up jj uq ˇ 2 C 2 ˇup ˇ ˇuq ˇ ˇKpq ˇ 1 C cos ˆpq ; (23)
where ˆpq D ı pq
pq . The analysis of the influence of correlation effects for the
phase ˆpq is conveniently carried out in terms of characteristic functions Rpq , rpq ,
2pq , which are given by
0 0
Rpq C Rpq D 1; Rpq D 1 C cos ˆpq ; Rpq D cos ˆpq ; (24)
0 0

rpq rpq D 1; rpq D 2= 1 Rpq ; rpq D 1 C Rpq = 1 Rpq ; (25)
2 2
0pq C 2pq D 1; 0pq 0
D cos2 ˆpq =2 D 1=rpq ; 2pq D sin2 ˆpq =2 :
(26)
346 V.S. Abramov
Einstein type (c, d) on j0 for squeezed states (p, q) D (5, 7) of branches u"1 (a, c), u"2 (b, d)
Functions Rpq from (24) have a sense of the normalized parameter describing
the population inversion between the states p and q taking into account correlations
(Figs. 6a, b, 7a, b).
0
Functions rpq , rpq from (25) make sense of the characteristic distribution func-
tions of Bose-Einstein type in ground, excited states, respectively (Figs. 6c, d, 7c,
0
d). Functions 2pq , (pq )2 from (26) make sense of the characteristic distribution
functions of Fermi-Dirac type in ground, excited states, respectively. Functions
0
2pq , (pq )2 also can be interpreted as the parameters of the polarization degree,
the entanglement degree of states [7]. Dependencies of the characteristic phase
functions R13 (Fig. 6a, b) and R57 for squeezed states (30 30) coincide.
By analogy dependencies of the characteristic phase functions R24 (Fig. 7 a, b)
and R68 for stretched states (40 40) coincide. Detailed behavior of the functions
h13 , h24 is shown in Fig. 8: observed shapingpulsesof complex form. Thus on the
whole interval j0 2 [33; 33] functions h13 , h24 are of variable sign and oscillate
around zero, but the behavior of these functions is different. Changing the sign of
the functions h13 , h24 is related to the change of sign of the second-order correlation
functions, which indicates a change of the nature of relations between the pair
operators of displacement b S pq .
Einstein type (c, d) on j0 for stretched states (p, q) D (6, 8) of branches u"1 (a, c), u"2 (b, d)
On the other hand, the interpretation to change the statistics for the pair
operators of displacement is allowed b S pq : from the statistic of Fermi-Dirac type
(positive values of the distribution functions h13 , h24 ) to the statistic of Bose-Einstein
type (negative values h13 , h24 ). However, dependencies of distribution functions
h13 , h57 for squeezed states (30x30) differ sharply. By analogy the dependencies of
distribution functions h24 , h68 for stretched states (40x40) also differ sharply. In this
case, the function h13 , h24 changes sign near zero, and there is a change in the type
of the interaction: repulsion or attraction for pair operators of displacement (similar
change in the type of interaction between biphonons).
5 Conclusions
Within the model of the fractal dislocation it is proposed to describe the fractal
nonlinear oscillator. We obtain a system of coupled nonlinear equations, which
is based on the theory of fractional calculus and the concepts of fractal. For a
description of other possible structural states of fractal dislocation corresponding
348 V.S. Abramov
Einstein type (c, d) on j0 for squeezed states (p, q) D (1, 3) of branches u"1 (a, c), u"2 (b, d)
Fig. 7 Dependencies of the inversion parameters (a, b) and the distribution functions of
Bose-Einstein type (c, d) on j0 for stretched states (p, q) D (2, 4) of branches u"1 (a, c), u"2 (b, d)
Fig. 8 Dependencies of the distribution functions on j0 for squeezed (p, q) D (1, 3) (a, b) and
stretched (c, d) states of branches u"1 (a, c), u"2 (b, d)
Hamilton operators are obtained. Within the two-point model for mixed states
(squeezed and stretched) the features of correlation and statistical properties of
deformations and stress fields of fractal dislocation are investigated. By the method
of numerical simulation it is shown that the fractal dislocation can be represented
as a set of coupled fractal oscillators. The appearance of the individual peaks in the
core of fractal dislocation (for the first and second branches of the displacement
functions) indicates the presence of separate fractal nonlinear oscillators with
special properties. It is shown that for the pair displacement operators there is the
change in the type of correlation connection and statistics and there is the effect of
the presence of entanglement states.
References
1. Abramov, V.S.: Fractal dislocation as one of non-classical structural objects in the nano-
dimensional systems. Metallofiz. i Noveishie Tekhnologii. 33(2), 247–251 (2011)
2. Abramov, V.S.: Behavior of the deformation field of fractal dislocation in the presence of
bifurcations. Bul. Donetsk Nat. Univers. Ser. A. 2, 23–29 (2011)
350 V.S. Abramov
3. Abramov, V.S.: Ordering and statistical properties of the deformation field of fractal dislocation
in a model nanosystem. Bul. Donetsk Nat. Univers. 1, 90–96 (2013)
4. Abramova, O.P., Abramov, S.V.: The influence of fractal fields of deformation and stress on
the energy spectrum of dislocation. Bul. Donetsk Nat. Univers. 1(1), 119–123 (2005)
5. Abramova, O.P., Abramov, S.V.: Features of behaviour of the deformation field of fractal quasi-
two-dimensional structures. Bul. Donetsk Nat. Univers. 1, 9–17 (2013)
6. Awrejcewicz, J., Krysko, V.A.: Chaos in Structural Mechanics. Springer, Berlin, Heidelberg
(2008)
7. Chekhova, M.V., Fedorov, M.V.: The Schmidt modes of biphoton qutrits: Poincare-sphere
representation. J. Phys. B At. Mol. Opt. Phys. 46(095502), 1–10 (2013)
8. Chou, C.W., Hume, D.B., Rosenband, T., Wineland, D.J.: Optical clocks and relativity. Science
329, 1630–1633 (2010)
9. Erhart, J., Sponar, S., Sulyok, G.: Experimental demonstration of a universally valid error-
disturbance uncertainty relation in spin measurements. Nat. Phys. 8, 185–189 (2012)
10. Gleyzes, S., Kuhr, S., Guerlin, C., et al.: Quantum jumps of light recording the birth and death
of a photon in a cavity. Nature 446, 297–300 (2007)
11. Gomes, K.K., Mar, W., Ko, W., et al.: Designer Dirac fermions and topological phases in
molecular graphene. Nature 483(7389), 306–310 (2012)
12. Mandelbrot, B.B.: The Fractal Geometry of Nature. Freeman, New York (1982)
13. Perrin, A., Bücker, R., Manz, S.: Hanbury Brown and Twiss correlations across the Bose-
Einstein condensation threshold. Nat. Phys. 8, 195–198 (2012)
14. Samko, S.G., Kilbas, A., Marichev, O.: Fractional Integrals and Derivatives. Theory and
Applications. Gordon and Breach Science Publication, New York (1990)
15. Scully, M.O., Zubairy, M.S.: Quantum Optics. Cambridge University Press, Cambridge (1997)
16. Skiadas, C.H., Skiadas, C.: Chaotic Modeling and Simulation: Analysis of Chaotic Models,
Attractors and Forms. Taylor and Francis/CRC, London (2009)
17. Stockmann, H.-J.: Quantum Chaos. An Introduction. Cambridge University Press, Cambridge
(1999)
18. Zakrzhevsky, M.V., Smirnova, R.S., Schukin, I.T., et al.: Nonlinear Dynamics and Chaos.
Bifurcation Groups and Rare Attractors. RTU Publishing House, Riga (2012)
Dynamical Statement Networks
Wojciech Cholewa
Abstract This paper addresses dynamics in diagnostic expert systems. The

introduced dynamic statement network may be used not only in static diagnosing
systems but also in dynamic monitoring systems. Within these networks, statements
consist of contents and values. Statement values are based on a concept introduced
in intuitionistic fuzzy sets, i.e. they contain independent belief about validity
and invalidity of information presented by statement contents. The relationships
between statements are modelled in the form of a set of necessary as well as
sufficient conditions. The author compared requirements placed against static and
dynamic statement networks and devised a manner in which a static network may be
transformed into a dynamic one which, in turn, facilitates non-monotonic reasoning
required for monitoring systems.
Keywords Expert system • Intuitionistic network • Knowledge base

• Diagnostic reasoning
1 Introduction
Systems facilitating evaluation of a technical state of objects are of crucial practical

significance. A recognition of the actual state or a change in this state frequently
provides argumentation for decision-making in reference to further operation of
objects in question. Such systems may be developed and devised as diagnostic
expert systems to support a recognition of the current state of an object [3].
Monitoring systems facilitate observations of objects in the course of time. They
may include diagnostic modules that would allow for recognition of the present
state on an object during its changes in time. The purpose of the considered systems
may refer to objects of different classes such as machines, industrial and economic
processes, patients at intensive care wards, meteorological processes, and many
more.
W. Cholewa ()
Faculty of Mechanical Engineering, Silesian University of Technology,
Konarskiego 18a, 44-100 Gliwice, Poland
e-mail: wojciech.cholewa@polsl.pl

352 W. Cholewa
One finds a great variety of the currently available expert systems and their
detailed overview would go beyond the purpose of the presented paper. Neverthe-
less, it is essential to note that before research into new classes of such systems
can proceed, the corresponding needs must be meticulously analysed. Such an
analysis should indicate those features of the available systems that require further
modifications.
The majority of expert systems that enable inference in conditions of inaccurate
and incomplete data require a considering of the complete knowledge base during
their development. Devising such systems by combination of independently devel-
oped and independently verified component subsystems constitutes a challenging,
or impossible, task.
Practical applications of expert systems in diagnostics involve an assumption
that we frequently do not have the complete data which are required to enable the
inference process. Some types of expert systems such as bayesian network-based
systems allow for substitution of incomplete data with default ones. A number
of methods were developed that enable arbitrary acknowledgment or acquisition
of such default data, e.g. on the basis of available historical data. The results of
inference related to application of default data are communicated to the system user
who is to take a decision concerning further operation of the object upon analysis of
received information.
While taking a decision, the user should know whether the result of the inference
process is conditioned by direct results of both observation and measurements or
whether it originates from the default data whose occurrence is expected for an
average object, however, not necessarily for the analysed object in question. The
decisions should base on a belief that there are indications of their validity, i.e.
plausibility, advisability, and legality. Systems basing on intuitionistic logic are
considered to be an interesting example of inference systems that enable collection
of judgments. In this logic, propositional calculus allows for a syntax transformation
that ensures its justification. This differentiates intuitionistic logic from classical
logic whose propositional calculus maintains the truthfulness of transformed syntax.
The following section of the paper presents essential operation activities of expert
systems in which knowledge base is saved as intuitionistic statement networks. Such
systems were introduced in order to limit any inconveniences occurring during
development and applications of discussed systems. This, in particular, applies to
difficulties with such systems resulting from:
• A need of investigation into a complete knowledge base during system develop-
ment
• No possibility of differentiation between conclusions formulated on the basis
of default data that substitute missing (unknown) data as well as conclusions
conditioned on a complete set of data
• No possibility of integration with a knowledge base of detailed explanations
and sets of source information that justify proper operation of the system and
facilitate interpretation
Dynamical Statement Networks 353
2 Statements
Diagnostic systems recognize the object state basing on the available diagnostic
knowledge directly pertaining to the object family to which the examined object
belongs as well as on available data determined as values of features (attributes)
of the given object. These features may directly concern the impact of the object
onto the surrounding, the impact of the surrounding onto the object, and may define
the structure of the object, its features, etc. The features may manifest ratio and
interval quantitative values as well as ordinal and nominal qualitative values. Strings
of signs come as a particular kind of feature values. Recording of such data in the
form of OA matrix (object-attribute matrix) whose elements are values of features
is a commonly applied representation method. The OA matrix is also referred to as
Aristotelian table [8] or an informational system [7]. The columns of the OA matrix
correspond to features (attributes), whereas rows reflect different instances of the
object, instances being various objects belonging to the studied object class, or an
object of the selected class at different time intervals.
A hypothesis stating that for the object class in question the OA matrix
contains unfailing regularities allowing for inference process, constitutes the basic
assumption related to diagnostic applications of the OA matrix.
Knowledge acquisition for the needs of diagnostic systems may be realized as a
result of applied processes of machine learning [9] as well as a result of cooperation
with domain experts. In each case, a need to define a degree of generality of
considered data arises. Not only a considerably high but also significantly low
degree of generality of data may hinder or even block entirely a possible definition
of a sustainable solution. An optimization should result in assuming appropriate
granularity of information represented by the data. One expects that the optimal
granularity should allow for identification and generalization of dependencies
occurring between the data. Furthermore, it is expected that the correct interpretation
of obtained information granules will be maintained.
A number of formalisms relating to possible methods of information granulation
is currently well known [2]. Statements, most suitable for practical uses, constitute
a kind of informational granules that facilitate development and construction of
knowledge models in a form of statement networks [4].
The noun statement has several meanings. Here, the statement is information on
recognition of proposition resulting from observed facts or representing an opinion.
The statement s consists of a content c.s/ and value b.s/
s D h c.s/; b.s/ i: (1)
The content c.s/ of a statement s is presented as an indicative sentence, i.e. a

sentence which may be either entirely true or entirely false. On the basis of the
content only, one may not conclude whether the statement has been recognized or
not. The information on recognition of the statement is represented as a value b.s/
of the statement s, independently of its content c.s/.
354 W. Cholewa
The values of statements may be analysed as elements of OS matrix (object-

statement matrix) which is analogically constructed as the OA matrix. The columns
of the OS matrix correspond to statements, whereas rows refer to object instances.
The values of statements may be determined on the basis of the values of features
contained in the OA matrix. It may be assumed that the OA matrix includes all data
required for definition or recognition of the statement value. This results in the OA
matrix being transformable into the OS matrix. Having considered this occurrence
it appears that the data on the object may be represented in a form of statements.
It is very important that not only data on the object design and working conditions
but also results of inferring on the object state may appear in a form of statements.
This, in turn, allows for an inference process carried out in a set of statements
in which diagnostic knowledge is represented by relations occurring between the
statement values. It should be assumed that
• the set of considered statements is a finite set in which contents of statements do
not change,
• the set of considered relations between statements is known and fixed for the
studied class of objects.
In inference processes realized by expert systems, conclusions are either acknowl-
edged or rejected on the basis of formerly recognized premisses. Sets of statements
studied as sets of premisses and conclusions are fixed and finite ones. They
are determined during system development and construction. During the system
operation no new conclusions, appearing as new statements, are generated or
discovered. This, subsequently, means that inference processes in diagnostic expert
systems that are realized with the use of OS matrices are processes conducted in
so-called closed worlds.
3 Intuitionistic Statement Value
A value of a statement constitutes a carrier of information on recognition or the lack

of recognition of the statement content. The boolean type is the fundamental type
of a statement value. It consists of logical values true and false which are frequently
encoded as numerical values, 1 and 0, respectively. If the value of a statement is not
equal to one of these values, then it must be equal to the other value. It means the
Boolean type complies with a so-called law of excluded middle.
The boolean type of statement values facilitates construction of accurate infer-
ence systems. Due to restricted quality of available knowledge and available
information, a use of approximate inference systems is frequently required. A degree
of belief about the truth of a statement content is an example of a value of statements
considered as nodes of belief networks which are also referred to as bayesian
networks [6]. Not only a certainty factor or subjective probability of a statement
truthfulness or degree of belief in the truth of a statement [5] but also a degree
of belonging to a fuzzy or rough set of statements recognized as true, and a large
number of similar concepts, may become a value of a statement. A common feature

that is shared by the enumerated types of statement values is that these values are
represented in a form of a point value by a one number only.
In practical applications of expert systems, one needs to determine statement
values in a manner that allows differentiation between the following groups of
statements:
• statements that are not recognized as either true or false,
• statements for which the numbers of for and against arguments on their recogni-
tion are equal
Additionally, these systems require that conclusions be formulated independent on
the basis of a belief about the truth of selected statements as well as on the basis of
belief about the falsity of other statements.
The expected manner of representing a statement value is hard to introduce if a
statement value is written down as a single number. For this purpose it is essential
to consider a statement value defined in a form of pairs of numbers. An example
of such a value may include a pair consisting of a necessity degree and possibility
degree originating from modal logic. Another concept of a statement value which
is worthy of attention is a definition of this value (2) derived from the theory of
intuitionistic fuzzy sets (IFS) [1].
An intuitionistic value b.s/ of statement s, also referred to as the statement value,
is represented by the ordered pair
b.s/ D h p.s/; n.s/ i for p.s/; n.s/ 2 Œ 0 ; 1 (2)
where p.s/ is a degree of validity (justification, truthfulness) and n.s/ is a degree

of invalidity (lack of justification, lack of truthfulness) of statement s. These values
are also referred to as evaluations of positive and negative information. Within IFS
it is assumed that
p.s/ C n.s/ 2 Œ 0 ; 1 ; (3)
and it is not assumed that
p.s/ C n.s/ D 1:0 : (4)
Considering (2), the value b.s/ of an unrecognized statement s, i.e. a value of the
statement not recognized as true or false, takes the following form:
b.s/ D h 0; 0 i (5)
Values p.s/ and n.s/ allow for defining a hesitation margin h.s/ for statement s
which is also known as a degree of nondetermination
h.s/ D 1 p.s/ n.s/ (6)

356 W. Cholewa
where (3) results in
h.s/ 2 Œ 0 ; 1 (7)
The literature on IFS uses notation .s/; .s/; .s/ instead of p.s/; n.s/; h.s/.
The statement values (2) may be determined directly with respect to the
assumption (3). They may also result from inference on the basis of other statements
in accordance with an assumed knowledge model. A potential imperfection of the
applied knowledge model and inconsistency of the studied set of statement values
may lead to unexpected results of the inference process manifested as disagreements
and contradictions in the set of selected statement values which may not satisfy
restrictions (3). Such possibility leads to an extension of the definition (6) of
hesitation margin

1 p.s/ n.s/ if p.s/ C n.s/ 1
h.s/ D (8)
0 if p.s/ C n.s/ > 1
In addition a disagreement level d.s/ for values p.s/ and n.s/ of statement s not
satisfying restriction (3) is also introduced:

0 if p.s/ C n.s/ 1
d.s/ D (9)
p.s/ C n.s/ 1 if p.s/ C n.s/ > 1
where
d.s/ 2 Œ 0 ; 1 : (10)
Occurrence of positive values d.s/ implies contradiction in either the applied

knowledge model or in the assumed statement values.
4 Statement Network
Elements which statement networks consist of are statements s 2 S considered as

pairs (1) of both contents c.s/ and values b.s/ of these statements. The state of
statement networks is determined by a set of statement values
b.S / D f b.s/ W s 2 S g (11)
The essence of statement networks is that a change in a value of a particular state-

ment occurring in the network may induce changes in values of other statements.
A statement network may be represented as a graph G
G D h C; V; E; A i where E D Ep [ En ; (12)
with a static subgraph H defining a constant structure of a statement network and

its elements
H D h C; V; E i ; (13)
and a dynamic directed subgraph D defining interrelations of statements and actual

statement values
D D h V; A i ; (14)
where:
C is a set of vertices representing statement contents,
V is a set of vertices representing values of degrees of validity,
E is a set of edges connecting vertices c 2 C which represent statement contents
with vertices v 2 V which represent degrees of validity or invalidity,
Ep is a set of p-edges ep .s/ D hc.s/; p.s/i connecting contents c.s/ of
statements s with values of degrees of their validity p.s/,
En is a set of n-edges en .s/ D hc.s/; n.s/i connecting contents c.s/ of
statements s with values of validity degrees of their contradictions, i.e. with
values of degrees of their invalidity n.s/,
A is a set of directed edges (arcs) defining relations between values of degrees of
validity (or invalidity).
The vertex v 2 V representing the value of degree of validity or invalidity may
be connected by means of edges e 2 E with numerous vertices c 2 C representing
statement contents, which allows for, inter alia, concurrent considering of equivalent
statement contents. Each vertex c 2 C may be connected with no more than one
p-edge and one n-edge.
Each edge a D hs1 ; s2 i 2 A of the graph G defines a true implicational
relationship s1 ! s2 between statements s1 and s2
. a D h s 1 ; s2 i 2 A / ) . s 1 ! s 2 / (15)
This relationship proves that the antecedent, i.e. statement s1 , is a sufficient

condition for the consequent, i.e. statement s2 . The sufficiency condition denotes
that recognition of the validity of statement s1 is always accompanied by recognition
of validity of statement s2 .
The necessary and sufficient conditions can be modelled by means of inequalities
[3]. For considered intuitionistic values of a pair of statements s1 and s2 connected
by an edge a D hs1 ; s2 i and for the selected value b.s/ the following sufficiency
condition is obtained:
. s1 ! s2 ; b.s1 / D hp.s1 /; n.s1 /i / ) . p.s2 / p.s1 / / (16)

358 W. Cholewa
Furthermore, for the given value b.s2 /, the following necessity condition is
obtained:
. s1 ! s2 ; b.s2 / D hp.s2 /; n.s2 /i / ) . n.s1 / n.s2 / / (17)
One considers, in the similar fashion, a sufficient condition including one

consequent s and a disjunction of a number of antecedents s1 ; s2 ;
. . _i si / ! s/ ) p.s/ max p.si / (18)

i
as well as conjunctive of a number of antecedents s1 ; s2 ;
. . î si / ! s/ ) p.s/ min p.si / (19)

i
In a set of vertices representing values of degrees of validity and invalidity one

can distinguish between three subsets:
V D Vor [ Vand [ Vf ix ; (20)
where:
Vor is a set of vertices whose values are defined according to (18), when they
appear as consequents of sufficient conditions,
Vand is a set of vertices whose values are defined according to (19), when they
appear as consequents of sufficient conditions,
Vf ix is a set of vertices whose values are arbitrarily acknowledged or are defined
by means of measurement sets or other systems.
The introduced concept of a statement network is illustrated in Fig.1 which
depicts a very simple network, dealing with a pocket flashlight. Statements s3 and
s4 present the manner in which different statement contents may be allocated to the
same vertices representing the degrees of validity.
5 Solving of Dynamic Statement Network
A statement network represents knowledge in a form of graph G (12). While

applying the statement network, statement values serve as variable elements of the
graph. The remaining elements, i.e. statement contents, undirected edges, as well as
directed edges, are not subject to change.
A statement network which is studied as a model of a dynamic object or process
should adapt to changing operation environment due to which values v 2 Vf ix
of selected statements may be subject to changes. It is essential to emphasize that
updates of values v 2 Vf ix may be performed in an asynchronous manner.
Fig. 1 An example of a statement network, where v1 2 Vand ; v2 v6 2 Vor (see text for details)
Solving of a statement network consists in application of known values of

vertices v 2 Vf ix as well as relation a 2 A according to (14) in order to determine
unknown values of vertices v 2 V nVf ix . A simple algorithm consists in assuming
initial zero values of degrees of validity for v 2 V nVf ix and, subsequently, in
iterative increasing selected degrees of validity that fulfils relations a 2 A. The
implemented process of inference is a monotonic one.
In order to limit the monotonic properties of inference, a process of gradual
forgetting of the defined degrees of validity may be introduced. This applies to all
values v 2 V , i.e. to both values of primary statements v 2 Vf ix and values of
secondary statements v 2 V nVf ix . The values of these degrees are studied as values
dependent on a discrete time t , i.e.
vi .t /; vi .t C 1/; : (21)
The process of forgetting may be considered in the following form:
vi .t C 1/ WD qi vi .t / I 0 < qi 1 ; (22)
where the introduced constants qi assigned separately to all vertices vi 2 V allow

for tuning of dynamic properties of the network.
The statement network integrity should be subject to verification during the sys-
tem development phase as well as application phase. It is assumed that potentially
contradicting elements of the network should be detected during network operation,
basing on the disagreement level (9)
360 W. Cholewa
d.s/ > 0 ; (23)
as elements that are conditionally contradicting, i.e. contradicting for a specific case.
6 Conclusions
A specific class of statement networks has been introduced in which statements

values are based on a concept used for intuitionistic fuzzy sets. Statement values
include independently determined degree of validity as well as degree of invalidity
of a statement. The domain knowledge is represented in a form of a set of
necessary and sufficient conditions occurring between the statements. While the
statement network is being solved, these conditions are considered as corresponding
inequalities between statement values.
The proposed networks with implemented processes of gradual forgetting of
statement values may be used as models of dynamic objects or processes. They
may be applied, among others, in diagnostic expert systems. The main advantages
of these networks include:
• a possibility of independent development of a set of subnetworks which,
subsequently, may be connected to form a general network,
• a possibility of evaluation of the disagreement level between values of studied
statements enabling identification of erroneous data as well as errors in network
design.
Acknowledgements Described herein are selected results of study, supported partly from the bud-
get of Research Task No. 4 implemented under The National Centre for Research and Development
in Poland and ENERGA SA Strategic Program of Scientific Research and Development entitled
Advanced Technologies of Generating Energy.
References
1. Atanassov, K.T.: Intuitionistic Fuzzy Sets: Theory and Applications. Springer, Berlin
Heidelberg (1999)
2. Bargiela, A., Pedrycz, W.: Granular Computing. An Introduction. Kluwer Academic Publish-
ers, Boston (2003)
3. Cholewa, W.: Expert systems in technical diagnostics. In: Korbicz, J., Kościelny, J., Kowal-
czuk, Z., Cholewa, W. (eds.) Fault Diagnosis: Models, Artificial Intelligence, Applications,
pp. 591–674. Springer, Berlin, Heidelberg (2004)
4. Cholewa, W.: Mechanical analogy of statement networks. Int. J. Appl. Math. Comput. Sci.
18(4), 477–486 (2008) DOI 10.2478/v10006-008-0042-7
5. Cholewa, W., Amarowicz, M., Chrzanowski, P., Rogala, T.: Development environment for
diagnostic multimodal statement networks. Key Eng. Mater. 588, 74–83 (2014) DOI 10.4028/
www.scientific.net/KEM.588.74
6. Jensen, V.J.: Bayesian Networks and Decision Graphs. Springer, New York (2002)
7. Pawlak, Z.: Information systems: Theoretical foundations. Informat. Syst. 6(3), 205–218
(1981)
8. Watanabe, S.: Pattern Recognition: Human and Mechanical. John Wiley & Sons, New York
(1985)
9. Witten, I.H., Frank, E.: Data Mining: Practical Machine Learning Tools and Techniques, 2nd
edn. Morgan Kaufman, San Francisco (2005)
The Shooting Method for Non-standard
Boundary Value Problem
Rafał Palej
Abstract The paper deals with a method for solving the non-standard two-point
boundary value problem in which the number of boundary conditions is greater
than the number of first-order ODEs containing a certain number of unknown
parameters. The total number of the unknown parameters depends on the number
of boundary conditions and the way they are distributed at both ends of the domain.
The presented method consists in the replacement of the boundary value problem
by the initial value problem formulated with the use of sensitivity functions. The
general discussion has been illustrated by numerical example consisting of two first-
order differential equations and four boundary conditions. The difference between
the standard and non-standard boundary value problems consists in the size of the
initial value problem that needs to be solved.
1 Introduction
The standard boundary value problem is characterized by the same number of

first-order ODEs and boundary conditions imposed at both ends of the domain.
There are several methods for solving such problems: finite difference method,
shooting method, monotype iterative method, quasilinearization method and others
[2, 5, 1, 3]. Sometimes the number of boundary conditions is greater than the num-
ber of differential equations. Finding a solution to such problems is conditioned by
the occurrence of a certain number of unknown parameters in differential equations.
In order to find the solution satisfying all boundary conditions it is necessary
to determine the values of the unknown parameters. Some computer software
offers procedures which can solve simple problems of such type. Unfortunately,
more sophisticated problems cannot be solved with the assistance of a ready-made
procedure. This paper presents a general concept for solving the non-standard
boundary value problem in which the number of boundary conditions may be twice
as high as the number of differential equations. The discussed method consists in the
R. Palej ()
Department of Applied Computer Science, Cracow University of Technology,
Al. Jana Pawła II 37, 31-864 Cracow, Poland
e-mail: palej@mech.pk.edu.pl

364 R. Palej
replacement of the boundary value problem by the initial value problem formulated
with the use of sensitivity functions [6, 7]. The general approach to solving the non-
standard boundary value problem has been illustrated by the problem of plane curve
which should go through two given points and be inclined at given angles at these
points [4].
We shall seek the solution to the non-standard two-point boundary value problem
consisting of a set of n coupled first-order ODEs containing r unknown parameters
(1 r n). The desired solution to this problem is supposed to satisfy q boundary
conditions at the starting point and n C r q boundary conditions at the final point.
We assume that the unique solution to such problem exists. The set of n first-order
ODEs may be written in the following form:
dyi
D fi .x; y1 ; : : : ; yn ; p1 ; : : : pr / ; i D 1; : : : ; n; a x b (1)
dx
At the starting point x D a the solution is supposed to satisfy
yi .a/ D yia ; i D 1; 2; : : : ; q .q n/ (2)
while at the final point x D b it is supposed to satisfy
yi .b/ D yib ; i D 1; 2; : : : ; n C r q (3)
where yia and yib denote some constants. Adjusting appropriately the values of the
guessed parameters we should find the solution to the set of Eqs. (1), satisfying n C r
boundary conditions (2) and (3).
3 General Procedure
The method is based on the concept of the shooting method which consists in
the replacement of the boundary value problem by the initial value problem. In
order to do that we need to complete values of the desired functions yi at one
end of the domain. If, for example, q > (n C r)/2, then n q new parameters
pi , i D r C 1, : : : , n C r q should be added to complete the boundary conditions
(2) at the starting point
yi .a/ D piCrq ; i D q C 1; : : : ; n (4)

The Shooting Method for Non-standard Boundary Value Problem 365
If q D n there is no need to enlarge the number of the unknown parameters.

The desired functions yi depend on variable x as well as on n C r q unknown
parameters

yi D yi x; p1 ; : : : ; pnCrq ; i D 1; 2; : : : ; n (5)
For the trial values of the unknown parameters p(1) i , i D 1, 2, : : : , n C r q the

boundary conditions (3) are not usually fulfilled, i.e.

.1/ .1/
yi b; p1 ; : : : ; pnCrq yib ¤ 0; i D 1; 2; : : : ; n C r q (6)
In order to decrease the discrepancies in (6) for the subsequent approximation of

the guess parameters, we expand all functions with imposed conditions on the final
point into Taylor’s series

.2/ .2/ .1/ .1/
yi b; p1 ; : : : ; pnCrq D yi b; p1 ; : : : ; pnCrq
ˇ
ˇ
ˇ
X @yi ˇ
nCrq
C ˇ p
.2/
p
.1/
C D yib ; i D 1; 2; : : : ; n C r q
j D1
@pj ˇˇ x D b j j
ˇ p D p.1/
(7)
The set of Eqs. (7) may be written in the matrix form

y b; p.1/ C A b; p.1/ : p.2/ p.1/ D yb (8)
where
2 3 2 3
.1/ .1/ .2/
y1 b; p1 ; : : : ; pnCrq p1
6 7 6 7
6 7 6 .2/ 7
6 .1/ .1/
y2 b; p1 ; : : : ; pnCrq 7 6 p 7
6 7
7 ; p.2/ D 6 7
2
y b; p.1/ D 6
6 7 6 7;
6
::
7 6 :
: 7
6 : 7 6 : 7
4 5 4 5
.1/ .1/ .2/
ynCrq b; p1 ; : : : ; pnCrq p nCrq
2 .1/
3
p1
6 7
6 .1/ 7
6 p2 7
6 7
p.1/ D6 : 7
6 : 7
6 : 7
4 5
.1/
pnCrq
366 R. Palej
2 @y1 @y1 @y1

3
@p1 @p2
::: @pnCrq ˇ 2 3
6 7 ˇ y1b
6 @y2 @y2 @y2 7 ˇ 6 y2b 7
6 ::: 7 ˇ 6 7
D6 7 ˇ
@p1 @p2 @pnCrq
A b; p.1/
6 :: :: :: :: 7 ˇx D b ; yb D 6 :: 7
6 : : : : 7 ˇ 4 : 5
4 5 ˇp D p.1/
@ynCrq @ynCrq
:::
@ynCrq y.nCrq/b
@p1 @p2 @pnCrq
From Eq. (8) we can derive the iteration formula for the subsequent approxima-
tions of the parameter values in the form

p.kC1/ D p.k/ A1 b; p.k/ : y b; p.k/ yb ; k D 1; 2; : : : (9)
The Jacobian matrix A(b, p(1) ) is composed of derivatives which must be deter-
mined. To this end, we differentiate Eqs. (1) with respect to individual parameters
X
n
@ dyi @fi @yk @fi
D C ; i D 1; ::; n; j D 1; : : : ; n C r q (10)
@pj dx @yk @pj @pj
kD1
By defining the sensitivity functions with the following formulas:
@yk
gkj D ; k D 1; ::; n; j D 1; : : : ; n C r q; (11)
@pj
we may express Eqs. (10) in the form
dgij X @fi n
@fi
D gkj C ; i D 1; ::; n; j D 1; : : : ; n C r q (12)
dx @yk @pj
kD1
The initial conditions for sensitivity functions (found from Eqs. (2), (4) and (11))
shall have the form

@yk .a/ 0; k D 1; : : : ; q; j D 1; : : : ; n C r q
gkj .a/ D D (13)
@pj ık;j Cqr ; k D q C 1; : : : ; n; j D 1; : : : ; n C r q
where ı k,j C q r is the Kronecker delta. Finally, in each iteration we have to solve the
initial value problem consisting of the system of n (n C r q C 1) first-order ODEs,
(1) and (12), and the same number of initial conditions, (2), (4) and (13). After
determining the values of all parameters from Eq. (9) with the assumed accuracy,
we shall obtain the solution to the non-standard boundary value problem, defined by
Eqs. (1)–(3), by solving the initial value problem consisting of Eqs. (1) and initial
conditions (2) and (4). The solution to this initial problem shall at the same time
constitute the solution to the boundary value problem formulated by Eqs. (1)–(3).
The presented method of determining the values of the unknown parameters, which
is in fact the Newton-Raphson method, is characterized by quadratic convergence
on condition that the first approximation is sufficiently close to the desired solution.
4 Numerical Example
To illustrate the presented method we shall solve a problem consisting of second-

order differential equation and four boundary conditions. The differential equation
relates the normal reaction force acting on a particle during its slide down the curve
to the function y(x) and its derivatives. If we assume that the normal reaction force
N is a linear function of x (N D p1 p2 x), then the differential equation takes the
following form [4]:
3
0 2 2
2
d y 1 .p 1 p 2 x/ 1 C .y / .y 0 /2 1
D (14)
dx 2 2 h C h0 y
where h denotes the height of the starting point while h0 denotes theheight above
p
starting point to provide the prescribed initial velocity v0 D 2gh0 .
Boundary conditions determine the coordinates of the two points the curve
should go through and the angles of tangents at these points. They should therefore
adopt the following form:

y.0/ D 0
(15)
y 0 .0/ D tan .˛/

y.1/ D h
(16)
y 0 .1/ D tan .ˇ/
To apply the presented method, it is necessary to replace the second-order

differential equation (14) with two first-order equations:
dy1
D y2
dx
3
dy2 1 .p1 p2 x/ 1 C y22 2 y22 1
D (17)
dx 2 h C h0 C y1
In this example we have n D r D q D 2. Now, the differential equations (12) will
take the following form:
dg1;1 dg2;1
D g2;1 ; D h1 .x/g1;1 C h2 .x/g2;1 C h3 .x/
dx dx
dg1;2 dg2;2
D g2;2 ; D h1 .x/g1;2 C h2 .x/g2;2 x h3 .x/ (18)
dx dx
368 R. Palej
where
3
1 .p1 p2 x/ 1 C y22 2 y22 1
h1 .x/ D
2 .h C h0 y1 /2
1
1 3 .p1 p2 x/ 1 C y22 2 y2 2y2
h2 .x/ D
2 h C h0 y1
3
1 1 C y22 2
h3 .x/ D
2 h C h0 y1
The initial conditions (2) and (13) in this case yield
y1 .0/ D 0; g1;1 .0/ D 0; g1;2 .0/ D 0

y2 .0/ D tan .˛/ ; g2;1 .0/ D 0; g2;2 .0/ D 0 (19)
For h D h0 D 0.5, tan(˛) D 0.2, tan(ˇ) D 1 and p(1) (1)

1 D 2.0, p2 D 0.5 we obtain
(5) (5)
after 4 iterations p1 D 1.869525017658811 and p2 D 0.3544242522592831. For
such values of the parameters, solving the non-standard boundary value problem
amounts to solving the initial value problem consisting of Eq. (14) and initial
conditions (15). The plot of the desired curve y(x) is shown in Fig. 1.
One of the important questions arising during the process of solving boundary
value problems of this type concerns the region of the solution convergence. Each
problem with a unique solution is characterized by a specific region of convergence.
The convergence region calculated for simple examples may be unbounded. The
region of convergence for parameters p1 and p2 is limited and its shape is shown in
Fig. 2.
If the starting point is placed inside the region of convergence, then the sequence
of subsequent approximations will converge with the solution, marked by a cross in
Fig. 2.
Fig. 1 The plot of the solution of the initial value problem given by (14)(15)
Fig. 2 The region of convergence for parameters p1 and p2
5 Conclusions
The method presented in the paper is based on Newton’s approach to iterative

determination of parameter values pi . The method allows solving the non-standard
boundary value problem in which the number of boundary conditions may exceed
the number of first-order differential equations. The difference between the standard
and non-standard boundary value problems consists in the size of the initial value
problem that needs to be solved in each iteration. When r D n the number of
sensitivity functions necessary for determination of the Jacobian matrix equals n2
which means that the initial value problem will consist of n (n C 1) first-order
ODEs.
References
1. Asher, U.M., Mattheij, R., Russel, R.: Numerical Solution of Boundary Value Problems for
Ordinary Differential Equations, SIAM, Philadelphia, 2nd edn. University City Science Center,
Philadelphia (1995)
2. Cuomo, S., Marasco, A.: A numerical approach to nonlinear two-point boundary value problems
for ODEs. Comput. Math. Appl. 55, 2476–2489 (2008)
3. Keller, H.B.: Numerical Solutions of Two Point Boundary Value Problems, SIAM, Philadelphia,
4th edn. University City Science Center, Philadelphia (1990)
4. Palej, R., Filipowska, R.: Mathematical modeling of the Inrun profile of a ski jumping hill with
the controlled track reaction force. J Theor. Appl. Mech. 47(1), 229–242 (2009)
370 R. Palej
5. Press, W.H., Teukolsky, S.A., Vetterling, B.F.: Numerical Recipes. The Art of Scientific
Computing, 3rd ed. Cambridge University Press, New York (2007)
6. Rao, S.S.: Applied Numerical Methods for Engineers and Scientists. Prentice Hall, New Jersey
(2002)
7. Sung, N.H.: A nonlinear shooting method for two-point boundary value problems. Comput.
Math. Appl. 42, 1411–1420 (2000)
Automatic Sleep Scoring from a Single Electrode
Using Delay Differential Equations
Claudia Lainscsek, Valérie Messager, Adriana Portman, Jean-François Muir,

Terrence J. Sejnowski, and Christophe Letellier
Abstract Sleep scoring is commonly performed from electroencephalogram

(EEG), electrooculogram (EOG), and electromyogram (EMG) to produce a so-
called hypnogram. A neurologist thus visually encodes each epoch of 30 s into one
of the sleep stages (wake, REM sleep, S1 , S2 , S3 , S4 ). To avoid such a long process
(about 3–4 hours) a technique for automatic sleep scoring from the signal of a single
EEG electrode located in the C3 /A2 area using nonlinear delay differential equations
(DDEs) is presented here. Our approach considers brain activity as resulting from
a dynamical system whose parameters should vary according to the sleep stages.
It is thus shown that there is at least one coefficient that depends on sleep stages
and which can be used to construct a hypnogram. The correlation between manual
hypnograms and the coefficient evolution is around 80%, that is, about the inter-rater
variability. In order to rank sleep quality from the best to the worst, we introduced
a global sleep quality index which is used to compare manual and automatic sleep
scorings, thus using our ability to state about sleep quality that is the final goal for
physicians.
1 Introduction
Up to 2007, polysomnographic recordings were scored into sleep stages according

to the rules introduced by Rechtschaffen and Kales [19] which are mainly based on
a spectral analysis. The scoring, accomplished by well-trained neurologist, consists
in scoring all 30 s epochs into one of the six stages of vigilance, namely awakeness,
C. Lainscsek () • T.J. Sejnowski

Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, CA 92037, USA
e-mail: claudia@salk.edu; terry@salk.edu
V. Messager • C. Letellier
Université de Rouen, CORIA, Avenue de l’Université, BP 12, F-76801 Saint-Etienne
du Rouvray cedex, France
e-mail: valerie.messager@coria.fr; christophe.letellier@coria.fr
A. Portman • J.-F. Muir
Hôpital de Bois-Guillaume, Bois-Guillaume, France
e-mail: adrianaportmann@yahoo.fr; jean-francois.muir@chu-rouen.fr

372 C. Lainscsek et al.
rapid eyes movement sleep (REM), and sleep stages S1, S2, S3, and S4. RK rules
were recently modified to overcome the inter-rater variability ([11]). The most
important change was that stages 3 and 4 merged into a single stage, named slow-
wave sleep or N3. In spite of that, recent studies only showed slight improvements
with the new rules ([6]) with an inter-rater agreement slightly greater than 72% ([3]).
Automatic sleep scoring techniques are thus welcome. Most of the computer-
assisted scoring techniques stages were based on RK rules ([10, 12, 18]). In fact,
most of them try to reproduce what is done by neurologists and which can lead
to an overall epoch-by-epoch agreement of 80%, and require a quite complex
decisional tree (see Fig. 2 in [2]). With the emergence of “chaos theory,” recurrence
plots quantifiers, Lyapunov exponents, or correlation dimension were used to
obtain hypnograms with an overall agreement which was rarely greater than 60 or
70% ([23]).
Neural networks were also used to distinguish different features exhibited in
the spectral domain but were not able to distinguish more than the REM sleep
from non-REM sleep ([9]). Another technique was correctly scoring sleep stages
but required two EEG channels, one horizontal electrooculogram channel and one
chin electromyogram channel ([20]). An automatic sleep classification was able to
distinguish wake, slow-wave sleep and rapid eye movements sleep stages ([22]),
but a specific sensor, a head accelerometer, was required and must be added to
conventional sensors.
Our aim is to develop a reliable automatic technique using a single EEG signal for
scoring hypnograms. The subsequent part of this paper is organized as follows. In
Sect. 2 the pool of patients which were recorded is described. Section 3 is devoted to
our automatic sleep scoring technique and to a new global sleep quality index used
to rank a set of hypnograms. In Sect. 4 the results are presented and Sect. 5 gives a
conclusion.
2 Patients
This retrospective observational study was conducted at the sleep laboratory at

the medical university hospital Intensive Care Unit in Rouen. We selected 38
recordings, but only 35 were associated with a reliable sleep scoring. These patients
were long-term ventilated for chronic respiratory failure and grouped into two
types. The first type corresponds to an obesity hypoventilation syndrome (OHS)
commonly seen in severely overweight people who fail to breathe normally resulting
in low blood oxygen levels and high blood carbon dioxide (CO2 ) levels. Many
of these patients have increased upper airway resistances during sleep (obstructive
sleep apnea). This induces a significant amount of wake after sleep onset (WASO)
leading to abnormal daytime sleepiness. This disease puts strain on the heart,
possibly resulting in heart failure, leg swelling, and various other related symptoms.
The second group of respiratory failure, considered here, is associated with chronic
obstructive pulmonary disease (COPD). This refers to small airway obstructions
Automatic Sleep Scoring Using Delay Differential Equations 373
Table 1 Main clinical Demographics and respiratory parameters Mean (SD)

characteristics of the patients
(n D 34) Age (year) 64.5 (11.7)
Male:female 24:11
Body mass index (kg.m2 ) 42.0 (10.5)
PaO2 (cmH2 O) 9.5 1.1
PaCO2 (cmH2 O) 5.8 (0.9)
Normal values: (10:7 < PaO2 < 12:0) cmH2 O, PaCO2
5:3 cmH2 O, (18:5 < BMI < 25) kg.m2 and obesity is
defined by BMI> 30 kg.m2
and emphysema, two commonly coexisting pulmonary diseases in which the

airways progressively narrow inducing shortness of breath. In these patients, the
airflow limitation is usually nonreversible when treated with bronchodilators and
progressively becomes more and more severe. One efficient treatment is to put these
patients under noninvasive mechanical ventilatory assistance. In the present case, all
patients were ventilated with the bilevel ventilator RESMED VPAP III. All patients
included in this study were in stable condition, as assessed by clinical examination
and arterial blood gases.
Main characteristics of the thirty-five patients for which the sleep was scored
during one night under mechanical ventilation are reported in Table 1. Twenty
patients (57%) had OHS and 15 patients (43%) had COPD. Thirteen patients (38%)
were diagnosed with obstructive sleep apnea syndrome (defined as more than 10
apneas per hour). Upon study inclusion, the patients were ventilated for a few
months. Nineteen patients (56%) were hypercapnic (PaCo2 > 5:6 cmH2 O).
3 Method
3.1 Automatic Sleep Scoring
A nonlinear delay differential equation has the general form

xP D a1 x1 C a2 x2 C a3 x3 C C ai1 xn C ai x21 C aiC1 x1 x2
CaiC2 x1 x3 C C aj 1 x2n C aj x31 C aj C1 x21 x2 C : : : C al xmn
(1)
where x D x.t / and xj D x.t j /. In this general form, the DDE has n delays,
l monomials with their corresponding coefficients ai , and a degree of nonlinearity
equal to m. In the subsequent part of this paper, we will define a k-term DDE as an
equation with only k < l monomials selected from the right-hand side of the general
form (1). As for any global modeling technique, there is a significant improvement
of capturing main characteristics of the underlying dynamics from observed data by
carefully selecting the structure of the DDE model ([1, 14–16]). The minimal mean
squared error is used for this process. By structure selection, we mean retaining
only monomials in the DDE that have the most significant contribution to classify
the data. An equally important task is to select the right time-delays, since linear
terms are directly related to the fundamental timescales and nonlinear terms to the
nonlinear couplings between them ([16]). This can be performed by using a genetic
algorithm ([15]) or by an exhaustive search for the best model among the general
form with n D 2 and m D 3 resulting in l D 9 monomials as performed in [16].
Here only models with up to three terms were considered (see Table 2 in [16]).
The variable x corresponds to the signal provided by the electrode located in the
C3 /A2 area of the scalp. We ran a genetic algorithm to minimize the least square
error of 30 s data windows to select the best models and delays for each 30 s window
([8, 15]). For 95% of the data windows (corresponding to the 35 patients), the four
models
xP D a1 x1 C a2 x2 C a3 x21 I (2)
xP D a1 x1 C a2 x2 C a4 x1 x2 I (3)
xP D a1 x1 C a2 x2 C a6 x31 I (4)
xP D a1 x1 C a2 x2 C a7 x21 x2 I (5)
were selected as well as delays between 1 ıt and 4 ıt with ıt D 64 1

s. Among
these four models, model (5) is the best to distinguish wake, REM, and S1 from the
sleep stages S2, S3, and S4 (see left panel from Fig. 1). Delay 1 D 1 is useful to
distinguish wake, S2, S3, and S4 from REM and S1 (right panel from Fig. 1). Delay
2 D 3 allows to distinguish wake from sleep stages. Thus, combining model (5)
with delays 1 D 1 and 2 D 3 provides the model with the most discriminative
ability. Among the three coefficients of model (5), parameter a2 was found to be the
most correlated (r D 0:95) to the manually scored hypnogram, as exemplified in
Fig. 2 in the case of patient 15. We then used this model and this coefficient to score
the sleep for our 35 patients.
It was then necessary to convert the a2 -time series which corresponds to the time
evolution of a real number sampled at 0.1 Hz (one point per 10 s) into a sequence of
integers from 1 (stage S1) to 6 (awake). This is the tricky part of our technique. In
the case of patient 15, we got an automatically scored hypnogram which was quite
close to the manually scored one (Fig. 2b).
3.2 Assessing the Sleep Quality
Since patients with chronic respiratory failures are ventilated during their sleep,
it is important to assess whether the ventilation improves the sleep quality or,
Fig. 1 Histograms of the number of time each of the four selective DDEs (left) and each delays
(right) were selected with minimum error for each sleep stage
a 0,4 b
0,3 Awake
Coefficient a2
0,2 REM Sleep

Automatic
S1 manual
0,1
S2
0
S3
-0,1
S4
-0,2
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time (hour) Time (h)
Fig. 2 Time series of coefficient a2 of the delay differential equation (5) and the corresponding
hypnogram. Case of patient 15 (male, 76 years, BMID50 kg.m2 ). The manually scored hypno-
gram (green) is also reported for comparison. (a) Raw a2 time series (b) Sequence of integers
at least, that it does not degrade it. In order to do that, it is necessary to be

able to rank hypnograms according to sleep quality. From a subjective point of
view, sleep quality refers to patient feelings about the refreshing effect of sleep
which can be assessed using some sleep diary or the Pittsburgh Quality Index
([4]). The characteristics commonly taken into account in such evaluation are sleep
latency, sleep duration, regular sleep efficiency, sleep disturbances (including sleep
disruptive events such as snoring, apnea, or pains), use of sleeping medication, and
daytime dysfunction ([4]).
Up-to-now, the objective evaluation of sleep quality was based on the same
characteristics but directly measured from hypnograms ([11]). Also considered
are the arousal index (number of arousals per hour) and the number of various
respiratory events. To assess the evolution of sleep quality, all these quantities are
then subjectively combined and compared since none of them can alone allow
to rank hypnograms according to sleep quality (see [17] for details). In order to
avoid this last subjective step, we introduced a new index which combines the
most important sleep characteristics. Thus, our global sleep quality index takes into
account the number of sleep cycles (each cycle, between 90 and 120 min, contains
some slow-wave sleep restoring physical functions and some rapid eye movements
restoring cognitive functions), the fraction of WASO, the fraction of stable sleep,
the number of micro-arousals, and the number of stage transitions. The global sleep
quality index GSQ is defined as
GSQ D cycle restoring stability .1 M frag / .1 frag / (6)

N
where cy D Max 6cy ; 1 and Ncy is the number of sleep cycles that saturates
to one when it exceeds 6 cycles; the restoring capacity of sleep is evaluated
according to

5 S3 C S4 C R
restoring D Min ;1 (7)
2 S1 C S2 C S3 C S4 C R
with i being the time duration spent in the i th sleep stage (i D S1, S2, S3, S4, and
R) and saturates to 1 when the restorative sleep (S3, S4, and R) exceeds 25 of the
effective sleep; the sleep stability is evaluated according to
0 0 0 0
S1 C S2 C S3 C S4 C R0
stability D (8)
effective sleep
with i0 being the time spent in the i th sleep stage without any micro-arousal and
not corresponding to an epoch connexe to a stage transition, and effective sleep being
the time duration of sleep stages (S1 C S2 C S3 C S4 C R ); the sleep macro-
fragmentation is evaluated according to
waso
M frag D I (9)
waso C effective sleep
the sleep micro-fragmentation is evaluated according to

0 0 0 0
.S1 S1 / C .S2 S2 / C .S3 S3 / C .S4 S4 / C .R R0 /
frag D
effective sleep
(10)
with i i0 being the time spent in an epoch of the i th sleep stage with a micro-
arousal or connection to a stage transition.
4 Results
The time series of coefficient a2 were found quite well correlated to the correspond-
ing hypnograms (r D 0:86 ˙ 0:1). To assess the quality of our sleep scorings using
the coefficient a2 we computed the confusion matrix ([13]) which is a specific table
layout used to assess performance of classifier. Each column of the matrix represents
the instances in a predicted class, while each row represents the instances in an
actual class. The confusion matrix for all epochs of all patients is reported in Fig. 3.
To get a graphical representation the numbers were also converted to a percentage.
A dark diagonal from the upper-left corner to the lower-right corner with all other
squares in white would indicate perfect scoring of each data window into the correct
sleep stage.
As additional measure of performance we used Cohen’s kappa [5, 7, 21] which
a pe
can be computed directly from the confusion matrix as [13]. D p1p e
, where
Pq Pq
pa D pkk , and pe D pkC pCk where q D 6 for the 6 classes, pa is the
kD1 kD1
observed percentage of agreement, pe is the expected percentage of agreement, pkC
is the percentage of actual classification, and pCk is the percentage of predicted
classification. We got D 0:51 ˙ 0:1 when comparing automatically scored
hypnograms with the manually scored ones. Detailed results are reported in Table 2.
The global sleep quality index GSQ was first computed from the hypnograms
scored by the neurologist. Patients were then ranked according to a decreasing GSQ
(Fig. 4). The hypnogram of the patient with the largest GSQ (35.4 %) is shown
in Fig. 5a: it presents 3 sleep cycles quite well structured. Contrary to this, the
hypnogram of patient 22 with the smallest GSQ (0.1 %) is shown in Fig. 5b: it does
not present a single well-structured sleep cycle and the effective sleep time duration
is small (effective sleep D 146 min).
The rates of each sleep stage was computed for each hypnograms which were
ranked according to decreasing GSQ (Fig. 6). The best hypnogram (patient 34,
GSQ D 35:4) presents a good proportion of restorative sleep. Contrary to this,
predicted
100
W R 1 2 3 4 62 34 3 1 0 0 90
W 3026 1642 147 29 1 4 10 55 32 3 0 0

80
70
R 225 1288 762 62 5 11 60
1 9 49 31 8 2
1 61 733 3841 2388 607 164 50
actual
0 1 10 36 27 27 40
2 7 13 191 703 524 528 30

0 1 13 26 56 3
3 1 34 394 790 1669 97 20
10
0 1 2 5 10 82
4 27 54 166 557 1050 8615 0
Fig. 3 Confusion matrix for all subjects: the table on the left side shows the numbers of predicted
and actual sleep stage windows. The plot on the right side visualizes the percentage of predicted
and actual sleep stage windows
Table 2 Correlation coefficient r and Cohen’s between the manually scored hypnograms and
the time series of coefficient a2 of model (5) for each subject
# r # r # r # r # r
1 0.82 0.36 9 0.91 0.53 17 0.70 0.28 24 0.95 0.65 32 0.91 0.55
2 0.95 0.61 11 0.80 0.36 18 0.81 0.44 25 0.78 0.41 33 0.84 0.41
3 0.89 0.59 12 0.90 0.64 19 0.87 0.53 26 0.82 0.50 34 0.93 0.66
5 0.91 0.63 13 0.91 0.57 20 0.78 0.59 27 0.92 0.64 35 0.82 0.37
6 0.92 0.51 14 0.76 0.36 21 0.79 0.39 29 0.94 0.61 36 0.89 0.55
7 0.92 0.68 15 0.95 0.67 22 0.80 0.40 30 0.87 0.51 37 0.91 0.59
8 0.79 0.43 16 0.90 0.54 23 0.80 0.41 31 0.79 0.43 38 0.91 0.51
40
Global Quality Sleep Inde (in %)
35
30
25
20
15
10
5
0
34 35 24 15 7 17 21 23 13 3 29 33 31 9 30 18 5 6 8 38 26 37 1 14 11 16 36 12 19 25 27 32 2 20 22
Patient Number
Fig. 4 Global sleep quality index computed from the manually scored hypnograms for the 35
patients of our protocol
the worst hypnogram (patient 22, GSQ D 0:1) associated with a very small fraction
of restorative sleep and a large one of WASO. Hypnograms are rather well ranked
since the rate of WASO and sleep micro-fragmentation are anticorrelated to GSQ
(r D 0:65, p < 0:0001 and r D 0:75, p < 0:0001, respectively). The rate
of slow-wave sleep (S3 and S4) and the rate of REM sleep are correlated to GSQ
(r D 0:83, D< 0:0001 and r D 0:59, p < 0:0001, respectively). These features
and others that are outside the scope of this paper correspond to an increase of the
sleep quality with GSQ .
We now computed the global sleep quality index from the automatically scored
hypnograms with our technique (Fig. 7). They were ordered in a slightly different
order than the manual hypnograms. In order to quantify this disagreement between
these two orders, let us designate by n the rank (n0 ) the rank obtained by computing
GSQ from the manual (automatic) hypnograms. Thus n D jn n0 j corresponds
to the rank shift observed between these two orders. We thus have n D 4:6 ˙ 5:4,
meaning that, in average, the good (bad) hypnograms remain the good (bad) ones.
There are four notable exceptions with the hypnograms for patients 11, 15, 24, and
35 for which n equals to 23, C15, C20, and +11, respectively.
a Manual scoring Automatic scoring

Awake Awake
REM Sleep REM Sleep
S1 S1
S2 S2
S3 S3
S4 S4
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
b
Awake Awake
REM Sleep REM Sleep
S1 S1
S2 S2
S3 S3
S4 S4
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Fig. 5 Hypnograms for two of the 35 patients corresponding to the largest and the smallest global
sleep quality index. The gender, age, body mass index, and the rate of synchronous breathing cycles
are also reported. (a) Patient 34 : male, 82 years, BMID44.1, 2.1% of asynchronous cycles, and
GSQ D 35:4%. (b) Patient 22 : male, 83 years, BMID36.3, 8.0% of asynchronous cycles, and
GSQ D 0:1%
120
100
Rate of sleep stage (%)
80
%I WASO
% REM
60
% S3+S4
% S2
40
% S1
20
0
34 35 24 15 7 17 21 23 13 3 29 33 31 9 30 18 5 6 8 38 26 37 1 14 1116 36 12 19 25 27 32 2 20 22
Patient number
Fig. 6 Fraction of time duration of each sleep stage. Patients are ranked according to the global
sleep quality index GSQ
The manually scored hypnogram of patient 11 (Fig. 8a) presents many fluctua-
tions between wake and stage S1 and a very few epochs in stages S3 or S4 and
REM sleep, thus associated with a small global quality sleep index (GSQ D 3:7%).
The evolution of the coefficient of the DDE fluctuates a lot between the values
corresponding to wake and S1 stages. Consequently, since REM sleep is between
these two stages from EEG, our technique returns too often REM sleep (and not
WASO). This is significantly increasing the global sleep quality index to 24.9.
Sleep Global Quality Index (in %)

30
25
20
15
10
0
23 11 34 7 33 13 17 21 18 31 8 9 35 5 30 3 29 15 26 38 36 6 24 14 37 12 16 1 22 19 25 27 32 2 20
Patient Number
Fig. 7 Global sleep quality index computed from the automatically scored hypnograms for the 35
patients of our protocol
a Manual scoring Automatic scoring

Awake Awake
REM Sleep REM Sleep
S1 S1
S2 S2
S3 S3
S4 S4
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time (h) Time (h)
Patient 11 : Male, 78 years, BMI=32.0, 5.5% of asynchronous cycles, ηGSQ = 35.4%
b
Awake Awake
REM Sleep REM Sleep
S S
1 1
S S
2 2
S S
3 3
S S
4 4
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time (h) Time (h)
Patient 24 : Male, 62 years, BMI=50.2, 1.7% of asynchronous cycles, ηGSQ = 0.1%
Fig. 8 Hypnograms for two badly scored using our automatic technique
It is important to note that a neurologist uses a lot the electrooculogram and the
electromyogram to distinguish REM sleep from awake and S1, two signals which
are not considered by our technique.
Contrary to this, the automatically scored hypnograms for patient 24 is char-
acterized by a global sleep quality index GSQ D 7:0% is significantly smaller
than the value (16.2%) obtained from the manual hypnograms (Fig. 8b). There are
few reasons explaining such a large departure between these two GSQ -values. The
global sleep duration (between the first and the last sleep epoch) is larger than the
one obtained from the automatic scoring (221.5 min and 198 min, respectively),
but the number of sleep cycles is 2 in both cases. The rate of WASO in the
automatic hypnogram is about three times the rate obtained from the manually
scored hypnogram (19.9 and 6.6, respectively). The rate of micro-fragmentation
obtained with our technique is about three times the rate returned by the neurologist
(31.8 and 11.1, respectively). The stability is smaller in the hypnogram provided by
our technique than in the one scored by the neurologist (38.2 and 58.3, respectively).
All these modifications tend to increase the global sleep quality index.
5 Conclusions
In 88% of subjects the overall sleep quality index computed from the DDE
hypnograms are in agreement with the sleep quality index computed from the
visually scored hypnograms. The difference in 12% of all patients results from
converting the real number outputs of the DDE to the integers used for indexing
sleep stages (S1, S2, S3, S4, R, and wake). This is the weakest part of the
present version of our technique. In spite of this, our hypnograms are already
sufficiently close to the manual hypnograms that are used to assess the sleep quality.
Importantly, this first study has led to the identification of possible improvements
that are currently being developed.
Our automatic scoring technique using DDEs is well correlated to the corre-
sponding visually scored hypnograms (r D 0:86 ˙ 0:1). This excellent agreement
becomes even more impressive when considering the use of only one scalp electrode
for the DDE method. Indeed, the most promising aspect of our technique is that only
one scalp electrode is sufficient to accurately score sleep stages.
References
1. Aguirre, L.A., Billings, S.A.: Improved structure selection for nonlinear models based on term
clustering. Int. J. Contr. 62(3), 569–587 (1995)
2. Anderer, P., Gruber, G., Parapatics, S., Woertz, M., Miazhynskaia, T., Klösch, G., Saletu,
B.,Zeitlhofer, J., Barbanoj, M.J., Danker-Hopfe, H., Himanen, S.-L., Kemp, B., Penzel, T.,
Grözinger, M., Kunz, D., Rappelsberger, P., Schlögl, A., Dorffner, G.: An E-health solution for
automatic sleep classification according to Rechtschaffen and Kales: validation study of the
somnolyzer 24x7 utilizing the SIESTA database. Neuropsychobiology 51(3), 115–133 (2005)
3. Basner, M., Griefahn, B., Penzel, T.: Inter-rater agreement in sleep stage classification between
centers with different backgrounds. Somnologie 12,75–84 (2008)
4. Buysse, D.J., Reynolds, C.F., Monk, T.H., Berman, S.R., Kupfer, D.J.: The Pittsburgh sleep
quality index: a new instrument for psychiatric practice and research. Psychiatric Res. 28, 193–
213 (1989)
5. Cohen, J.: A coefficient of agreement for nominal scales. Educ. Psychol. Meas. 20(1), 37–46
(1960)
6. Danker-Hopfe, H., Anderer, P., Zeitlhofer, J., Boeck, M., Dorn, H., gruber, G., Heller, E.,
Loretz, E., Moser, D., Parapatics, S., Saletu, B., Schmidt, A. Dorfner, G.: Interrater reliability
for sleep scoring according to the Rechtschaffen and Kales and the new AASM standard.
J.Sleep Res. 18(1), 74–84 (2009)
7. Fleiss, J.L., Cohen, J.: The equivalence of weighted kappa and the intraclass correlation
coefficient as measures of reliability. Educ. Psychol. Meas. 33, 613–619 (1973)
8. Goldberg, D.E.: Genetic Algorithms in Search, Optimization and Machine Learning. Addison-
Wesley, Wokingham (1998)
9. Grötzinger, M., Wolf, C., Uhl, T., Schäffner, C., Röschke, J.: Online detection of REM sleep
based on the comprehensive evaluation of short adjacent EEG segments by artificial neural
networks. Progr. Neuro-Psychopharmacol. Biol. Psychiatr. 21(6), 951–963 (1997)
10. Harper, R.M., Schechtman, V.L., Kluge, K.A.: Machine classification of infant sleep state using
cardiorespiratory measures. Electroencephalogr. Clin. Neurophysiol. 67(4), 379–387 (1987)
11. Iber, C., Ancoli-Israel, S., Chesson, A., Quan, S.F. (eds.): The AASM Manual for the Scoring
of Sleep and Associated Events: Rules, Terminology, and Technical Specification. American
Academy of Sleep Medicine, Westchester (2007)
12. Jansen, B.H., Dawant, B.M.: Knowledge-based approach to sleep EEG analysis-a feasibility
study. IEEE Trans. Biomed. Eng. 36(5), 510–518, (1989)
13. Kohavi, R., Provost, F.: Glossary of terms. Mach. Learn. 30(2/3), 271–274 (1998)
14. Lainscsek, C., Letellier, C.,Gorodnitsky, I.: Global modeling of the Rössler system from the
z-variable. Phys. Lett. A, 314 409–127 (2003)
15. Lainscsek, C., Rowat, P., Schettino, L., Lee, D., Song, D., Letellier, C., Poizner, H.: Finger
tapping movements of Parkinson’s disease patients automatically rated using nonlinear delay
differential equations. Chaos, 22, 013119 (2012)
16. Lainscsek, C., Sejnowski, T.J.: Electrocardiogram classification using delay differential equa-
tions. Chaos 23(2), 023132 (2013)
17. Messager, V., Portmann, A., Muir, J.-F., Letellier, C.: A global sleep quality index for ranking
hypnograms. in preparation (2013)
18. Principe, J.C., Gala, S.K., Chang, T.G.: Sleep staging automaton based on the theory of
evidence. IEEE Trans. Biomed. Eng. 36(5), 503–509 (1989)
19. Rechtschaffen, A., Kales, A. (eds.): A Manual of Standardized Terminology, Techniques and
Scoring System for Sleep Stages of Human Subject. US Government Printing Office, National
Institute of Health Publication, Washington (1968)
20. Schaltenbrand, N., Lengelle, R., Toussaint, M., Luthringer, R., Carelli, G. Jacqmin, A., Lainey,
E., Muzet, A., Macher, J.P.: Sleep stage scoring using the neural network model: comparison
between visual and automatic analysis in normal subjects and patients. Sleep 1, 26–35 (1996)
21. Scott, W.: Reliability of content analysis: The case of nominal scale coding. Publ. Opin. Q.
19(3), 321–325 (1955)
22. Sunderam, S., Chernyy, N., Peixoto, N., Mason, J.P., Weinstein, S.L., Schiff, S.J., Gluck-
man, B.J.: Improved sleep-wake and behavior discrimination using MEMS accelerometers.
J.Neurosci. Meth. 163(2), 373–383 (2007)
23. Šušmáková, K., Krakovská, A.: Discrimination ability of individual measures used in sleep
stages classification. Artif. Intell. Med. 44(3), 261–277 (2008)
Methods for the Quick Analysis of Micro-chaos
Gergely Gyebrószki and Gábor Csernák
Abstract Micro-chaos is a phenomenon when sampling, round-off and processing

delay (shortly, digital effects) lead to chaotic oscillations with small amplitude. In
previous works [1], the so-called micro-chaos maps of various digitally controlled
unstable linear mechanical systems were derived and the possibility of the coexis-
tence of several disconnected attractors was highlighted. The typical size of these
attractors is usually negligible from the practical point of view, but the distance
of the farthest attractor from the desired state can be rather large. This is why
the phenomenon of micro-chaos can be the source of significant control error. In this
paper, a set of numerical methods (e.g. cell mapping techniques for the exploration
of the phase-space structure) is assembled in order to create a toolkit for the quick
analysis of micro-chaotic behaviour. The elaborated methods are tested on models
of PD-controlled unstable systems and the practically important characteristics of
chaotic behaviour are determined.
1 Introduction
Chaotic vibrations arising due to the digital effects in control are well known for
the last 20 years [3, 5]. Sampling and delay are commonly taken into account in
control problems; however, rounding is usually neglected. It has been shown [4, 5]
that rounding leads to small amplitude chaotic oscillations—referred to as micro-
chaos because of the small amplitude—where several disconnected attractors may
coexist. In a couple of cases (inverted pendulum stabilized with D control without
delay [5], D control with delay [1] and PD control without delay [2]), it has been
rigorously proven that the vibrations are indeed chaotic. In [2], the coexistence of
G. Gyebrószki ()
Budapest University of Technology and Economics, Department of Applied Mechanics,
H-1111 Budapest, Műegyetem rkp 5., Hungary
e-mail: gyebro@mm.bme.hu
G. Csernák
HAS-BUTE Research Group on Dynamics of Machines and Vehicles, H-1111 Budapest,
Műegyetem rkp 5., Hungary
e-mail: csernak@mm.bme.hu

384 G. Gyebrószki and G. Csernák
chaotic attractors was already shown. Since chaotic behaviour is generally found
in theoretical models of digitally controlled systems, in this paper we are focusing
on numerical methods to characterize and examine chaotic behaviour. Section 2
introduces the digitally controlled inverted pendulum—as the subject of chaos
investigation—while in Section 3 we give a short overview of methods used for
characterizing chaotic behaviour. Section 4 shows different scenarios to select the
region of interest for numerical methods. Section 5 is devoted to describe the
simple cell mapping (SCM) method and other supporting methods (such as fractal
dimension calculation) for investigating chaotic behaviour and shows the SCM
results for various parameters.
2 System Under Examination
The pendulum-on-a-cart is one of the simplest devices used for demonstrating

control problems. Neglecting the cart part of this device leads to a simple inverted
pendulum with control. In this section, the equation of motion of the digitally
controlled inverted pendulum is presented and the derivation of the so-called micro-
chaos map is shown. The map corresponding to the inverted pendulum is used to
illustrate the application of numerical methods described in the next section.
2.1 Inverted Pendulum with Damping and PD Control
Consider an inverted pendulum with damping and digitally implemented PD control

with zero-order hold (i.e. the control torque is kept constant between two successive
sampling instants) as shown in Fig. 1. The actuator (the neglected cart) is considered
to be ideal, i.e. produces the desired control effort immediately. The equation of
motion of this system is
φi , φi
M
g φ
3
PC
2
C M
1
h
k
t
ti–1 ti ti+1 ti+2
Fig. 1 The digitally controlled inverted pendulum and the control torque with respect to time
Methods for the Quick Analysis of Micro-chaos 385
R / D mgh sin.'.t // k '.t

J '.t P / p 'i d 'Pi ; t 2 Œi ; .i C 1//; (1)
where m is the mass of the pendulum, J is the mass moment of inertia about
the axis of rotation, h is the distance between the centre of mass and the axis of
rotation, p and d are control parameters, k is the linear damping coefficient, g
is the gravitational acceleration and is the sampling time, while 'i D '.i /
and 'Pi D '.i
P / are sampled values of the angular position and angular velocity,
respectively (at the beginning of the i th time interval). Rearranging, linearizing and
dividing (1) by J yields
R / C 2ˇ '.t
'.t P / ˛ 2 '.t / D P 'i D 'Pi ; (2)
where ˛ 2 D mgh
J
; 2ˇ D Jk ; P D Jp ; D D Jd .
One can rewrite (2) as a system of first-order differential equations:
P / D ˛ 2 '.t / 2ˇ '.t
!.t P / P 'i D !i ; t 2 Œi ; .i C 1// ; (3)
P / D !.t /;
'.t
with initial conditions !.i / D !i ; '.i / D 'i .

T
Using the notation yi D '.i / !.i / the solution of this equation formulates
a 2D map
yiC1 D .A C BK/ yi ; i 2 N; (4)
where

e ˇ cosh. / C ˇ sinh. / sinh. /

AD ; (5)
˛ 2 sinh. / cosh. / ˇ sinh. /
" #
e ˇ ˇ sinh. /C .cosh. /e ˇ /
BD ˛2 ; (6)
sinh. /

KD P D ; (7)
p
where D ˛ 2 C ˇ 2 .
Without considering rounding, one can acquire the dimensionless form of the
map, using the notations ˛O D ˛; ˇO D ˇ; pO D P 2 ; dO D D; x D '='ref ; v D
T
!=!ref ; yO D x v
O C BO K/
yO iC1 D .A O yO i ; (8)
where
O
" #
O D e
ˇ
O cosh.O / C ˇO sinh./ O O
sinh./
A ; (9)
O O
˛O 2 sinh./ O ˇO sinh./
O cosh./ O
O
" O
Ǒ
#
e ˇ ˇ sinh.O /CO cosh.O /e
BO D ˛2 ; (10)
O sinh./O
h i
KO D pO dO : (11)
q
Here O D ˛O 2 C ˇO 2 .
2.2 Rounding at the Output
Consider the case when rounding is applied to the calculated control effort (which
is the output of the control system). Introducing the rounding to Equation (8), the
so-called micro-chaos map (or -chaos map) is obtained:
O yO i C BO Int.K
yO iC1 D A O yO i /: (12)
Here the reference angle in yO is 'ref D rout 2 , and the reference angular velocity
is !ref D rout , while rout rad
s2
is the resolution of the actuated control effort and
Int.n/ denotes rounding towards zero (or truncating or taking the integer part of n).
2.3 Rounding at the Input
When rounding is applied to the measured position and velocity the corresponding
form of the micro-chaos map is
O yO i C BO K
yO iC1 D A O Int.Oyi /; (13)
where the reference angle is 'ref D rin (i.e. the resolution of the position
measurement), while the reference angular velocity is !ref D rin = since we assume
that the velocity is calculated from the measured position. Function Int./ calculates
the integer part of every element in the vector, returning a vector of the same
dimension.
3 Numerical Methods for Examining Chaotic Behaviour
In this section the general overview of numerical methods which are useful for the
quick characterization of control-related chaotic phenomena is presented. Methods
are divided into two groups according to the size of the region in which the
chaotic behaviour is examined. The first group shows methods which are useful
to characterize a single attractor, while the second group enumerates methods for
examining an arbitrarily large region in the state space.
3.1 Methods for Examining Local Chaotic Behaviour

3.1.1 Iteration of the Map
One can start repeatedly applying (12)—the micro-chaos map for the case when
rounding is at the output—to a chosen initial state and will eventually find that the
iteration will not converge to a specific stable equilibrium but after N iterations it
will arrive at one of the several coexisting chaotic attractors instead (Fig. 2). The
following figures are obtained for a set of realistic parameters (˛O D 6:85 103 ,
ˇO D 0, pO D 5:5 105 , dO D 2:5 103 ) [4]. Since the number of necessary
iterations N (i.e. duration of transients) is unknown, a criterion is needed to
determine whether the solution has reached an attractor. Having the exact Lyapunov
exponents in hand, an appropriate condition could be derived based on the time
necessary for the synchronization of coupled maps used in the Lyapunov exponent
estimation method [7].
v v
0.4 0.4
0.2 0.2
0.0 x 0.0 x
54 547 54 549 72 722 72 724
−0.2 −0.2
−0.4 −0.4
Fig. 2 Chaotic attractors when rounding is applied at the output

3.1.2 Searching for Periodic Points
One can search for periodic points of the map as shown in [2]. A point yO 0 is
p-periodic if yO p D yO 0 . Groups of periodic points (periodic orbits) with high period
usually correspond to the skeleton of a chaotic attractor. Consider the case when
rounding is applied to the output. Since
O yO 0 C BO Int.K
yO 1 A O yO /; (14)
„ ƒ‚ 0…
WDm1
O yO 1 C BO Int.K
yO 2 A O yO / D A.
O AO yO 0 C BO m1 / C BO m2 ; (15)
„ ƒ‚ 1…
WDm2
X
p
O p yO 0 C B
yO p A O O pi mi D yO 0 ;
A (16)
iD1
X
p
O p /1 BO
yO 0 D .I A O pi mi ;
A (17)
iD1
where mi are integers corresponding to constant control effort. For a p-periodic

orbit, checking all mi integer combinations requires very high computational
capacity. Moreover, prior to searching for periodic points, a global analysis should
be performed to find specific locations of interest.
3.2 Methods for Examining Global Chaotic Behaviour
Methods mentioned in the previous subsection generally give information about a

single attractor (corresponding to a specific set of initial states—i.e. the domain
of attraction). However, in the case when several chaotic attractors coexist, the
exploration and examination of a large region of the state space are necessary.
3.2.1 Systematic Iterations
Starting a set of iterations (repeatedly applying the micro-chaos map) from different
initial states—selected either randomly or systematically in a region within the state
space—one can explore the questioned region and find all attractors within it. Doing
so however requires knowledge about the duration of transients or criteria for testing
whether the solution has reached an attractor. Moreover executing a large number of
simulations requires a large computational capacity and no information on unstable
equilibria can be found.
3.2.2 Cell Mapping Methods
Cell mapping (CM) methods are tools for the global investigation of the long-term
behaviour of nonlinear dynamical systems [6]. Using CM methods, periodic and
chaotic solutions of the equations of motion can be found; moreover, the basin of
attraction can also be determined.
4 Selecting the Region of Interest
To obtain a global image of the state space, one needs to estimate the region of
chaotic behaviour. This can be done using the formula for the maximal possible
norm (y1 ) presented in [2] or the location of the attractors can be calculated based
on topological assumptions.
4.1 Topological Pattern
Consider the case when the rounding is applied to the output. A vector plot of (12)
in case of parameters ˛O D 6:53 103 , ˇO D 0, pO D 5:5 105 , dO D 2:5 103
can be seen in Fig. 3. The location of unstable fixed points can be derived from the
following equation:
100
50
–50
–100
1 1 2 2 3
x
0 p 2 p 2 p
a a
Fig. 3 Vector plot with switching lines (black lines), unstable fixed points (intersection of dashed
lines) and some example trajectories leading to different attractors (curves)

xu O xu O O xu
DA C B Int K ; (18)
0 0 0
O
e ˇ sinh.O / Int.pO xu / ˛O 2 xu
0D : (19)
O
O
Since e ˇ sinh./
O O 1 ¤ 0,
k
Int.pO xu / D ˛O 2 xu ! xu D ; k 2 Z: (20)
˛O 2
The equation of switching lines
l pO x
Int.pO x C dO v/ ! v D ; l 2 Znf0g: (21)
dO
Since the dynamics of the system between two switching lines is unstable, the stable
equilibria (i.e. attractors) are expected to be on the switching lines. The intersections
between the switching lines and the x-axis are
l
xs D ; l 2 Znf0g: (22)
pO
Based on topological assumptions, the unstable fixed points and attractors should
occur alternately. Since we restrict control parameters to the stable domain, pO > ˛O 2 ,
therefore the alternating pattern of unstable fixed points and attractors will end with
a virtual unstable fixed point (at lmax =˛O 2 ); thus
lmax lmax C 1 lmax

< < 2 : (23)
pO pO ˛O
Here the first term corresponds to the location of the last attractor. See Fig. 4. The
index of the last attractor is therefore
! !
1 1
lmax D Int pO C 1 D Floor pO (24)
˛O 2
1 ˛O 2
1
So the region of interest along the x-axis is x 2 Œlmax =p; O Clmax =p.
O The region
of interest along the y-axis was chosen to include the intersection of the stable
manifold of the unstable fixed point in the origin and the neighbouring switching
lines. One can also observe the location of unstable fixed points and stable attractors
by plotting the left- and right-hand side of Int.pO xu /=˛O 2 D x (see Fig. 4).
v
4
2
a
3
2
a
2
2
a
1
2
a
2 1 2 2 3 34 5 4
x
p 2 p 2 p 2 pa2
a a a p
Fig. 4 Solutions of Int.pO xu /=˛O 2 D x (in case of parameters ˛O D 6:53 103 , pO D 5:5
105 ) yielding the locations of unstable fixed points (dashed lines) and attractors (thin lines). Here
lmax D 4, the last attractor is at lmax =pO and the virtual unstable fixed point is at lmax =˛O 2
5 Cell Mapping Method and Results
5.1 Simple Cell Mapping
In simple cell mapping (SCM) [6], the Euclidean state space RN of a dynamical
system is restricted to a bounded region denoted by ˝, which is divided into M
(generally rectangular) cells, indexed by j 2 f1; ::; M g. The region RN n˝ is called
sink-cell (indexed with j D 0). The general idea behind SCM is that the state of
the examined system is no longer described by its state vector but with the index
of the cell corresponding to that state. For each cell one or more image cells can be
determined where the dynamics of the system leads to. In SCM only one image cell
is determined for each cell using the centre point of the cell (i.e. for flows the set of
ODEs describing the system are integrated for a fixed time period, or for maps the
map is applied to the state corresponding to the centre of the cell). The image cell
corresponding to cell z is denoted by C.z/. The mapping z.n C 1/ D C.z.n// C W
N ! N is called an SCM and generally means that the next state of the system is
completely described by its current state (and explicitly independent of the mapping
step n). In SCM, two kinds of cells are distinguished:
• Periodic cells: for which z D C m .z/ is true, for m 2 N. In this case z is an
m-periodic cell. If a cell z is m-periodic, cells C.z/; C 2 .z/; : : : ; C m1 .z/ are also
m-periodic cells, and such a group of periodic cells is called an m-periodic group.
By definition the sink cell is a 1-periodic cell (once the system enters into it, it
stays there forever).
• Transient cells: which are not periodic cells. Transient cells are mapped to a
periodic cell (or the sink-cell) in finite number of steps, thus representing the
basin of attraction of periodic groups.
The main procedure of the SCM method is determining the type and properties of
every cell. This is done by generating the sequence C.z/; C 2 .z/; : : : (while marking
cells as visited) for every cell z until the sequence returns to a previously visited cell.
If that cell is the part of the current sequence, a new periodic group and possibly
some transient cells leading to that group are found. If the cell is a previously
found transient cell, the current sequence contains only transient cells leading to
the same periodic group. Similarly in case the sequence leads to a previously found
periodic group, all cells in the sequence are marked as transient cells leading to that
group. (See the SCM algorithm in [6].) It is clear that in the context of SCM only
periodic motions occur, yet cell mapping methods are applicable for chaotic systems
by taking the following assumptions stated in [6]:
• Chaotic behaviour is represented by periodic groups with relatively high period.
• A chaotic attractor is represented by a set of periodic cells covering a part of the
attractor in the state space.
5.2 SCM Results
Some modifications were made to the SCM algorithm [6] to suit the micro-chaos
map. Adjacent periodic groups are considered to be the part of the same attractor;
therefore after the main procedure of SCM periodic groups close to each other were
joined. Moreover in order to increase the precision of the method around unstable
fixed points, the image cells were determined using multiple steps of the micro-
chaos map (i.e. if the image cell of a cell was itself, the map was applied once
again—until a fixed number of repetitions—to find the image cells of cells where
the dynamics of the system is slow). This way false 1-periodic cells were eliminated.
5.2.1 Rounding at the Output
When rounding is applied at the output (Section 2.2), the SCM method shows
the expected topological pattern. The chaotic attractors are indeed found on the
switching lines and all unstable fixed points are discovered as a set of 1-periodic
cells. The area of the basin of attraction of an attractor is larger, the larger the
distance of the closest neighbouring unstable fixed point is (i.e. the attractor
featuring the largest basin of attraction is the one equally far away from the two
neighbouring unstable fixed points). See Fig. 5. One can observe gateways on
switching lines, which are leading to a specific chaotic attractor (See also Fig. 4.)
Fig. 5 SCM results with switching lines (white), stable and unstable manifolds of unstable fixed
points (white, dashed) and chaotic attractors (black dots) for parameters ˛O D 6:8 103 , ˇO D 0,
pO D 5:5 105 , dO D 2:5 103 . Here lmax D 6, and the 3rd attractor has the largest basin of
attractions
Fig. 6 SCM results for ˛O D 6:8 103 , ˇO D 4:0 103 , pO D 5:5 105 , dO D 2:5 103
5.2.2 Effect of Damping
If the damping ratio is non-zero, the eigenvectors of (9) are rotating in clockwise
direction; therefore the previously mentioned gateways are moving and stretching
in the direction of the y-axis (Fig. 6).
5.2.3 Rounding at the Input
In case of realistic parameters rounding at the input leads to a solution where small
amplitude chaotic motion is superposed on a periodic orbit (Fig. 7).
Fig. 7 SCM results for ˛O D 5:0 103 , ˇO D 0, pO D 5:5 105 , dO D 2:5 103
Fig. 8 SCM images of the same chaotic attractor using different cell sizes (from left to right 940,
1100 and 1220 cells along each axis): ˛O D 6:85 103 , ˇO D 0, pO D 5:5 105 , dO D 2:5 103
5.2.4 Fractal Dimension Calculation with SCM
Since SCM utilizes rectangular cells, it could be suitable to determine the fractal
dimension (box counting or Minkowski dimension) of chaotic attractors. SCMs
with different cell sizes (") were applied to the close bounding region of chaotic
attractors, while the total number of periodic cells N."/ was stored. Then, the box-
counting dimension by definition is
log N."/
Dbox D lim : (25)
"!0 log 1="
The box-counting dimension was calculated by using least squares linear regression
to fit a line on .log 1="; log N."// points. The slope of the fitted line is Dbox .
However, SCM provides greatly different images even when the cell size varies a
little (Fig. 8). Because of this phenomenon, the calculated box-counting dimension
will be inaccurate.
6 Conclusion
It has been shown that simple cell mapping (SCM) is a suitable tool to examine
systems with complex chaotic behaviour, while it is inaccurate for box-counting
dimension calculation. For the case when rounding is applied to the output
(Sect. 2.2), the expected topological pattern was obtained by SCM and further
examination of gateways on switching lines (defined by the stable and unstable
manifolds of unstable fixed points) could help reveal unknown properties of the
2D micro-chaos map. On the other hand, rounding at the input (Sect. 2.3) shows a
completely different behaviour where chaotic motion is superposed on a periodic
orbit. In both cases, further investigation with more advanced CM methods is
planned.
Acknowledgements This research was supported by the Hungarian National Science Foundation
under grant no. OTKA K 83890.
References
1. Csernák, G., Stépán, G.: Digital control as source of chaotic behavior. Int. J. Bifur. Chaos 20(5),
1365–1378 (2010)
2. Csernák, G., Stépán, G.: Sampling and round-off, as sources of chaos in PD-controlled systems.
Proceedings of the 19th Mediterranean Conference on Control and Automation (2011)
3. Delchamps, F.D.: Stabilizing a linear system with quantized state feedback. IEEE Trans.
Automat. Contr. 35, 916–924 (1990)
4. Enikov, E., Stépán, G.: Micro-chaotic motion of digitally controlled machines. J. Vib. Contr. 4,
427–443 (1998)
5. Haller, G., Stépán, G.: Micro-chaos in digital control. J. Nonlinear Sci. 6, 415–448 (1996)
6. van der Spek, J.A.W.: Cell Mapping Methods: Modifications and Extensions. Eindhoven
University of Technology, Eindhoven (1994)
7. Stefański, A., Kapitaniak, T.: Estimation of the dominant Lyapunov exponent of non-smooth
systems on the basis of maps synchronization. Chaos Sol. Fract. 15, 233–244 (2003)
Bouncing Ball Dynamics: Simple Motion of the
Table Approximating the Sinusoidal Motion
Andrzej Okniński and Bogusław Radziszewski
Abstract We study nonlinear dynamics of a small ball moving vertically in a

gravitational field and colliding with a moving table. Displacement of the table is
approximated in one period of its motion by four cubic polynomials (we shall refer
to this model as MC ). The Poincaré map, mapping a state of the ball at an impact
to a new state at the next impact, is used to investigate the ball dynamics. Results
obtained for the model MC are used to elucidate dynamics of the standard model
of bouncing ball with sinusoidal motion of the limiter (model MS ). Bifurcation
diagrams are computed for the models MC and MS . Then fixed points and their
stability are determined for the MC model and compared with our earlier results
for the model MS . Finally, the birth of the low-velocity 2-cycle is investigated
analytically for the MC model.
1 Introduction
In the present paper we investigate dynamics of a small ball moving vertically in

a gravitational field and impacting with a periodically moving limiter (a table).
This simple but important model has been extensively studied within the framework
of nonsmooth and nonlinear dynamical systems which have many applications in
technology [1, 2, 5, 11].
In dynamics with impacts it is usually difficult or even impossible to solve
nonlinear equation for an instant of the next impact. For example, in the bouncing
ball models, the table’s motion has been usually assumed in sinusoidal form. This
choice of the limiter’s motion leads to a complicated nonlinear equation for time
of the next impact. To address this problem we proposed a sequence of models in
which periodic motion of the table is assumed (in one period of limiter’s motion) as
a low-order polynomial of time; see [6] and references therein.
A. Okniński ()
Kielce University of Technology, 25-314 Kielce, Poland
e-mail: fizao@tu.kielce.pl
B. Radziszewski
Collegium Mazovia Innovative University, 08-110 Siedlce, Poland
e-mail: b@nazu.org

398 A. Okniński and B. Radziszewski
In the present work we conduct analytical and numerical investigations of the

model in which sinusoidal displacement of the table is approximated in one period
by four cubic polynomials. We shall refer to this model as MC . We hope that
rigorous results obtained for the model MC may cast light on dynamics with
sinusoidal motion of the limiter. The latter model is referred to as MS .
The paper is organized as follows. In Sect. 2 a one-dimensional dynamics of a
ball moving in a gravitational field and colliding with a table is reviewed and the
corresponding Poincaré map is constructed and models of the limiter’s motion MC
and MS are defined. Bifurcation diagrams are computed for MC and MS . In Sect. 3
fixed points and their stability are investigated for the MC model and compared with
our earlier results for the MS model. In Sect. 4 birth of the low-velocity 2-cycle is
investigated analytically for the MC model. We summarize our results in the last
section.
2 Bouncing Ball: A Simple Motion of the Table
We assume that a ball moves vertically in a constant gravitational field and collides
with a periodically moving table. The ball is treated as a material point and it
is assumed that the limiter’s mass is so large that its motion is not affected at impacts.
Dynamics of the ball is studied via a Poincaré map which takes a state of the ball
at an impact to a new state at the next impact. The corresponding Poincaré map
in nondimensional form reads (cf. [7], see also Ref. [4] where analogous map was
derived earlier):
Y .TiC1 / D Y .Ti / 2iC1 C iC1 Vi ; (1a)

ViC1 D RVi C 2RiC1 C .1 C R/ YP .TiC1 / ; (1b)
where Ti denotes time of the i -th impact and Vi is the corresponding postimpact
velocity while iC1 TiC1 Ti . The parameters , R are a nondimensional
acceleration and the coefficient of restitution, 0 R < 1 [11], respectively, and
the function Y .T / represents the limiter’s motion. The limiter’s motion has been
usually assumed in sinusoidal form, YS .T / D sin.2T /. Equations (1) and Y D YS
lead to the model MS . This model of limiter’s motion leads to serious difficulties in
solving the first of Eqns.(1) for TiC1 , making analytical investigations of dynamics
virtually impossible. We have thus decided to simplify the limiter’s periodic motion
to make (1a) solvable. The function YC .T / defined below:
8
ˆ
ˆ f1 .T / ; 0 TO < 14
<
f2 .T / ; 14 TO < 12
YC .T / D (2)
ˆ f3 .T / ; 12 TO < 34
:̂
f4 .T / ; 34 TO 1
Bouncing Ball Dynamics 399
f1 .T / D .32 128/ TO 3 C .16 C 48/ TO 2 C 2 TO (3a)

O3
f2 .T / D .128 32/ T C .144 C 32/ T C .48 10/ TO 4 C
O2 (3b)
f3 .T / D .128 32/ TO 3 C .240 C 64/ TO 2 C .144 42/ TO 28 C 9 (3c)
O3
f4 .T / D .32 128/ T C .336 80/ T C .288 C 66/ TO C 80 18 (3d)
O2
provides a reasonably good approximation of the function YS D sin.2T / on all

intervals Œk; k C 1, k D 0; 1; : : : ; with TO D T bT c, where bxc is the floor
function – the largest integer less than or equal to x.
The model MC consists of equations (1), (2), (3) with control parameters R, .
Velocities of the limiter are computed as gi .T / D dtd
fi .T /, i D 1; : : : ; 4.
In Fig. 1 above the bifurcation diagram with velocities and impact times versus
computed for growing and R D 0:85 has been shown. It follows that dynamical
system MC has several typical attractors which undergo a period doubling scenario:
fixed points, with one impact per k periods and Vi D V D k, and 3-cycles with
Vi k. There are also several small attractors. We shall investigate some of these
attractors in the next section combining analytical and numerical approach.
In Fig. 2 bifurcation diagram for the sinusoidal motion has been shown. Simi-
larity of Figs. 1 and 2 suggests that analytical results obtained for model MC may
shed light on the problem of sinusoidal motion, MS .
3 Fixed Points and Their Stability
We shall first study

periodic
solutions of the model MC with one impact per k
periods and T 2 0; 14 since it is suggested by the bifurcation diagram that they
are stable. Such states have to fulfill the following conditions:
.k=1/ .k=1/
VnC1 D Vn V ; TnC1 D Tn C k T Ck .k D 1; 2; : : :/ ; (4)
where

> YPc1 T
.k=1/ .k=1/ .k=1/
T 2 0; 1
4
; V : (5)
The demanded (stable) solution is given by

q
.k=1/
T.s/ D 3
6.4/
1
24.4/
4 . 6/2 C 6B . 4/; BD k 1R
1CR
(6a)
.k=1/
V D k: (6b)
Fig. 1 Bifurcation diagram for the model MC , R D 0:85: velocities after impacts (top figure) and
times after impacts (bottom figure) versus control parameter
Fig. 2 Bifurcation diagram for the model MS , R D 0:85: velocities after impacts (top figure) and
times after impacts (bottom figure) versus control parameter
Since T 2 Œ0; 1 we demand that T > 0 and it follows from (6a) that physical
.k=1/
solution appears for lower critical value > cr1 where
.k=1/
cr1 ;C D k 1R
2 1CR
: (7)
We have checked by stability analysis that the solution (6a), (6b) is stable for
.k=1/
> cr1 ;C , i.e. when it is physically acceptable. To determine upper critical value
of when dynamics loses stability we put into (1):
.k=1/ .k=1/
Ti D T.s/ C "i ; TiC1 D T.s/ C k C "iC1 ; (8)
Vi D V C i D k C i ; ViC1 D V C iC1 D k C iC1 ; (9)
with Y .T / given by (3), and keep only terms linear in perturbations "i , "iC1 , i ,
iC1 of the fixed point to get
0 1
1 k
"iC1 f1 .T /Ck "i
D@ A (10)
iC1 .1 C R/ g1 .T / k 2RC.1CR/g 1 .T /
R i
f1 .T /Ck
.k=1/
where T T.s/ , f1 .T / is given by (3a), and g1 .T / D d
dT
f1 .T /.
Since the characteristic polynomial is
8 2
<X Cq
ˆ ˛X C ˇ D 0
˛ D 4 4 . 6/2 C 6k . 4/ .1CR/
1R
.1 C R/2 R2 1 (11)
:̂
ˇ D R2
application of the Schur-Cohn criterion [3]:
ˇ < 1; j˛j < ˇ C 1 (12)
.k=1/ .k=1/
leads finally to the following localization of the fixed points (6): cr1 ;C < < cr2 ;C ,
.k=1/
where cr1 ;C is given by (7) and
q
2 2
.k=1/ 6k.4/.R2 1/C 36k 2 .4/2 .1R2 / C4.6/2 .1CR2 /
cr2 ;C D ; R < 1: (13)
8.6/2 .1CR/2
In Fig. 3 stability regions in .R; / plane for the MC model are shown.
In the case of the model MS we have
.k=1/ .k=1/
cr1 ;S D cr1 ;C ; (14)
p
.k=1/ k 2 2 .1R2 /2 C4.1CR2 /2
cr2 ;S D 2 2 .1CR/2
; R < 1; (15)
see [7] (note that in [7] we used Y.T / D sin.T / rather than YS .T / D sin.2T / and
it follows that all values of the control parameter must be rescaled, D .2/
2 ),
and stability regions are very similar to those of model MC , cf. Fig. 3.
0.5
γ
0.45
0.4
0.35
0.3 k=3
0.25
0.2 k=2
0.15
0.1
k=1
0.05
0
R
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
γ
0.45
0.4
0.35
0.3 k=3
0.25
0.2 k=2
0.15
0.1 k=1
0.05
0 R
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Fig. 3 Stability regions for the model MC (top figure) and the model MS (bottom figure)
4 Birth of Low-Velocity 2-Cycle
In this section we shall study birth of low-velocity k-cycles which can be seen in
the bifurcation diagram, Figs. 1 for > 0:03 and V < 1. In the case of such cycles
T1 ; T2 ; : : : ; Tk 2 .0; 1/ and TkC1 1 D T1 . Numerical tests show that 2-cycle
fulfilling conditions T1 2 0; 14 ; T2 2 12 ; 34 and T3 D T1 C 1 are stable. This
2-cycle can be seen in the bifurcation diagram in Fig. 1 for & 0:0366 and V1 Š
0:51, V2 Š 0:55 (R D 0:85). Equations to determine T1 ; T2 , and V1 ; V2 are shown
below:
8
ˆ
ˆ f3 .T2 / D f1 .T1 / .T2 T1 /2 C .T2 T1 / V1
ˆ
ˆ
ˆ
ˆ V2 D RV1 C 2R .T2 T1 / C .1 C R/ g3 .T2 /
ˆ
<
f1 .T3 1/ D f3 .T2 / .T3 T2 /2 C .T3 T2 / V2
(16)
ˆ
ˆ V3 D RV2 C 2R .T3 T2 / C .1 C R/ g1 .T3 1/
ˆ
ˆ
ˆ
ˆ T3 D T1 C 1
:̂
V3 D V1
where fi .T /’s and gi .T /’s are defined in Eqn. (3) and the text below.
We were able to simplify Eqns. (16) significantly obtaining equation for
T2 T1 only:
X9
F ./ D dj j D 0; (17)
j D0
where di ’s are given in the Appendix in [9]. Numerical computations suggest that
.2/ .2/
the 2-cycle appears for D cr;C and fixed R, where cr;C is a critical value,
.2/
as a double (and stable) solution of Eqns. (16). For > cr;C there are two real
solutions, one stable (seen in the bifurcation diagram) and another unstable. On the
.2/
other hand, for < cr;C , the solutions are complex conjugated and thus unphysical.
.2/
Moreover, at D cr;C , the stability matrix has unit eigenvalue. Therefore this is a
tangent (saddle-node) bifurcation; see [10] for elementary discussion of the tangent
bifurcation in the logistic map when the 3-cycle is born.
To determine critical value of the parameter let us note that double solution
of the polynomial equation (17) is also the solution of G ./ D 0 where G ./ D
d
d
F ./. For example, solving for R D 0:85 the system of equations:
X9
F ./ D dj j D 0; (18)
j D0
Xj
G ./ D jdj j 1 D 0; (19)
j D1
.2/
we get cr;C D 0:036 617 052 682 892 250 62, cr D 0:634 279 960 677 747 355 95
(and many other, unphysical solutions) in perfect agreement with numerical compu-
tations, see also Fig. 1.
5 Summary
We have studied dynamics of a bouncing ball impacting with a periodically moving

limiter within two frameworks of the table motion: MC and MS defined in Sect. 2.
Stability conditions of fixed points have been determined and results for the models
MC and MS have been compared. Then we have found for the model MC that
the low-velocity 2-cycle is born in tangent bifurcation. Equations for this 2-cycle
have been written and the condition for the onset of this attractor was determined
and solved numerically. The second part of this study will be devoted to systematic
investigation of birth of attractors and chattering in the models MC and MS [8].
References
1. Awrejcewicz, J., Lamarque, C.-H.: Bifurcation and Chaos in Nonsmooth Mechanical Systems.
World Scientific Publishing, Singapore (2003)
2. di Bernardo, M., Budd, C.J., Champneys, A.R., Kowalczyk P.: Piecewise-Smooth Dynamical
Systems. Theory and Applications. Springer, Berlin (2008)
3. Jury, E.I.: Inners and Stability of Dynamic Systems. Krieger, New York (1982)
4. Luo, A.C.J., Han, R.P.S.: The dynamics of a bouncing ball with a sinusoidally vibrating table
revisited. Nonlinear Dynam. 10, 1–18 (1996)
5. Mehta, A. (ed.): Granular Matter: An Interdisciplinary Approach. Springer, Berlin (1994)
6. Okniński, A., Radziszewski, B.: Dynamics of impacts with a table moving with piecewise
constant velocity. Differ. Equ. Dyn. Syst. 21(1&2) 165–171 (2013)
7. Okniński, A., Radziszewski, B.: Grazing dynamics and dependence on initial conditions in
certain systems with impacts, arXiv: 0706.0257v2, 17 pages (2007)
8. Okniński, A., Radziszewski, B.: Int. J. Nonlinear Mech. 65, 226–235 (2014)
9. Okniński, A., Radziszewski, B.: Bouncing ball dynamics: simple model of motion of the table
and sinusoidal motion, arXiv: 1302.0369v1, 2013, 23 pages.
10. Peitgen, H.O., Jürgens, H., Saupe D.: Fractals for the Classroom. Part Two: Complex Systems
and Mandelbrot Set. Springer, New York (1992)
11. Stronge, W.J.: Impact Mechanics. Cambridge University Press, Cambridge (2000)
Periodic Motions of Coupled Oscillators Excited
by Dry Friction and Harmonic Force
Madeleine Pascal and Sergey Stepanov
Abstract Vibrating systems excited by dry friction are frequently encountered in

technical applications. These systems are strongly nonlinear, and they are usually
modeled as spring-mass oscillators. One of the most popular models of stick-slip
oscillators consists of several masses connected by linear springs; one (or more)
of the masses is in contact with a driving belt moving at a constant velocity. In
the past, several authors investigated the behavior of this system, with different
friction laws and with or without external actions and damping. In this work, we
consider a system composed of two masses connected by linear springs. One of the
mass is in contact with a driving belt moving at a constant velocity. Friction force,
with Coulomb’s characteristics, acts between the mass and the belt. Moreover, it
is assumed that the mass is also subjected to a harmonic external force. Several
periodic orbits including stick phases and slip phases are obtained in closed form. In
particular, the existence of periodic orbits including an overshooting part is proved.
In the case of a nonmoving belt, a set of nonsticking periodic solutions is obtained,
and we prove that these orbits are symmetrical in space and in time.
1 Introduction
This paper is a continuation of several investigations [5–7, 10] related to vibrating

systems excited by dry friction. One of the most popular models of stick-slip
oscillators consists of several masses connected by linear springs; one (or more)
of the masses is in contact with a driving belt moving at a constant velocity. In
the past, several authors investigated the behavior of this system, with different
friction laws and with or without external actions and damping [1, 2, 4, 11], mainly
via the numerical approach. However, assuming Coulomb’s laws of dry friction,
M. Pascal ()
Universite d’Evry, IBISC, 40 rue du Pelvoux, 91020, Evry, France
e-mail: madeleine.pascal3@wanadoo.fr
S. Stepanov
Department of Mechanics, Computing Center of the RAS,
40 Vavilova street, 119991, Moscow, Russia
e-mail: stepsj@ya.ru

408 M. Pascal and S. Stepanov
the corresponding dynamical model is a piecewise linear system, and even for
multi-degree-of-freedom cases, some analytical results about the existence and the
stability of periodic orbits including stick-slip phases have been obtained [5–7]. One
interesting phenomenon is the existence, inside periodic orbits with stick and slip
parts, of an “overshooting” slip phase. During this part of the orbit, the mass in
contact with the belt moves in the same direction at a higher speed than the belt
itself. In this work, we consider the same model of dry friction oscillator subjected
to a harmonic external force. Several periodic orbits including stick phases and slip
phases are obtained in closed form. In particular, the existence of periodic orbits
including an overshooting part is proved. In the case of a nonmoving belt, a set
of nonsticking periodic solutions is obtained, and we prove that these orbits are
symmetrical in space and in time.
2 Problem Formulation
The system (Fig.1) is composed of two masses m1 ; m2 connected by two linear

springs of stiffness k1 ; k2 . The second mass is in contact with a belt moving at
a constant velocity 0 . A friction force FQ acts between the mass and the belt.
Moreover, the second mass is also subjected to a driving force RQ D pQ cos .!t C '/
Q !; ' are constant parameters). The motion of this system are given by equations
(p;
x100 C x1 %x2 D 0; x200 C % .x2 x1 / D u C p cos .!t C '/ ; (1)
m1 k2 FQ pQ
D ; %D ; uD ; pD ;
m2 k 1 C k2 k 1 C k2 k 1 C k2
s
k 1 C k2 d .O/
0
t D˝t; ˝D ; .O/0 D ;
m1 dt
x1 ; x2 are the displacements of the masses.
Fig. 1 Dry friction oscillator

Periodic Motions of Coupled Oscillators Excited by Dry Friction and Harmonic Force 409
The dry friction force u is obtained from Coulomb’s laws:

u D us sign V x20 ; if x20
V ¤ 0;
u D % .x2 x1 / r.t /; if x20
V D 0; j% .x2 x1 / r.t /j < ur ;
u D us ; if x20
V D 0; % .x2 x1 / r.t / > ur ;
u D us ; if x20
V D 0; % .x2 x1 / r.t / < ur ;
0 p
0 < us < ur ; V D ; r.t / D cos .!t C '/ ;
˝
ur is the static friction force, and us is the dynamic friction force.
3 Three Modes of Motion Exhibited by the System
The dynamical behavior of this oscillator includes several phases of slip and stick
motion of m2 . For each kind of motion, the closed form solution is available.
3.1 Slip Motion of m2 with x20 < V
The solution is obtained from a modal analysis of (1) where u D us

R .t /
Z .t / D H .t / .Z0 F0 / C F .t / ; F .t / D ;
R0 .t /
R.t / D Q cos.!t C '/; F0 D F .0/;

z x1
Z D 0 ; Z0 D Z.0/; Z D X d0 ; X D ;
z x2

H1 .t / H2 .t / d01 us d01
H .t / D ; d0 D ; d01 D ; d02 D ;
H3 .t / H1 .t / d02 1% %

q1 p% p.1 ! 2 /
QD ; q1 D 2 ; q2 D q1 :
q2 .! !1 /.! !2 /
2 2 2 %
The two-by-two matrices Hi .t / .i D 1; 2; 3/ and the natural frequencies !1 ; !2 are

obtained in analytical form [5].
3.2 Slip Motion of m2 with x20 > V (Overshooting)
The solution is obtained from (1) where u D us

! !
H1 .t / I 10
Z .t / D H .t / .Z0 F0 / C F .t / C 2L .t / d0 L .t / D ; I D :
H3 .t / 01
3.3 Stick Motion of m2 (x20 D V )
This motion is related to the dynamical system
x100 C x1 %x2 D 0; x200 D 0: (2)
The solution [5] is given by

1 .t / 2 .t /
Z .t / D .t / Z0 ; .t / D :
3 .t / 1 .t /
i .t / .i D 1; 2; 3/ are obtained from a modal analysis of the system (2). Moreover,

during all this kind of motion, the following constraint holds:
j% .x2 x1 / p cos .!t C '/j < ur :
4 Periodic Orbits Including an Overshooting Part
Several periodic orbits including stick phases and slip phases have been obtained [7].
Among these orbits, a set of periodic orbits including an overshooting part is found.
For each period .0 < t < 2=!/, the motion is composed of three parts. The first
one .0 < t < / is a slip motion of m2 with x20 < V , the next part (0 < t < 1 ) is
an overshooting slip motion of the mass (x20 > V ), and the last part (0 < t 1 <
T; T D 2=! 1 ) is a stick motion of m2 . At the beginning of the motion
for t D 0, we assume that
x20 .0/ D V; % .x2 .0/ x1 .0// D p cos ' C ur
and at t D , we assume the conditions
x20 ./ D V; % .x2 . / x1 . // p cos .! C '/ C ur < 0: (3)
The last condition leads to an overshooting motion for t > .

Fig. 2 Phase portrait of the overshooting periodic solution
An example of a periodic orbit with an overshooting part is obtained for the set of
data
% D 0:2; D 3:8; us D 0:059; ur D 0:5347; V D 1; ! D 0:6 ; p D 0:1; ' D 0:
The other parameters are computed:
D 4:775; 1 D 2:65; T D 3:047; z1 .0/ D 1:1471; z2 .0/ D 3:6572; z01 .0/ D 0:7281:
The phase portraits of the system are shown in Fig. 2. The solid lines show slip
motion (thick black lines correspond to the overshooting motion); the dotted lines
are related to the stick motion.
5 Nonsticking Periodic Solutions
In industrial applications, avoiding sticking phases of motion is sometimes nec-

essary. In the past, several authors [3, 8] investigated the existence of periodic
nonsticking solutions of a one-degree-of-freedom oscillator subjected to simple
harmonic loading. The mass is in contact with a fixed surface and a dry friction
force acts between the mass and the surface. The aim of these works is to obtain
some estimates of the minimum external force amplitude needed to prevent this
sticking motion. The nonsticking orbit involves for each period a slip motion with a
negative mass velocity and a slip motion with a positive mass velocity (overshooting
motion). Moreover, the authors assumed that the motion is symmetrical in space and
time [3, 8].
In the following, this problem is revisited for the two-degree-of-freedom oscil-

lator considered in this work. Let us consider the system described in Fig.1 with
initial conditions
0
x20 D V; % .x20 x10 / > p cos ' C ur : (4)
The nonsticking periodic orbit is composed of two parts: for 0 < t < the system
undergoes a slip motion (x20 < V ); the second part for < t < 2=! is an
overshooting motion (x20 > V ). At t D the conditions (3) are fulfilled. We prove
[7] that this kind of periodic orbits exists only if
V D 0; D =!; x .0/ D x ./ ; x 0 .0/ D x 0 ./ :
The nonsticking periodic orbits are symmetrical in space and time.

A numerical computation is performed for the parameters
% D 0:3; D 4; ! D :6; p D 1; us D 0:1; ur D 0:2996 :
The corresponding values of the initial conditions and of the time lag ' are obtained:
0
x10 D 1:5608; x20 D 3:3295; x10 D 0:1523; ' D 0:3925 :

The phase portraits xi ; xi0 ; i D .1; 2/ of the two masses are shown in Fig. 3 (the
thick black parts of the curves are related to the overshooting motion). These curves
are symmetrical with respect to the origin.
Fig. 3 Phase portrait of the nonsticking orbit

6 Periodic Orbits with Abnormal Stops
The nonsticking orbit investigated in the last paragraph involves for each period two
normal stops
x20 .0/ D x20 . / D 0:
These stops occur when the displacement of the second mass reaches a local
extremum and the mass reverses its direction of motion at the turning point. In [9],
for a one-degree-of-freedom oscillator, with dry friction and harmonic load, a set of
periodic orbits including abnormal stops have been obtained. Abnormal stops occur
when at the turning point the mass moves in the same direction as its motion prior
to the stop. The same phenomenon can be observed for the two-degree-of-freedom
oscillator investigated in this paper. At t D 0 the initial conditions (4) are assumed
and for 0 < t < , the system undergoes a slip motion (x20 < 0). Let us assume that
at t D
x20 . / D 0; % .x2 . / x1 .// p cos .! C '/ ur > 0:
A new phase of slip motion occurs. This motion ends at t D C 1 if at this time
x20 . C 1 / D 0; % .x2 . C 1 / x1 . C 1 // p cos .! . C 1 / C '/ C ur < 0:
For t > C1 the system undergoes an overshooting motion (x20 > 0). A periodic
orbit of period 2=! is obtained, but due to the symmetry in space and in time of
this nonsticking orbit, the overshooting motion ( C 1 < t < 2=!) involves
another abnormal stop for t D C 1 C T , with the conditions
x20 . C 1 C T / D 0;
% .x2 . C 1 C T / x1 . C 1 C T // p cos .! . C 1 C T / C '/ C ur < 0:
Due to the symmetrical property we obtain
C 1 D =!; T D ; x . C 1 / D x .0/ ; x 0 . C 1 / D x 0 .0/ :
This kind of orbit exists only if us D ur .

An example of periodic orbits with abnormal stops is obtained for the data
% D 0:7; D 3:8; ! D 2=11; p D 1; ur D us D 0:1717; ' D 2:6455 :
The other parameters are computed:
D T D 4; 1 D 1:5; x1 .0/ D 2:1249; x2 .0/ D 2:0451; x10 .0/ D 0:7267 :

Fig. 4 Phase portrait of the orbit with abnormal stops
The phase portrait of the system is shown in Fig. 4 (the thick black parts of the
curves are related to the overshooting motion).
References
1. Andreaus, U., Casini, P.: Dynamics of friction oscillators excited by a moving base and/or
driving force. J. Sound Vib. 245, 685–699 (2001)
2. Awrejcewicz, J., Olejnik, P.: Stick-slip dynamics of a two-degree-of-freedom system. Int.
J. Bifur. Chaos. 13(4), 843–861 (2003)
3. Csernak, G., Stepan, G.: On the periodic response of a harmonically excited dry friction
oscillator. J. Sound Vib. 295, 649–658 (2006)
4. Olejnik, P., Awrejcewicz, J.: Application of Henon method in numerical estimation of the stick-
slip transitions existing in Filippov-type discontinuous dynamical systems with dry friction.
Nonlinear Dyn. 73(1–2), 723–736 (2013)
5. Pascal, M.: Dynamics of coupled oscillators excited by dry friction. ASME J. Comput.
Nonlinear Dyn. 3, 20–26 (2008)
6. Pascal M.: New events in stick slip oscillators behavior. J. Appl. Math. Mech. (PMM). 75,
10–17 (2011)
7. Pascal, M.: New limit cycles of dry friction oscillators under harmonic load. Nonlinear Dyn.
70, 1435–1443 (2012)
8. Hong, H.K., Liu, C.S.: Non-sticking oscillation formulae for Coulomb friction under harmonic
loading. J. Sound Vibr. 244(5), 883–898 (2001)
9. Hong, H.K., Liu, C.S.: Coulomb friction oscillator: Modeling and responses to harmonic loads
and base excitations. J. Sound Vib. 229(5), 1171–1192 (2000)
10. Stepanov, S.: Limit cycles of a double oscillator excited by dry friction. Int. J. Bifur. Chaos.
21(10), 3043–3046 (2011)
11. Van de Vrande, B.L., Van Campen, D.H., De Kraker, A.: An approximate analysis of dry-
friction-induced stick-slip vibrations by a smoothing procedure. Nonlinear Dyn. 19, 157–169
(1999)
Limit Cycle Oscillations of an Aerodynamic
Pendulum
Yury D. Selyutskiy
Abstract Dynamics of an aerodynamic pendulum in low-speed airflow is studied

using the phenomenological model, where the internal dynamics of the flow is
simulated using an oscillator attached to the pendulum. Limit cycle oscillations
occurring in the vicinity of the “along the flow” equilibrium at a certain range of
wind speeds are studied. Their evolution under changing system parameters (wind
speed, holder length, etc.) is analyzed. Experimental study of such oscillations of
the aerodynamic pendulum is performed in the subsonic wind tunnel of the Institute
of Mechanics of Lomonosov Moscow State University for different wind speeds
and different structural parameters of the pendulum. It is shown that results of
experiments are in good enough agreement with simulation results.
1 Introduction
Aerodynamic pendulum, that is, a wing mounted a flow so that it can rotate around
a fixed axis (like a weathercock), belongs to the waste family of aeroelastic systems.
Study of different systems of this class began as early as in 30th of the last century
in different countries and was driven by problems of quickly developing aviation,
and since then, an extensive literature appeared dedicated to research of various
issues related with such objects. It is well known that the “trivial” equilibrium of
such objects (namely, situation when the wing is stretched along the flow) becomes
unstable in a certain range of parameters, and a limit cycle appears. Prediction
and control of such cycles is important, as for some technical objects, they are
undesired or even dangerous, while for others they can represent working regimes
(for instance, for systems for converting flow energy into electric power).
Y.D. Selyutskiy ()

Institute of Mechanics of Lomonosov Moscow State University, 119192,
Michurinsky pr. 1, Moscow, Russia
e-mail: seliutski@imec.msu.ru

416 Y.D. Selyutskiy
This effect is described for systems with two degrees of freedom (e.g., [6,14,15])
allowing both pitching and plunging motion of the wing. The work [10] shows
existence of this phenomenon also for systems with one degree of freedom (only
rotational). Papers [8, 9] show experimentally and numerically that such cycles
with small amplitude appear in a certain range of wind speeds for the aerodynamic
pendulum with one degree of freedom.
It should be noted that these results pertain to the case of elastically mounted
pendulum with elastic axes located within the chord of the wing, and hence, the
characteristics of limit cycles are influenced by combined action of aerodynamic
load and elastic forces (as a rule, assumed nonlinear). However, from the point
of view of both theory and applications, it is interesting to examine the pure effect
of aerodynamic forces and to study the evolution of limit cycles when the position
of elastic axis changes in wide enough range. Works [5, 11] considered behavior of
one- and two-degree-of-freedom aerodynamic pendulum without spring mounting
under the assumption of quasi-steady aerodynamics, which did not allow describing
limit cycles in the domain of small angles of attack.
Another important issue concerning aeroelastic (and not only) systems is devel-
oping a plausible and simple enough (that is, containing small number of additional
degrees of freedom and relatively few parameters) phenomenological mathematical
model of unsteady interaction between the wing and the flow. The actuality of
this task is related to the fact that engineering practice requires in many cases
to determine values of parameters that are in a certain sense optimal for the
given problem. In order to fulfill this requirement, it is necessary either to set up
multiple experiments or perform numerous numerical simulation runs. However,
making real experiments for studying the dynamical behavior of bodies in flow is
very cumbersome, which makes mass testing impracticable. Performing numerical
simulation using different CFD-type packages (or vortical methods like those
described in [2]) in such kind of problems requires considerable computational
resources. This makes it troublesome (if ever possible) to perform quick and efficient
parametric analysis in a wide enough range of parameter values.
Therefore, different types of such phenomenological approaches are developed
by various groups of researchers (e.g., [3, 4, 7]). All these methods are based
on introducing additional empirical variables, so that original dynamic system is
supplemented with additional ODEs for aerodynamic forces and/or moments.
The present work develops another approach proposed in [12] where the internal
dynamics of the flow is simulated with an additional (“dummy”) degree of freedom
(“attached oscillator”). Stability of the “along the flow” equilibrium of this object is
studied, and stability domains are determined for different combinations of system
parameters. Evolution of limit cycle is studied in wide range of wind speeds and
positions of rotation axis. Results of mathematical simulation are verified against
experiments made in the wind tunnel of the Institute of Mechanics of Lomonosov
Moscow State University.
Limit Cycle Oscillations of an Aerodynamic Pendulum 417
Fig. 1 Aerodynamic pendulum with the attached oscillator
2 Problem Statement and Motion Equations
Consider an aerodynamic pendulum, that is, a rigid body consisting of a weightless

holder OA and a wing with symmetric airfoil (Fig. 1) that can rotate about a fixed
axis O. The system is placed in airflow having constant speed V at infinity. Denote
the angle of rotation of the pendulum about the axis with
.
We neglect the aerodynamic load upon the holder and suppose that the flow is
two-dimensional. In order to take into account unsteady effects arising due to the
nonuniform motion of the object under consideration, it is necessary to describe
the internal dynamics of the flow. For that, we use the approach proposed in [12],
where the internal dynamics of the flow is simulated using an oscillator G attached
to the wing. In [13] it was shown that this method allows to provide good enough
agreement with results of numerical calculations according to the modified method
of discrete vortices [1].
Within the framework of this approach it is assumed that the oscillator is attached
to the chord in the center of pressure C of the airfoil and that point G can move along
the normal to the chord. Let be the oscillator length C G, and l be the distance OC .
Assume also that the aerodynamic force is applied to the point G, and its normal and
tangential components have the following structure:
S 2 S 2
Fn D V Cn .˛G /; F D V C .˛G / (1)
2 G 2 G
Here is the air density, S is the wing area, VG is the effective flow speed, that
is, the speed of point G with respect to the flow at infinity, ˛G is effective angle of
attack, and Cn , C are nondimensional coefficients of normal and tangential force,
correspondingly. Their dependence on the effective angle of attack is the same as in
steady case and can be determined from static experiments.
Evidently, the effective angle of attack should contain information about orien-
tation and the angular speed of the airfoil, as well as about the attached oscillator
speed. Taking this into account, we use the following formula for ˛G :
.l C L/
P P
˛G D
(2)
V V
Here we suppose that L > 0.
418 Y.D. Selyutskiy
From steady experiments it is known that for thin enough airfoils and small
angles of attack the tangential force coefficient is much smaller than the normal
force coefficient. So, we neglect this component of the aerodynamic force. Then the
motion equations of the aerodynamic pendulum with the attached oscillator look as
follows:
J
R C 2m
P P C m2
R C lm
P 2 D D
P lFe (3)
mR C ml
R m
D Fn C Fe
P2 (4)
Here J is the moment of inertia of the pendulum with respect to the point O, D
is the structural damping coefficient, Fe is elastic force, and m is the mass of the
oscillator.
In order to describe nonlinear effects arising during the unsteady interaction
between the pendulum and the flow, we introduce nonlinearity into the elastic
force Fe :
Fe D k hP H
P 2 (5)
Equations (3)–(4) along with relations (1), (2), and (5) make a closed system of
equations.
As for symmetric airfoil we have Cn .0/ D 0, it is evident that the “along the
flow” position (
D 0, D 0) is equilibrium. We restrict our subsequent analysis
with the domain of small angles of attack. For such values of ˛G , the following
relation holds:
Cn .˛G / D Cn˛ ˛G (6)
where Cn˛ is a positive constant.

Introduce nondimensional time D b=V0 where b is the airfoil chord and V0
is characteristic flow speed. Then the following nondimensional coordinates and
parameters appear:
V l 2D 2J N D L;
uD ; lN D ; DN D ; JN D ; L
V0 b S V0 b 2 S b 3 b
2m 2kb 2h 2H b 2
N D ; N D
m ; kN D ; hN D ; HN D
b S b S V 2 S V S V
In order to simplify the notation, denote the derivative with respect to with dot and
omit bars over nondimensional values. Then, taking into account (6), the motion
equations look as follows:

J
R C 2m
P P C m2
R C lm
P 2 D D
P C l ku2 C huP C H u
P 2 (7)

mR C ml
R m
P 2 D ku2 huP H u
P 2 Cn˛ u2
C u.l C L/
P C uP (8)
In what follows we suppose that the moment of inertia J of the pendulum is

large.
3 Stability of the Along the Flow Equilibrium
Motion equations linearized in the vicinity of the abovementioned equilibrium are

as follows:

J
R D D
P C l ku2 C huP (9)

mR C ml
R D ku2 huP Cn˛ u2
C uL
P C uP (10)
The Hurwitz criterion yields a necessary and sufficient condition of the asymp-
totic stability of the “along the flow” equilibrium that can be written in the following
structural form:
Q D A2 k 2 C A1 k C A0 > 0 (11)
or
X
6
QD Bi l i > 0 (12)
iD0
where A0;1;2 , B0::6 are some combinations of system parameters. In particular,
A2 D u4 .ul.l C L/Cn˛ C D/ ..J mlL/J uCn˛ C hu.J C ml 2 /2 C Dm2 l 2 /;

B6 D mhCn˛ ku6 .Cn˛ h C km/ > 0;

B0 D D.h C Cn˛ /u3 k J 2 ku2 C DJ u.h C Cn˛ / C D 2 m > 0 (13)
Note that A2 > 0 for large enough J .

In order to understand the behavior of the system at different values of parame-
ters, consider first some limit cases (taking into account that J 1).
For small values of k we obtain the following relations for the roots of the
characteristic polynomial:
s
D .l C L/.C ˛
C h/ m Cn˛ lh
1;2 D Cn˛ lhu
n
2 ˙ iu C O.J 3=2 /
2J 2J Cn˛ C h J.Cn˛ C h/
uk Cn˛ C h
3 D C o.k/; 4 D u C o.1/
h m
420 Y.D. Selyutskiy
Under the assumption that m < L.Cn˛ C h/ real parts of these roots are negative,
hence, the equilibrium is asymptotically stable.
For larger values of k the following relations hold:
r
D Cn˛ ul..l C L/k Cn˛ / Cn˛ l
1;2 D ˙ iu C O.J 3=2 /
2J 2kJ J
q 2
.h C Cn˛ / C h C Cn˛ 4km
3;4 D u C o.1/
2m
From the previous formulae for 1;2 one can readily show that the studied equilib-
rium is unstable, if parameter l satisfies the following inequalities:
s 2 s 2
Cn˛ kL Cn˛ kL D Cn˛ kL Cn˛ kL D
< l < C
2k 2k uCn˛ 2k 2k uCn˛
Note, in particular, that if Cn˛ < kL, then for all l > 0 the equilibrium is
asymptotically stable. However, if Cn˛ > kL, then there exists a range of positive l
where the stability is lost, at least, for large enough J .
For h 1 roots of the characteristic polynomial look as follows:
r
D Cn˛ ul.l C L/ Cn˛ l
1;2 D ˙ iu C O.J 3=2 /
2J 2J J
hu uk
3 D C O.1/; 4 D C o.h1 /
m h
Apparently, in this case the “along the flow” equilibrium is asymptotically stable.
Note that from Hurwitz criterion one can readily obtain that for h D 0 the
equilibrium is asymptotically stable, if
uCn˛ 2 l
k> C O.J 1 / (14)
uCn˛ l.l C L/ C D
Note also that from (13) it follows that for l 1=J 1 and for l 1 the
equilibrium in question is asymptotically stable.
Basing on the above results, as well as on representations (11–12) and for-
mula (14), the evolution of the asymptotic stability domain in the plane of
parameters .k; h/ with the change of l is qualitatively shown in Fig. 2. Here solid
lines mean boundaries of stability domains at different l (areas located below each
of these lines are domains of instability for the corresponding value of l).
Consider now the situation when the equilibrium position “along the flow” is
unstable in more detail.
Fig. 2 Evolution of stability domains when the holder length changes (l1 < l2 < l3 )
4 Limit Cycles
Let’s search for limit cycles arising in the system in case of unstable position “along
the flow” in the following form:

.t / D
c sin.!c t /; .t / D c sin.!c t C ˚/ (15)
In order to determine the characteristics of the cycle, use the conventional

harmonic balance method. Suppose that !c 1 and J !c2 D O.1/. Then,
substituting (15) into (7–8) and leaving only terms up to the first order of smallness
yield the following equations:
J
c !c2 C Cn˛ l
c u2 Cn˛ luc !c sin ˚ D 0
Cn˛ luc !c cos ˚ C Cn˛ .l C L/u

c !c C D
c !c D 0
Cn˛ u2
c C c ku2 cos ˚ uhc !c sin ˚
1
uCn˛ c !c sin ˚ u3c !c H sin ˚ D 0 (16)
4
Cn˛ .L C l/u
c !c C c ku2 sin ˚ C uc !c h cos ˚C
1
Cn˛ uc !c cos ˚ C uc H !c cos ˚ D 0
4
Note that from the second equation of (16) it follows that cos ˚ < 0.
One can readily show that amplitudes of
.t / and .t / and the phase difference
can be expressed in terms of !c with the following relations:
422 Y.D. Selyutskiy
Cn˛ luc cos ˚

c D ;
Cn˛ l.l C L/u C D
s
.!c D.h C Cn˛ / C !c hCn˛ l.l C L/u C .Cn˛ ul.l C L/ C D/ku tan ˚/
c D 2 ;
H !c .Cn˛ l.L C l/u C D/
J !c2 Cn˛ lu2

tan ˚ D
.Cn˛ l.l C L/u C D/!c
(17)
Solving the system (16) yields the following formula for the frequency of the limit
cycle:
v
u q 2
u
u 2kJ 2 !02 kX12 C DulCn˛ 2 4X1 X2 kJ 2 !02 C kX12 DulCn˛ 2
!c D t
2J kJ Cn˛ 2 .L C l/l
(18)
where
r
Cn˛ l
!0 D u ; X1 D D C uCn˛ l.l C L/; X2 D ulCn˛ 2 kX1
J
In order to verify the accuracy of the obtained approximation, numerical
calculations were performed using the following values of parameters:
J D 500; Cn˛ D 5; D D 1; u D 0:5

m D 1:5; k D 1; h D 5; L D 1; H D 200
Parameter l varied from 0 to 2.

Comparison between exact solution of the motion equations and approxima-
tions (17)–(18) is shown in Fig. 3 a,b (boxes denote exact solution, and solid lines
denote approximations).
Apparently, the quality of approximation of the cycle frequency can be deemed
quite reasonable (Fig. 3a) in the whole range of values of l, for which the limit cycle
exists. The quality of approximation of the cycle amplitude (Fig. 3b) is good enough,
though it gets somewhat worse near the maximum of the curve and near the ends of
the interval where the limit cycle exists.
5 Comparison with Experiments
In order to verify the model, experiments were performed in the subsonic wind
tunnel of the Institute of Mechanics of Lomonosov Moscow State University.
Characteristics of the wind tunnel are summarized in the following Table 1:
a b
Fig. 3 Cycle frequency (a) and cycle amplitude (b) as functions of l: comparison between exact
solution and approximation formulae
Table 1 Wind tunnel characteristics

Characteristic Value
Dimensions of the working part 3m x 5m, elliptic
Flow speed range 0–50 m/s
Nonuniformity of the wind speed across the cross section 0:5 %
Flow angularity (both in horizontal and vertical directions) < 0:25 deg
Gradient of the static pressure 0:002 m1
Longitudinal component of the turbulence intensity < 0:2 %
Table 2 Parameters of the experimen-

tal setup
Parameter Case 1 Case 2 Case 3
l 1.12 0.69 0.25
J 419 338 289
The experimental setup represented a pendulum consisting of a frame (made

of thin metal rods) and wooden wing installed inside the frame. The structure
was mounted in the wind tunnel so that its rotation axis was vertical. The wing
parameters were as follows: chord 0:07 m, length 0:32 m, and maximum thickness
0:009 m. Position of the wing with respect to the frame could be changed. Test
series were performed for three configurations. Nondimensional parameters of the
pendulum for all cases are given in Table 2. The wind speed corresponding to
Reynolds number Re D 105 was taken as the characteristic flow speed V0 .
424 Y.D. Selyutskiy
Fig. 4 Normal force coefficient from steady experiments vs steady angle of attack (rad.)
In order to assess the structural damping, a series of experiments was carried

out without flow where the rotation axis was horizontal (so that the system behaved
like a conventional physical pendulum). In the result, the following estimation was
obtained for the nondimensional damping coefficient: D D 1:6.
In order to determine static characteristics of the airfoil, a series of experiments
was performed, where the pendulum was fixed at different angles, and aerodynamic
load was measured acting on the wing. Results are presented in Fig. 4 for two values
of the wind speed (boxes denote experimental values of the normal force coefficient
Cn , and solid line denotes the approximation curve Cn .
/ D Cn˛
, where Cn˛ D 5:2.
Evidently, agreement with experiments is good for smaller values of
(up to
about 0.15 rad.).
Then dynamic experiments were performed at different flow speeds ranging from
8mps to 25mps (which corresponds to Reynolds numbers from 3:7 104 to 1:2
105 ). In the course of testing, the pendulum was deflected from the “along the flow”
equilibrium for a certain angle, after which it was released and moved freely. The
pendulum rotation angle
was registered as a function of time using a conventional
encoder.
It was found that for case 1 the “along the flow” equilibrium remained asymp-
totically stable in the entire range of wind speeds. In the same time, for cases 2
and 3 in a certain range of wind speeds (approximately from 10mps to 17mps), this
equilibrium became unstable, and a limit cycle appeared. This fact agrees with [8].
Evidently, this phenomenon is due to flow rearrangements taking place in this
interval of Reynolds numbers. Such rearrangements having considerable influence
upon the interaction between the body and the flow were registered and discussed,
in particular, in [9]. In terms of the proposed model, this situation can be interpreted
as attached oscillator parameters’ dependence on the Reynolds number (or on the
normalized flow speed u).
Series of numerical simulations were performed, and parameters of the attached
oscillator were chosen (if necessary, as functions of u) in such a way as to provide
possibly good agreement with experimental data. In the result, the following values
and dependences were selected:
a b
Fig. 5 Cycle frequency (a) and cycle amplitude (b): comparison between experimental data and
numerical simulation
m D 1:5; Cn˛ D 5:2; H D 200; L D 1

8 8
< 1:0; u 0:61 < 7:5; u 0:61
k D 2u 0:19; u < 0:99; h D 31:16.u 0:61/2 C 7:5; u < 0:99
: :
2:4; u 0:99 12:0; u 0:99
Simulation and test data (limit cycle amplitudes and frequencies) are shown in
Fig. 5 a,b. Experimental values are represented with boxes (for the case 2) and
crosses (for the case 3), and the corresponding calculated curves are shown with
solid (case 2) and dashed (case 3) lines. In case 1, the studied equilibrium is asymp-
totically stable both in simulation and in wind tunnel experiments (as mentioned
before) in the whole range of test wind speeds.
Evidently, calculation results qualitatively agree with experimental data. How-
ever, the difference between measured and calculated amplitudes becomes larger
near the ends of the range of u where the limit cycle exists. This is quite natural, as
the cycle amplitude depends rather strongly on u in these areas, and even small error
in the determination of the mentioned end points causes relatively large errors in
the amplitude. Besides, calculated frequencies in the second case are slightly lower
than measured ones. These discrepancies can be due to the presence of additional
dissipation factors in the system and to neglection of aerodynamic load upon the
frame of the pendulum.
426 Y.D. Selyutskiy
6 Conclusion
This work proposes a novel nonlinear phenomenological model for simulating

unsteady interaction between wing-type body and flow. This model is used for the
examination of dynamics of the aerodynamic pendulum. Basing on it, stability of
the “along the flow” equilibrium of this object is studied, and stability domains
are determined for different combinations of system parameters. Formulae are
obtained for amplitude and frequency of limit cycles arising upon stability loss
of this equilibrium. Comparison with results of wind tunnel test performed in the
Institute of Mechanics of Lomonosov Moscow State University shows that the
model allows giving adequate qualitative description of experimentally registered
features of behavior of the aerodynamic pendulum. This gives ground for applying
the model for performing qualitative analysis of dynamics of different aeroelastic
systems.
Acknowledgements The work is supported by Russian Foundation for Basic Research (grant No.
12-01-00364).
References
1. Andronov, P.R., Guvernyuk, S.V., Dynnikova, G.Ya.: Vortex Method of Calculation of

Unsteady Hydrodynamic Loads. Moscow University Press, Moscow (2006)
2. Belotserkovsky, S.M., Lifanov, I.K.: Method of Discrete Vortices. CRC Press, Boca Raton,
Florida (1993)
3. Goman, M.G., Khrabrov, A.N.: State-space representation of aerodynamic characteristics of
an aircraft at high angles of attack. J. Aircraft 31, 1109–1115 (1994)
4. Hansen, M.H., Gaunaa, M., Madsen, H.A.: A Beddoes-Leishman type dynamic stall model in
state-space and indicial formulations. Risø-R-1354(EN) (2004)
5. Lokshin, B.Ya., Samsonov, V.A.: On heuristic model of aerodynamical pendulum. Fundam.
Prikl. Mat. 4 (3), 1047–1061 (1998)
6. Marsden, C.C., Price, S.J.: The aeroelastic response of a wing section with a structural freeplay
nonlinearity: An experimental investigation. J. Fluids Struct. 21, 257–276 (2005)
7. Meyer, M., Matthies, H.G.: State-space representation of instationary two-dimensional airfoil
aerodynamics. J. Wind Eng. Ind. Aerod. 92, 263–274 (2004)
8. Poirel, D., Harris, Y., Benaissa, A.: Self-sustained aeroelastic oscillations of a NACA0012
airfoil at low-to-moderate Reynolds numbers. J. Fluids Struct. 24, 700–719 (2008)
9. Poirel, D., Price, S.J.: Random binary (coalescence) flutter of a two-dimensional linear airfoil.
J. Fluids Struc. 18, 23–42 (2003)
10. Runyan, H.R.: Single-degree-of-freedom-flutter calculations for a wing in subsonic potential
flow and comparison with an experiment. NACA Reports 1089, 1–8 (1952)
11. Samsonov, V.A., Sapunkov, G.I.: Stability research of double pendulum construction in a flow.
In: Proceedings of 5th International Conference Of Cranes and Textile Mach. EUROCRANE
96, 250–256 (1996)
12. Samsonov, V.A., Selyutskiy, Yu.D.: Phenomenological model of interaction of a plate with a
flow. J. Math. Sci. 146(3), 5826–5839 (2007)
13. Selyutskiy, Yu.D., Samsonov, V.A., Andronov, P.A.: On oscillations of aerodynamic pendulum.
Int. J. Struct. Stability Dynam. (2013), doi: 10.1142/S0219455413400105
14. Shao, S., Zhu, Q., Zhang, C., Ni, X.: Airfoil aeroelastic flutter analysis based on modified
Leishman-Beddoes model at low mach number. Chinese J. Aeronaut. 24, 550–557 (2011)
15. Strganac, T.W., Ko, J., Thompson, D.E., Kurdila, A.J.: Identification and control of limit cycle
oscillations in aeroelastic systems. J. Guid. Contr. Dynam. 23(6), 1127–1133 (2000)
Vibratory Energy Localization by Non-smooth
Energy Sink with Time-Varying Mass
Alireza Ture Savadkoohi and Claude-Henri Lamarque
Abstract We study nonlinear interactions of two coupled oscillators at different

time scales. The main oscillator which is linear is coupled to a nonlinear energy
sink with non-smooth (piecewise linear) potential and time-dependent mass. The
overall time is embedded to fast and slow time scales and the behavior of the system
at each one of them is revealed.
The invariant of the system at fast time scale is detected. Then we try to
have further information about the overall system behavior at the first slow time
scale. Finally, analytical developments are compared with numerical results and the
possibility of the passive control of the main system by means of the time-dependent
NES is commented upon.
1 Introduction
A special kind of light oscillators with essentially nonlinear potential (mainly

cubic) is called nonlinear energy sink (NES). It can be coupled to some important
systems for passively controlling them and localizing their vibratory energy [16,17].
The efficacy of NES devices has been verified and proved experimentally as well
[3, 4, 13]. Some works were carried out to consider and to design other types
of nonlinear NES: namely, vibro-impact [1, 10], non-polynomial [2], and non-
smooth [5, 14]. Some works have been carried out in order to consider non-smooth
main structure to be controlled by smooth or non-smooth NES systems [6, 11, 15].
In all of those systems, masses of the main system and NES were constant. [7]
studied the energy pumping phenomenon between a linear system with time-
dependent mass and a non-smooth NES. In this paper we try to study the energy
pumping phenomenon between a linear oscillator and a non-smooth NES where the
mass of the NES is time-dependent. The organization of the paper is as follows:
General presentation of the system, its rescaling process, shifting to the complex
A. Ture Savadkoohi () • C.-H. Lamarque

ENTPE, Université de Lyon, LGCB and LTDS UMR CNRS 5513, Rue Maurice Audin,
69518 Vaulx en Velin, France
e-mail: alireza.turesavadkoohi@entpe.fr; lamarque@entpe.fr

430 A. Ture Savadkoohi and C.-H. Lamarque
domain, and averaging are explained in Sect. 2. Behavior of the system at fast and
slow time scales is described in Sect. 3. A numerical example is given in Sect. 4 and
results are compared with analytical predictions. The application of the proposed
system in the passive control of main structures is demonstrated in Sect. 5. Finally
summary of the work is collected in Sect. 6.
2 The System and Preliminary Treatments
Let us consider an academic model of a 2-dof system: it consists of a main structure

with mass, damping and elastic stiffness as M , C , and k1 which is coupled to a
non-smooth NES with the time-dependent mass m.t Q /. The mass m Q can move freely
in a distance equal to 2ı until it reaches to elastic springs, k2 , at two sides. The
overall damping of the NES system is supposed to be . Q If we assume that the mean
velocity at which the mass is “leaving or entering” the mass of the NES in the x
direction is zero [18], then governing equations of the system can be summarized as
Q xP 1 xP 2 / D sin.˝t /
M xR 1 C C xP 1 C k1 x1 C FQ .x1 x2 / C .
(1)
Q xP 2 xP 1 / C m.t
Q /xR 2 C FQ .x2 x1 / C .
m.t PQ /xP 2 D 0
The non-smooth potential FQ of the NES is defined as

8
<0 ı z ı
@V .z/
FQ .z/ D D FQ .z/ D k2 .z ı/ zı (2)
@z :
k2 .z C ı/ z ı
Q / reads
and the rule of the time dependency of the mass m.t
Q / D m0 .1 C m.t //
m.t (3)
is a small parameter which corresponds to the ratio of the initial mass of NES and
m0
the mass of the main system, i.e., 0 < D << 1. We assume that .1C m.t //
M
0 for a time long enough. We would r like to rescale the system with respect to the
k1
new time domain T where T D t D t #. Equation (1) reads (xi .t / ! yi .T /):
M
yR1 C yP1 C y1 C FO .y1 y2 / C .yP1 yP2 / D f0 sin.!T /

(4)
yR2 C .1 C m.T //1 .yP2 yP1 / C .1 C m.T //1 FO .y2 y1 /
C 2 .1 C m.T //1 m.T
P /yP2 D 0
Vibratory Energy Localization by Non-smooth Energy Sink with Time-Varying Mass 431
C 1 Q 1 k2 Q 1
where p D , F D FO , k D , p D , D f0 , and
M k1 k1 k1 M k1 k1
˝
D !. We assume that k D o. 0 / and scaled potential of the NES is
#
8
<0 ı z ı
FO .z/ D k.z ı/ zı (5)
:
k.z C ı/ z ı
Taking into account that .1 C m.T //1 ' .1 m.T // C o. 2 /, we will have
yR1 C yP1 C y1 C FO .y1 y2 / C .yP1 yP2 / D f0 sin.!T /

(6)
yR2 C .1 m.T //.yP2 yP1 / C .1 m.T //FO .y2 y1 /
C 2 .1 m.T //m.T
P /yP2 D 0
Let us transfer the system to the coordinates v D y1 C y2 and w D y1 y2 ;

equation (6) will be equivalent to
1 1
vR C .Pv C w/
P C P C 2 m.T /FO .w/
.v C w/ C 2 m.T /w
1C 1C
1
C 2 .1 m.T //m.T
P / .Pv w/
P D f0 sin.!T /
1C
1 1
wR C .Pv C w/
P C .v C w/ C .1 C m.T //wP
1C 1C
1
C.1 C m.T //FO .w/ .1 m.T //m.T
P / .Pv w/
P D f0 sin.!T /
1C
(7)
p complex variables of [8] ('1 e
We introduce i!T
D vP C i !v and '2 e i!T D wP C i !w
with i D 1) to the system. We can present the functions FO .w/ and m.T / in the
form of Fourier series:
i C1
X
FO .w/ D FO .'2 e i!T '2 e i!T / D fj .'2 ; '2 /e i!j T (8)
2! j D1
C1
X
m.T / D mj e i!j T (9)
j D1
so,
C1
X
P /D
m.T i !j mj e i!j T (10)
j D1
where the : represents the complex conjugate of the function under consideration.
C1
X C1
X
We assume that jmj j < C1 and j!j mj j < C1. We would like
j D1 j D1
to analyze the averaged form of equations (7) around the 1 W 1 resonance, i.e.,
! D 1 C , by studying its detailed components at different scales of the . After
ignoring some obvious higher-order terms of in system (7) we will have
i i
'P1 C !'1 C .'1 C '2 / .'1 C '2 / D
2 2.1 C / 2.1 C /!
i f0

2
i i (11)
'P2 C !'2 C .'1 C '2 / .'1 C '2 /
2 2.1 C / 2.1 C /!
i 1C
.1 C /'2 G1 .j'2 j2 / C m1 G2 .j'2 j2 / C '2
2 2
! i f0
.m0 '2 C m2 '2 / C i m2 .'1 '2 / D
2 1C 2
where the functions G1 .%/ and G2 .%/ are defined as [14]

8 p
ˆ %
ˆ
<0 if <ı
! !
G1 %/ D p
ˆ k 2ı! p ı! % (12)
:̂ % .ı!/ C 2 arccos p
2 if ı
! % % !
8 p
ˆ %
<0 if <ı
G2 %/ D p p! (13)
:̂ 2k % .ı!/2 if %
ı
! !
We introduce fast (T D 0 ) and slow (1 D 0 , : : :) time scales to the system and
then we will consider its behavior at different orders of [9].
3 Time Multi-scale Analysis of the System
0 order of the (11) is equivalent to
@'1
D 0 ) '1 D '1 .1 ; 2 ; : : :/
@0 (14)
@'2 i 1 G1 j'2 j2 C i
C '2 D '1
@0 2 2
@'2
Fixed points of the system, i.e., ˚.1 / (0 ! 1, D 0), can be defined as
@0

i 1 G1 j˚j2 C i
˚D '1 (15)
2 2
By considering '1 D N1 e iı1 and ˚ D N2 e iı2 , the following invariant manifold of

the system at 0 time scale can be obtained:
q 2
N1 D N2 2 C 1 G1 .N22 / (16)
Stable areas of the linearized system at the 0 time scale can be defined by the
following criterion: (H.N22 / D 2N2 G 0 .N22 /) [14]:

2 C 1 G1 .N22 / 1 H.N22 /N2 G1 .N22 / > 0 (17)
Let us study the system at the 1 order; equations (11) read:

@'1 1 i
C i C C 2i '1 i'2 D f0
@1 2 2
@'2 i
C 2G2 j'2 j2 m1 C C i.1 C / '1 C 2i m2 '2 (18)
@1 2
!
i

C iG1 j'2 j C m0 C i.1 C / '2 m2 .2i C /'2 D f0
2
2
We would like to analyze the behavior of the structure at the slow time scale 1
around obtained invariant manifold at 0 time scale. By imposing equation (15) in
the second equation of the system (18), assuming '1 D N1 e iı1 , '2 D N2 e iı2 and
taking into account general forms for m0 , m1 and m2 as
mj D mjr C i mj i j D 0; 1; 2 (19)
the following reduced order forms of system equations can be obtained:
@ı2
D ±1 .ı2 ; N2 /
@1
(20)
@N2
D ±2 .ı2 ; N2 /
@1
where
1 1 1
N2 ±1 .ı2 ; N2 / D G1 .N22 /N2 C N2 m0i N2 N2 C G1 .N22 /N2
2 2 2
3
f0 cos.ı2 / G2 .N2 /m1i cos.ı2 / C G2 .N2 /m2r N2 cos.2ı2 / C N2 m2i cos.2ı2 /
2 2
2
3
CG2 .N2 /m1r sin.ı2 / C G2 .N2 /N2 m2i sin.2ı2 / N2 m2r sin.2ı2 /
2 2
2
(21)
1 1 1 1
±2 .ı2 ; N2 / D N2 C G1 .N22 /N2 C N2 m0r C N2
2 2 2 2
3
G2 .N22 /m1r cos.ı2 / G2 .N22 /N2 m2i cos.2ı2 / C N2 m2r cos.2ı2 / f0 sin.ı2 /
2
3
G2 .N2 /m1i sin.ı2 / C G2 .N2 /N2 m2r sin.2ı2 / C N2 m2i sin.2ı2 /
2 2
2
(22)
Equilibrium points of the system can be obtained by setting ±1 D ±2 D 0.
Stable areas of the invariant manifold of the system at 0 time scale which are given
in Eq. (17) are characterized by two border lines, namely fold lines, which can be
obtained by the following equation:

D D 2 C 1 G1 .N22 / 1 H.N22 /N2 G1 .N22 / D 0 (23)
Ture Savadkoohi et al. [10] demonstrated that simultaneous intersections of these

two lines with the ±1 D ±2 D 0 provide singularities of the system that are called
fold singularities.
4 A Numerical Example
We choose a smooth and periodic profile of mass for the NES which reads

m.T / D m0r C2 m1r cos.!T /m1i sin.!T / C2 m2r cos.2!T /m2i sin.2!T /
(24)
For our numerical example we assume that m0r D 1:57, m0i D 0, m1r D 0, m1i D
0:5, m2r D 0, and m2i D 0:25. Assumed profile of the mass of the NES is depicted
in Fig. 1. Let us consider the following system parameters: D 103 , k D 1:5,
ı D 1, D 0:2, D 0:1, D 1. Moreover, we suppose that the system is under
external forcing term f0 D 1:176. Dynamics of the system until infinity of the 1
time scale is depicted in Fig. 2.
We can see that the system possesses one fold singular point (no: 1) and an
equilibrium point (no: 2). A zoomed area which covers these points is depicted at
the same figure. Phase portraits of the system around mentioned points are depicted
in Figs. 3 and 4. The singular point no: 1 is saddle and unstable and the equilibrium
point no: 2 is unstable as well.
2.5
2
m(T)
1.5
0.5
0
0 5 10 15 20 25 30
T (Sec)
Fig. 1 The profile of periodically varying mass of the NES: m0r D 1:57, m1i D 0:5, m1r D 0,
m2i D 0:25, and m2r D 0
4
N2
2
1
2
0
0 1 2 3 4 5 6
δ2
1.6
1.4
1
N2
1.2
1 2
0.8
0.6
2.6 2.65 2.7 2.75 2.8 2.85 2.9
δ2
Fig. 2 Positions of equilibrium and fold singularities of the system under f0 D 1:176: ±1 D 0
( ), ±2 D 0 (—), D D 0 or fold lines ( )
1.24
1.235
N2
1.23
1.225
1.22
2.71 2.715 2.72 2.725 2.73 2.735 2.74
δ2
Fig. 3 Phase portraits of the system around fold singular point no: 1: f0 D 1:176, D 1
1.17
1.16
1.15
N2
1.14
1.13
1.12
2.78 2.79 2.8 2.81 2.82

δ2
Fig. 4 Phase portraits of the system around fixed point no: 2: f0 D 1:176, D 1
In order to have better envision about the behavior of the system, we present
numerical results which are obtained by direct integration of the system (4) by Mat-
lab package. Assumed initial conditions are y10 D 1:5 and y20 D yP10 D yP20 D 0.
The invariant manifold of the system and its stability borders which are obtained
from equations (16) and (17), respectively, are illustrated in Fig. 5. It can be seen
that although the analytical invariant manifold is obtained based on keeping first
harmonic and truncating higher ones, the averaged form of the real behavior of
the system, which contains all harmonics, follows this invariant. When the system
reaches to stability border of the invariant, it faces bifurcation to jump to other stable
branch. Histories of the change of system amplitudes are depicted in Figs. 6 and 7.
Fig. 5 Invariant manifold (), stability borders ( ), and numerical results (—). Starting
values are .N2 ; N1 / D .1:5; 1:5/: f0 D 1:176, D 1
7
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=1
6 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
m2r=0, f0=1.176
5
4
N2
0
0 2 4 6 8 10
T
x 104
Fig. 6 History of N2 : f0 D 1:176, D 1
These figures show that the system experiences one reverse bifurcation and finally is
attracted by the equilibrium point no:2. The reason for this behavior is the existence
of fold singular point no: 1 which is saddle (see Figs. 2 and 3) that makes the system
aims to the first fold line to be ready to present a reverse bifurcation (see Figs. 5, 6,
and 7). After one complete cycle of bifurcations that forces the system to present
beating response, it finally aims to the fixed point no: 2 that is unstable during the
very long time for 1 time scale, i.e., 1 D 12 D 106 Sec. Although this equilibrium
2
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=1
1.8 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
1.6 m2r=0, f0=1.176
1.4
N1
1.2
0.8
0.6
0.4
0 2 4 6 8 10
T
x 104
Fig. 7 History of N1 : f0 D 1:176, D 1

8
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=1
6 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
4 m2r=0, f0=1.176
2
y2
−2
−4
−6
−8
0 2 4 6 8 10
T x 104
Fig. 8 History of y2 : f0 D 1:176, D 1
point is unstable after a very long time span, but makes the system oscillates around
it for quite long time. This oscillation corresponds to low energy levels for the main
system that is interesting from passive control view point.
The displacement histories of two oscillators which are illustrated in Figs. 8 and 9
show that the main structure during the time scale 1 behaves in a controlled manner
with very low displacement levels. Other behaviors can be observed for the system
at other time scales, but we are not going to discuss about them and they are out of
the scope of the current work.
1.5
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=1
1 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
m2r=0, f0=1.176
y1 0.5
−0.5
−1
−1.5
0 2 4 6 8 10
T x 104
Fig. 9 History of y1 : f0 D 1:176, D 1
5 Passive Control of Main Systems by NES

with Time-Dependent Mass
In order to show the capability of the NES with varying mass in controlling main
structures, we present a numerical example by showing the displacement histories
of a main structure without and with coupled NES. We consider a worst case for
the main system when it is in an exact 1 W 1 resonance with the external forcing
terms. This means that we set D 0 so ! D 1. The system in both cases is under
external forcing term f0 D 1:176 and assumed initial conditions and also mass
profile for the NES are the same as the described example in Sec. 4. The results
of the direct integration of system equations for two cases are depicted in Fig. 10.
This figure shows that while the single dof main structure (without coupled NES) is
presenting a kind of diverging behavior at the exact 1 W 1 resonance, by coupling a
NES with varying mass we are able to control the same structure. Its displacement
is reduced in a considerable amount and it presents permeant strongly modulated
response (SMR) (at least until the maximum time of integration) [5, 12]. This SMR
is more distinct by looking at the histories of amplitudes of the main system and
the NES which are illustrated in Figs. 11 and 12 showing a beating response of two
oscillators due to the SMR.
6 Conclusions
We studied the energy exchange problem between a linear system and a non-smooth
energy sink whose mass is time-dependent. Detected invariant manifold of the
system at slow scale and its stable zones let us predict the overall surfaces of all
15
single dof main system
2 dof system (main dof+NES)
10
y1 5
−5
−10
−15
0 2 4 6 8 10
T
x 104
Fig. 10 Histories of y1 for two cases: the bare structure without coupled NES and 2-dof system.
Parameters of the main system are D 0:1, D 0, f0 D 1:176. Parameters of the coupled NES
are k D 1:5, ı D 1 and D 0:2, m0r D 1:57, m0i D 0, m1r D 0, m1i D 0:5, m2r D 0, and
m2i D 0:25
7
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=0
6 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
m2r=0, f0=1.176
5
4
N2
0
0 2 4 6 8 10
T
x 104
Fig. 11 History of N2 and beating response of the NES due to the SMR: f0 D 1:176, D 0
fixed and/or pseudo-fixed points of the system by explanations of global bifurcation

due to the existence of unstable zone. Enlightened system behavior at first fast time
scale let us take further steps for detecting equilibrium points and fold singularities
which can explain the behavior of the system during closed-loop bifurcations and/or
its attractions to fixed points. By demonstrating numerical examples we showed the
capability of the non-smooth nonlinear energy sink in passively controlling a main
1.6
k=1.5, δ=1, λ=0.2, ξ=0.1, σ=0
1.4 m0r=1.57, m1i=0.5, m1r=0, m2i=0.25
m2r=0, f0=1.176
1.2
1
N1
0.8
0.6
0.4
0.2
0 2 4 6 8 10
T
x 104
Fig. 12 History of N1 and beating response of the main oscillator due to the SMR: f0 D 1:176,
D0
structure during a worst case, i.e., exact 1 W 1 resonance. Further studies should be
carried out for designing “optimized” nonlinear energy sink devices with varying
mass for the aim of passive control of civil/mechanical structures.
Acknowledgements The authors would like to thank the following organization for supporting
this research work: LABEX CELYA (ANR-10-LABX-0060) of the “Université de Lyon” within
the program “Investissement d’Avenir” (ANR-11-IDEX-0007) operated by the French National
Research Agency (ANR).
References
1. Gendelman, O.V.: Analytic treatment of a system with a vibro-impact nonlinear energy sink.
J. Sound Vib. 331, 4599–4608 (2012)
2. Gendelman, O.V.: Targeted energy transfer in systems with non-polynomial nonlinearity.
J. Sound Vib. 315, 732–745 (2008)
3. Gourdon, E., Alexander, N.A., Taylor, C.A., Lamarque, C.-H., Pernot, S.: (2007) Nonlinear
energy pumping under transient forcing with strongly nonlinear coupling: theoretical and
experimental results. J. Sound Vib. 300, 522–551 (2007)
4. Kerschen, G., Kowtko, J.J., McFarland, D.M., Lee, Y.S., Bergman, L.A., Vakakis, A.F.:
Experimental demonstration of transient resonance capture in a system of two coupled
oscillators with essential stiffness nonlinearity. J. Sound Vib. 299, 822–838 (2007)
5. Lamarque, C.-H., Gendelman, O.V., Ture Savadkoohi, A., Etcheverria, E.: Targeted energy
transfer in mechanical systems by means of non-smooth nonlinear energy sink. Acta Mech.
221, 175–200 (2011)
6. Lamarque, C.-H., Ture Savadkoohi, A., Etcheverria, E., Dimitrijevic, Z.: Multi-scales dynamics
of two coupled non-smooth systems. Int J. Bifurcat Chaos. 22, 1250295:1–18 (2012)
7. Lamarque, C.-H., Ture Savadkoohi, A., Dimitrijevic, Z.: Dynamics of a linear system with
time-dependant mass and a coupled light mass with non-smooth potential. Meccanica. (2013)
doi: 10.1007/s11012-013-9778-8
8. Manevitch, L.I.: The description of localized normal modes in a chain of nonlinear coupled
oscillators using complex variables. Nonlinear Dynam. 25, 95–109 (2001)
9. Nayfeh, A.H., Mook, D.T.: Nonlinear Oscillations. John Wiley and Sons, New York (1979)
10. Nucera, F., Vakakis, A.F., McFarland, D.M., Bergman, L.A., Kerschen, G.: Targeted energy
transfers in vibro-impact oscillators for seismic mitigation. Nonlinear Dynam. 50, 651–677
(2007)
11. Schmidt, F., Lamarque, C.-H.: Energy pumping for mechanical systems involving non-smooth
Saint-Venant terms. Int J. Nonlinear Mech. 45, 866–875 (2010)
12. Starosvetsky, Y., Gendelman, O.V.: Strongly modulated response in forced 2DOF oscillatory
system with essential mass and potential asymmetry. Physica D. 237, 1719–1733 (2008)
13. Ture Savadkoohi, A., Vaurigaud, B., Lamarque, C.-H., Pernot, S.: Targeted energy transfer with
parallel nonlinear energy sinks, part II: theory and experiments. Nonlinear Dynam. 67, 37–46
(2012a)
14. Ture Savadkoohi, A., Lamarque, C.-H., Dimitrijevic, Z.: Vibratory energy exchange between
a linear and a non-smooth system in the presence of the gravity. Nonlinear Dynam. 70,
1473–1483 (2012b)
15. Ture Savadkoohi, A., Lamarque, C.-H.: Dynamics of coupled Dahl type and non-smooth
systems at different scales of time. Int J. Bifurcat Chaos. 23, 1350114: 1–14 (2013)
16. Vakakis, A.F., Gendelman, O.V., Bergman, L.A., McFarland, D.M., Kerschen, G., Lee, Y.S.:
Nonlinear trageted energy transfer in mechanical and structural systems I. Springer, Berlin
(2008a)
17. Vakakis, A.F., Gendelman, O.V., Bergman, L.A., McFarland, D.M., Kerschen, G., Lee, Y.S.:
Nonlinear Trageted Energy Transfer in Mechanical and Structural Systems II. Springer, Berlin
(2008b)
18. Van Horssen, W.T., Pischanskyy, O.V.: On the stability properties of a damped oscillator with
a periodically time-varying mass. J. Sound Vib. 330, 3257–3269 (2011)
Coordinate-Free Formulation of Nonholonomic
Constraints for Wheeled Robots
Sanjuan Szklarz Paweł Cesar and Elżbieta Jarzȩbowska
Abstract A coordinate-free formulation of nonholonomic constraint equations for

wheeled vehicles is presented. The no-slip condition for the contact point of the
wheel with the ground is translated to a pure geometric formula. Based on this
general geometric interpretation, explicit formulas for nonholonomic constraints are
derived. The final calculations of the constraint equations are realized in Maxima, a
computer algebra system. The general no-slip condition formulation is validated via
the direct derivation of nonholonomic constraint equations for well-known robotic
systems. The presented approach provides a systematic procedure to calculate the
nonholonomic constraint equations for wheel components. The final equations are
suitable for symbolic calculations and computer aimed modeling of mechanical
systems.
1 Introduction
This paper presents a systematic coordinate-free approach to formulate the no-

slip condition for wheeled robot models. The no-slip condition ensures that the
contact point between a wheel and a ground has zero velocity. To calculate the
velocity of the contact point in a coordinate-free fashion, the wheel displacement
and rotations are defined as reference frame isomorphisms over time. This provides
constraint equations that can be calculated in any chosen coordinate system in the
state manifold.
Based on this general geometric interpretation, explicit formulas for nonholo-
nomic constraints are derived. The final calculations of the constraint equations are
realized in Maxima, a computer algebra system. The general no-slip condition for-
mulation is validated via the direct derivation of nonholonomic constraint equations
for well-known robotic systems. Presented calculations are symbolically realized
in the computer algebra system Maxima [9]. LaTeX formulas are automatically
generated.
S.S.P. Cesar () • E. Jarzȩbowska

Warsaw University of Technology, Institute of Aeronautics and Applied Mechanics,
Warsaw, Poland
e-mail: pszklarz@meil.pw.edu.pl; elajarz@meil.pw.edu.pl

444 S.S.P. Cesar and E. Jarzȩbowska
The novelty of the approach relies in the use of the symmetric properties of
the wheel geometry. The final nonholonomic equations which are obtained are
equivalent to the classical well-known equations [1, 2, 4–8]. However, they provide
a considerable improvement in symbolical and numerical calculations for complex
system models.
Nonholonomic constraint equations are the basis for the derivation of kinematic
control models for wheeled robots and car-like vehicles. Most of these constraints
are related to the no-slip condition of the contact point between wheels and the
ground. The constraint equations are formulated using selected reference frames
and coordinate systems. For multi-wheeled robots or wheeled vehicles of complex
designs, the formulation and validation of the resulting constraint equations are not
automated and it leads to many mistakes, and it is difficult to verify and debug
the final model formulation; additionally the transformations between a global and
a wheel reference frames are not shown explicitly; see, e.g., [1, 2, 4–8]. In the
case of linearization of the nonholonomic constraints, they require transformations
[10]. Also, there are vehicle models derived automatically by computer codes; see,
e.g., [11] and references therein. In [11] the code AutoSim was used to generate a
variety of linear and nonlinear bicycle models in symbolic forms. Yet, they are not
standard in the kinematic model derivations. The validation of models like in [11]
requires the verification of the mapping between handmade formulas and computer
model descriptions. In the absence of systematic tools for developing the constraint
equations, there exist many models derived under special assumptions that make the
constraint formulation viable. This approach, in turn, leads to models of the same
mechanical systems that are difficult to compare and adopt for other applications.
A coordinate-free formulation of nonholonomic constraints presented in the
paper does not require any special assumptions about a mechanical system model
and the resulting constraint equations are easily verifiable and repeatable by other
users.
2 Notation
It is assumed that the center of the wheel coincides with the origin of the wheel
reference frame. The constant r is the wheel radius. Let k be the global reference
frame and K the wheel reference frame. The wheel reference frame motion is
defined in terms of a translation:
T W k ! k; T .a/ WD a C % % 2 R3 (1)
and a rotation B 2 SO.3/—the 3D rotation group:
B W K ! k: (2)
Coordinate-Free Formulation of Nonholonomic Constraints for Wheeled Robots 445
Rotation transformations are defined using the base

0 1 0 1 0 1
1 0 0 cos a 0 sin a cos a sin a 0
B C B C B C
˝x .a/ WD @0 cos a sin aA ; ˝y .a/ WD @ 0 1 0 A ; ˝z .a/ WD @ sin a cos a 0A :
0 sin a cos a sin a 0 cos a 0 0 1
The translation transformation T is uniquely defined by the displacement vector

% 2 R3 . We use this relation to identify the space of translations in R3 .
Vectors in the global reference frame k are denoted by lower case letters and the
corresponding vectors in the wheel reference frame K are denoted by upper case
letters. This means that for A 2 K the corresponding a 2 k is
a D B.A/:
Displacement D D T ı B is defined as
D W K ! k; D./ WD T .B.//: (3)
For a mechanical system equipped with a wheel, let a state manifold be defined
by ˚ and a state variable q 2 ˚. The wheel reference frame location will be
specified with respect to the system state, i.e.,
B W ˚ ! SO.3/; T W ˚ ! R3 : (4)
For velocity calculation the following notations are used:
d
qP WD q;
dt
P q/ d @ d
B.q; P WD B.q/ D B.q/ q;
dt @q dt
d @ d
P q/
%.q; P WD %.q/ D %.q/ q:
dt @q dt
3 Geometric Formulation of the No-Slip Condition
The no-slip condition states that a contact point between a wheel and a ground
has zero velocity. In this section a detailed formulation of the no-slip condition is
presented. It is based upon the geometric transformation between the ground and
the wheel reference frames.
z
P
z
q
P
disk
x y= N
disk
f
x y= N
C C
2D projection of the wheel

3D projection of the wheel reference frame to the plane
reference frame. xz.
Fig. 1 The ground normal vector P and the wheel normal vector N in wheel reference frame
3.1 Contact Point Location
Define a contact point between the wheel and the ground as the lowest located point
with respect to the ground normal direction.
The contact point definition uniquely identifies a point on the wheel, and it
is independent from the fact that the wheel center location must match specific
conditions to ensure that the contact point in global coordinates is located in the
ground plane. Thanks to this property, the contact point for any system state q is
well defined and depends only on the rotation B.q/.
Following the upper/lower case letter convention described in Sect. 2, define
vectors (see Fig. 1):
• c; C is the vector pointing to the contact point position from the wheel center.
• p; P is the normal vector to the ground plane.
• n; N is the normal vector to the wheel plane.
Figure 1 shows the wheel normal vector N and the ground normal vector P in
the wheel reference frame. The following properties uniquely identify the contact
point vector C :
• It is located in the wheel plane: C N D 0.
• It is aligned with N; P : C 2 spanfP; N g.
• It is directed to the ground: C P < 0.
• It is located on the wheel circumference: kC k D r.
There is a singular case when the wheel lies on the floor, P D N . The presented
equations are not well defined in this case. In practice, such positions are strongly
unfortunate and should be avoided. This is why we do not discuss them.
The first approach to find C may be to project P on the place fN g? and
normalize it using the wheel radius r
P .P N /N
C D r :
kP .P N /N k
This solution is correct, but the term kP .P N /N k does not behave well
during the equation simplification.
The best approach seems to be the use of the following equation:
N P
C D rN . /: (5)
kN P k
Equation (5) is a direct consequence of the contact point geometric properties.

N P
The intermedium vector X WD kN P k
is a unitary vector perpendicular to N and P ,
located in the wheel plane. The contact point is then perpendicular to N and X
simultaneously. As N and X are unitary and perpendicular, we obtain equation (5).
Equation (5) exhibits better behavior during its symbolic simplifications. It is
due to the fact that in the ground reference frame the vector p D uz WD .0; 0; 1/T
and, in the wheel reference frame, the vector N D uy WD .0; 1; 0/T . Then, the term
kN P k is much simpler than kP .P N /N k. Note that vectors P and N in the
wheel reference frame are given by P D B 1 uz and N D uy . Then, equation (5)
expands to
uy B 1 uz
C D ruy . /: (6)
kuy B 1 uz k
Finally, note that for any vector v WD .vx ; vy ; vz /T the calculation
uy v D .vz ; 0; vx /T
is just a simple permutation and a sign change. This makes equation (6) easy to use.
Note that equation (6) is independent of any state manifold coordinate selection.
It is then the general formulation of the vector C pointing to the contact point
position.
3.2 Velocity of Wheel Fixed Points
In a given position of a mechanical system, the contact point is a fixed point in the
wheel reference frame. It means that the contact point position depends only on the
wheel position and not on velocity.
Location of the contact point in the global reference frame is given by
p.q/ WD D.C.q// D B.q/C.q/ C %.q/: (7)
Then, the velocity in the global frame is
P q/
p.q; P D %.q;
P q/ P q/C.q/;
P C B.q; P (8)
where the term B.q/CP .q; q/ P is not present as the vector C points to the instanta-
neous contact point between the wheel and the ground.
Expressing the contact point position in terms of the global reference frame C D
B 1 .c/, the contact point velocity vector can be written as
P q/
p.q; P D %.q;
P q/ P q/B
P C B.q; P 1 .c.q//: (9)
This leads to the well-known formula
P q/
p.q; P C !.q; q/
P D vo .q; q/ P c.q/; (10)
where vo D %P is the linear velocity of the wheel reference frame, operator !.q; q/
P
P q/B
./ WD B.q; P 1 ./ is the angular velocity of the wheel reference frame, and
c.q/ D B.q/C.q/ is the position of the contact point with respect to the global
reference frame orientation.
Then, the formula (10) and the no-slip condition lead to the nonholonomic
constraint formulation as
0 D vo .q; q/
P C !.q; q/
P c.q/: (11)
Equation (11) is the desired nonholonomic constraint equation. Note that it is not
related to any coordinate selection for the manifold ˚.
3.3 Final Formulation of the No-Slip Condition
Using the constraint equation (11) and the contact point location equation (6), we
can obtain a general formula for nonholonomic constraint equations for wheeled
vehicle models. It is of the form

P 1 uy B 1 .q/uz
0 D vo .q; q/
P C B.q; q/BP .q/ rB.q/.uy . // : (12)
kuy B 1 .q/uz k
It can be simplified to
P q/.u uy B 1 .q/uz
P C r B.q;
0 D vo .q; q/ P y . //: (13)
kuy B 1 .q/uz k
Since B.q/ 2 SO.3/, the calculation of B 1 .q/ is straightforward; in the matrix

representation B 1 .q/ D B T .q/.
Equation (13) is expressed in terms of the rotation B.q/ 2 SO.3/. It is
independent of any coordinate selections, and it does not need any additional
assumptions about the model adopted for a mechanical system. Equation (13)
presents the no-slip condition and it is the nonholonomic constraint for the wheel.
4 Examples—No-Slip Condition Validation

for Robotic Models
Using equation (13) nonholonomic constraints can be calculated for models of

well-known wheeled vehicles. The vertical and falling disk constraints calculated
using (13) match the equations in [3]. Note that to match the signs in the final
constraint equations as in [3], counterclockwise rotations have been adopted.
Models presented in [3] do not define the global to the wheel reference frame
transformation explicitly.
4.1 Vertical Rolling Disk
Following [3], the configuration space for the vertical rolling disk is Q D R2
S 1 S 1 and it is parametrized by the coordinates q D .x; y;
; /, where:
• x; y is the projection of the center of the wheel in the xy-plane,
•
is the roll angle of the disk, and
• is the heading angle of the disk.
The ground to the wheel transformation of reference frames is given by
Bv .q/ D ˝z .'/ ˝y .
/ :
Then
0 1
sin
Bv .q/1 uz D @ 0 A
cos
and
0 1
cos
uy Bv1 .q/uz D @ 0 A :
sin
Since uy and Bv .q/uz are perpendicular, then
kuy Bv .q/uz k D 1:
Then, we obtain
0 1
sin
uy Bv .q/uz
Cv D ruy . / D r @ 0 A: (14)
kuy Bv .q/uz k
cos
This is correct because in the global coordinates the contact point is always below
the center of the wheel and
0 1 0 1
sin
0
cv D Bv Cv D Bv @ 0 A D @ 0 A :
cos
1
Nonholonomic constraints (13) for the vertical rolling disk are

0 1 0 1
xP cos '
P
0 D vv C !v cv D vv C BP v .q; q/.C
P @ A C r @ sin '
P A :
v / D yP (15)
zP 0
4.2 The Falling Rolling Disk
Following [3], the configuration space for the falling rolling disk is Q D R3 S 1
S 1 S 1 , and it is parametrized by the coordinates q D .x; y; z;
; '; / where:
• x; y; z are the position coordinates of the center of the disk in the global reference
frame,
•
is the angle between the plane of the disk and the vertical axis,
• ' is the heading angle, and
• is the roll angle.
In [3], the coordinates are that of the contact point. Later we will change the
coordinate system to check that the constraints calculated in this paper match the
previous result.
The ground to the wheel transformation of reference frames is given by
Bf .q/ D ˝z .'/ ˝x .
/ ˝y . / :
Then
0 1
sin cos
Bf .q/1 uz D @ sin
A
cos cos
and
0 1
cos cos
uy Bf1 .q/uz D @ 0 A:
sin cos
Note that
kuy Bf1 .q/uz k D abs.cos

/:
Since
2 . 2 ; 2 /
kuy Bf1 .q/uz k D cos

:
Then, we obtain that

0 1
sin
uy Bf .q/uz
Cf D ruy . / D r @ 0 A:
kuy Bf .q/uz k
cos
Global coordinates of the contact point are

0 1 0 1
sin sin ' sin
rBf @ 0 A D r @ cos ' sin

A : (16)
cos cos
Nonholonomic constraints (13) for the falling rolling disk are

0 1
xP
0 D vf C !f cf D vf C BP f .q; q/.CP @ A
f / D yP C
zP
0 1
sin ' cos

P cos ' 'P sin
P cos '
r @ cos ' cos

P sin ' 'P sin
P sin ' A : (17)
sin

P
Equation (17) is not as simple as expected. This is because the constraints are
calculated assuming that .x; y; z/ are the coordinates of the center of the wheel.
Assume now that the coordinates of the contact point are used as provided by (16).
They yield
0 01 0 1 0 1
x x sin ' sin
@y 0 A D @y A C r @ cos ' sin

A ; (18)
z0 z cos
then
0 01 0 1 0 1
xP xP sin ' cos

P cos ' 'P sin
@yP 0 A D @yP A C r @ cos ' cos

P sin ' 'P sin
A (19)
zP0 zP sin

P
and equation (17) can be then rewritten as

0 01 0 1
xP P cos '
@yP0 A C r @ P sin ' A D 0: (20)
zP0 0
4.3 Tricycle Robot
A tricycle robot is presented in Fig. 2. The geometry of the tricycle is used to define
the state manifold and to illustrate the use of generalized coordinates. Note that
calculations of the nonholonomic constraints are realized based on equation (13),
i.e., using T .q/ and B.q/ only. Any change of a coordinate system will only
affect the translation T .q/ and rotation B.q/ mappings. The final calculation of
the nonholonomic constraints is automatic so that it can be performed symbolically
by any computer algebra system; in this paper Maxima [9] has been used.
The configuration space is parametrized by the coordinates q D Œx; y;
; ˇ; l ;
r ; f T :
• x; y—position of the rare axis center,
•
—heading angle of the tricycle body,
• ˇ—heading angle of the front wheel,
• l —left wheel roll angle,
• r —right wheel roll angle, and
• f —front wheel roll angle.
The ground to wheel transformations are denoted by
• Tl ; Bl —left wheel translation and rotation,
• Tr ; Br —right wheel translation and rotation, and
• Tf ; Bf —front wheel translation and rotation.
y ff
fl
q
d l
fr
Fig. 2 Tricycle model
Rotation transformations are as
Bl .q/ D ˝z .
/ ˝y .l / ;
Br .q/ D ˝z .
/ ˝y .r / ;

Bf .q/ D ˝z .
C ˇ/ ˝y f ;
and translation transformations as
Tl .q/ D .x; y; 0/T C ˝z .

/.0; d; 0/T ;
Tr .q/ D .x; y; 0/T C ˝z .
/.0; d; 0/T ;
Tf .q/ D .x; y; 0/T C ˝z .
/.l; 0; 0/T :
The final result for the nonholonomic constraints is presented. These formulas
are calculated symbolically in Maxima, and the LATEX equations are generated
automatically. The z component is omitted, as it is just 0 D 0.
Nonholonomic constraints (13) for the front wheel are

xP l sin

P C P f cos .
C ˇ/ 0
D : (21)
yP C l cos

P C P f sin .
C ˇ/ 0
Nonholonomic constraints (13) for the left wheel are

xP d cos

P C P l cos
0
D : (22)
yP d sin

P C P l sin
0
Nonholonomic constraints (13) for the right wheel are

xP C d cos

P C P r cos
0
D : (23)
yP C d sin

P C P r sin
0
The systematic approach to nonholonomic constraint equation derivation pre-

sented herein can be applied to any wheeled robotic system like the one presented
in [7, 8]. This approach allows a straightforward derivation of kinematic control
models for them.
5 Conclusions
The paper presents a systematic coordinate-free approach for the formulation of

the no-slip condition for wheeled robot models. Based upon this one formulation,
nonholonomic constraint equations can be derived for many wheeled robotic system
models in a unified, verifiable, comparable, and repeatable way. The constraint
equations, which are the basis for the derivation of kinematic control models for
the robotic systems, help in their derivation in a more systematic and simpler way.
Acknowledgements This work was supported by the Polish National Science Center (NCN)
under the Grant No.: 2011/01/N/ST8/01894.
References
1. Awrejcewicz, J.: Classical Mechanics: Dynamics. Advances in Mechanics and Mathematics.

Springer, NewYork (2012)
2. Basu-Mandal, P., Chatterjee, A., J.M. Papadopoulos, J.M.:Hands-free circular motions of a
benchmark bicycl. Proc. Roy. Soc. A: Math. Phys. Eng. Sci. 463(2084), 1983–2003 (2007)
3. Bloch, A.M.: Nonholonomic Mechanics and Control, vol. 24. Springer, NewYork (2003)
4. Bonilla, M., Reyes, F., Mendoza, M: Modelling and simulation of a wheeled mobile robot in
configuration classical tricycle. In: Proceedings of 5th WSEASA International Conference on
Instrumentation, Measurement, Control, Circuits and Systems. (2005)
5. Boussinesq, J.: Aperçu sur la théorie de la bicyclette. J. de Mathématiques Pures et Appliquées.
117–136 (1899)
6. Franke, G., Suhr, W., Rieß, F.: An advanced model of bicycle dynamics. In: Eur. J. Phys. 11)(2),
116 (2000)
7. Kumar, U., Sukavanam, N.: Dynamic modeling and tracking control of a four wheeled
nonholonomic mobile robot. In: Proc. 13th national conference on mechanics and machines
(NaCoMM07). 127–134. (2007),
8. Limebeer, D., Robin, S.: Sharp. Bicycles, motorcycles, and models. In: Contr. Syst. IEEE 26
(5), 34–61 (2006)
9. Maxima. A computer algebra system. version 5.31.0 (2012)http://maxima.sourceforge.net/
10. Meijaard, J.P., Papadopoulos, J.M., Ruina, A., Schwab, A.L.: Linearized dynamics equations
for the balance and steer of a bicycle: a benchmark and review. In: proc. Roy. Soc. A: Math.
Phy. E. Sci. 463(2084), 1955–1982 (2007)
11. Schwab, A.L., Meijaard, J.P., Papadopoulos, J.M.: Benchmark results on the linearized
equations of motion of an uncontrolled bicycle. In: J. Mech. Sci. Techno. 19 292–304 (2005)
12. Tsai, C., Meng-Bi C., Lin, S.: Dynamic modeling and tracking control of a nonholonomic
wheeled mobile manipulator with dual arms. In: J. Intell. Robot. Syst. 47(4) 317–340 (2006)
Optomechatronic Choppers with Rotating
Elements: Design Programs
O. Cira and V.-F. Duma
Abstract Choppers are most used devices for the modulation of light. We have
developed a unitary theory of the optical choppers with rotating wheels working
with top-hat laser beams (with constant intensity distribution). A novel type of
chopper has been introduced, with windows with non-linear edges. The different
modulation functions (of the transmitted light flux) were obtained. From this
analysis, design programs are developed and presented in this study to tackle with
the diverse cases represented by the chopper configurations and by their constructive
and functional parameters. Numerical simulations and modeling are carried out
based on the mathematical theory we have previously developed. Rules of thumb
result to choose the most appropriate type of device and its characteristic parameters
with regard to the particular requirements of a specific application. Top-hat (with
constant intensity), Gaussian, and Bessel laser beams are considered in this study.
Keywords Optomechatronics • Optical devices • Choppers • Laser beams

• Top-hat beams • Gaussian beams • Modulation functions • Analysis
Optomechatronics • Design programs
1 Introduction
Optical choppers [1] are devices utilized for a variety of purposes in optical
systems, including for the generation of light impulses, the elimination or selection
of spectral domains (or of certain wavelengths), or the controlled attenuation
of light. The optical setups in which they are used include thermometers and
pyrometers [10], photometers and colorimeters [1], lasers [3], telescopes or lidars
[11], manufacturing [16], as well as spectral [9, 12] or biomedical systems [13, 14].
Numerous studies, including early ones, were dedicated to the development of
both macroscopic and micro-choppers—the latter built as MEMS (micro-electrical-
mechanical systems) [15]. Nowadays they are considered an established type of
O. Cira • V.-F. Duma ()

Aurel Vlaicu University of Arad, Arad, Romania
e-mail: duma.virgil@osamember.org

458 O. Cira and V.-F. Duma
optical modulator, almost as used in laser setups as optical components like lenses,
prisms, filters, and mirrors. However, there is not yet a systematic theory for them,
which has led us to develop an analysis of “classical” chopper wheels [5]—with
rotating wheels with windows with linear edges. We have also introduced [6],
to the best of our knowledge, a novel type of device, with rotating wheels with
windows with semicircular edges. Due to the way their semicircular edges obscure
progressively the section of the laser beam in the plane of the wheel—which
resembles a planetary eclipse—we have proposed for these devices the name of
“eclipse” choppers [6].
We have presented in previous studies [7, 4] the different types of modulation
functions we had developed for these different choppers [5, 6]. Experimental
modules were also designed and manufactured, as well as prototype wheels [7] with
the different profiles presented above. The next step of this work is to achieve design
programs for the two main types of choppers. A first presentation of this program
for the classical devices was done in [4], for top-hat (with constant intensity over the
entire section) and Gaussian laser beams. In the present chapter this work is further
expanded, but for both types of scanners—classical and eclipse—and for three types
of the most used laser beams: top-hat, Gaussian, and Bessel.
2 Modulation/Transmission Functions of Classical

and Eclipse Choppers
Macroscopic optical choppers (Fig. 1) consist of a rotating wheel with “n” windows
and blades. They obscure the laser beam and generate controlled light impulses of
frequency f D n!/2 , where ! D cst. is the rotational speed of the wheel.
The three types of choppers pointed out in the Introduction are presented in
Fig. 1. The classical chopper (Fig. 1a) is thus compared with the two types of
eclipse choppers: with outward edges (Fig. 1b) and with inward edges (Fig. 1c).
Fig. 1 Optical choppers: (a) classical configuration with windows with linear edges, analyzed
in [5]; eclipse choppers with windows with outward (b) and inward (c) semicircular edges,
introduced and studied in [6]
Optomechatronic Choppers with Rotating Elements: Design Programs 459
One may remark that the classical choppers can be obtained as a particular case
of the eclipse ones for the radius of the edges ¡ ! 1. The modulation functions
we have deduced for the classical devices [5] are therefore a particular case of the
modulation functions we have developed [6] for the eclipse choppers.
The other characteristic parameters of the devices are (Fig. 1) ˛, the window
angle; , the blade angle; r, the radius of the section of the beam in the plane of the
wheel; R, the position of the center of the beam with regard to the pivot O of the
wheel; and 2ı, the angle the beam section is seen from the pivot O.
With regard to the relationships between the angle 2ı and the angles ˛ and ,
different types of modulation/transmission functions (radiometric flux of the output
beams) were obtained analytically and studied in [5, 6]:
(i) Laser impulses with approximate rectangular profiles are obtained for any type
of chopper if the beam is focused in the plane of the wheel.
(ii) Approximate trapezoidal laser impulses were obtained for choppers with
blades larger than the beam section ( 2ı) [5]. This case has also been
approached experimentally and our theory was confirmed by another group [2].
(iii) Approximate sinusoidal profiles of the output signals can also be obtained
for the previous case—if ˛ D 2ı D . In this case both the full transmission
and the full obscuration periods are reduced to zero, and the function of the
transmitted flux only consists of transition periods.
(iv) A non-null modulation function results for a single blade in front of the section
of the beam (for < 2ı and ˛ > 2ı).
(v) Non-null sinusoidal-type profiles are obtained for ˛, < 2ı, i.e., when there
are always multiple blades placed in front of the beam section.
(vi) Approximate trapezoidal or triangular profiles of the output laser beams
can be obtained, as demonstrated in [6], with the eclipse choppers. Thus,
while approximate trapezoidal impulses may also be obtained with classical
choppers—as pointed out above, at (ii)—for classical devices the transition
portions of the graphs of the flux are tangent to the time-axis; this does not
happen for the eclipse choppers (except for particular cases). Therefore, if
the ˛ D 2ı D condition is valid, with eclipse choppers not approximate
sinusoidal but approximate triangular impulses will be obtained—as the
general case of ¡ ¤ 1.
3 Design Programs for Classical Choppers
The analysis achieved for the classical choppers (with windows with linear edges)
[5] has led to the above rules of thumb to choose the appropriate wheels in order to
obtain a desired profile of the laser impulses.
The next steps in our work have thus been (i) to design and manufacture prototype
wheels and modules [7] to carry on experimental validations of the theory developed
and (ii) to develop a design program to obtain the modulation functions for a given
b
a 0.3
A(θ)
0.2
0.1
c
0
1 0 1 2
d 1
Φ(ω.t) 0.5
Φ(θ) 0.5
0
0
0 1 2
0 0.1 0.2 θ[rad/s]
t[s]
Fig. 2 Results of the design program for the classical chopper for a wheel with n D 6 wings, for
r D 5 mm, R D 30 mm, for Gaussian beams, with m D 1: (a) the chopper wheel with the section of
the beam in its plane; (b) graph of the area covered by the wing with regard to the rotation angle
D !t, where ! D cst.; (c, d) graphs of the transmitted flux (normalized, unitless) with regard to
time (c) and to the angle
(d)
set of parameters of the chopper. This program also allows the user to adjust the
constructive and functional parameters of the setup in order to obtain certain profiles
of the laser impulses generated through the chopper wheel.
A brief overview of these possibilities will be done in the following for the types
of chopper wheels presented in Fig. 1 and for different types of laser beams.
To characterize the overlapping of the blades and of the beam section, a
parameter m was defined [4] as the angular width of the beam section (as seen from
the pivot O) and the ratio between the angular width of the window; thus m D 2ı/˛.
In Fig. 2 an example of the information provided by the program is shown, for a
chopper with n D 6 wings, with r D 5 mm and R D 30 mm, working with Gaussian
laser beams. We have considered m D 1 (as an input parameter for the program);
thus it generated (Fig. 2a) a wheel with the margins of the window tangent to the
circular section of the beam. The program gives the A(
) function, which represents
the area of the beam section that is covered by the blades—as a function of time
or, equivalent, as a function of the rotation angle of the wheel,
D !t (the latter in
Fig. 2b). The final result of the program is the function of the transmitted flux, ˆ(t),
Fig. 2c (obtained in this example for ! D 200 rad/s), and ˆ(
), Fig. 2d.
Another example, this time with fractional values of the parameter m (and with
a different number n of blades), is considered in Fig. 3. Only two values of m were
considered, from the row of possible values [4], to make a comparison between
the modulation functions ˆ(
) obtained with a smaller and a larger value of the
parameter m.
A(θ) 0.4
0.3
0.2
0.1
0
(b1)
0 0.5 1 1.5
Φ(θ) 0.5
(a1) 0
(c1)
0 0.5 1 1.5
θ[rad/s]
A(θ)
0.1
(b2)
0.05
0
0 0.5 1 1.5
1
(c2)
Φ(θ) 0.5
0
(a2)
0 0.5 1 1.5
θ[rad/s]
Fig. 3 Classical chopper: applications of the design program developed, for a wheel with
(a) n D 11 and (b) n D 12, for r D 5 mm, R D 30 mm, for Gaussian beams, with (1) m D 11/10,
(2) m D 15/10. For each case considered, the program provides: (a) the chopper wheel with the
section of the beam in its plane; (b) graph of the area covered by the wing; (c) graph of the
transmitted unitless flux—both with regard to the rotation angle
D !t, where ! D cst
In the examples considered in Fig. 3, these graphs are given in (c1) for m D 11/10
and in (c2) for m D 15/10. In both cases m > 1, which means that the beam section
is larger than the window of the wheel. One may remark the irregular shape of the
transmitted flux with regard to the regular shape obtained in Fig. 2, in the case of
m D 1. The irregularities increase, as it can also be seen comparing Fig. 2b1 with
Fig. 2b2 with the increase of the difference of m from m D 1. Also, comparing
the functions in Fig. 3c1, 3c2, the obvious decrease in the overall transmission
coefficient of the device with the increase of m can be concluded.
Another comparison is made in Fig. 4, where we have considered other values
of the chopper parameters, with three different values of m (one on each column)
and three different types of chopper beams (one on each row). The study thus
(a1) (a2) (a3)
1 1
1
(b1) (b2) (b3)
Φ(θ)0.5 Φ(θ) 0.5 Φ(θ)0.5
0 0
0
0.5 1 0 0.5 1 0 0.5 1
0
θ[rad/s] θ[rad/s] θ[rad/s]
1 1 1
(c1) (c2) (c3)
Φ(θ)0.5 Φ(θ) 0.5 Φ(θ)0.5
0 0 0
0 0.5 1 0 0.5 1 0 0.5 1

1 1 1
(d1) (d2) (d3)
Φ(θ) 0.5 Φ(θ)0.5 Φ(θ)0.5
0 0
0
0 0.5 1 0 0.5 1
0 0.5 1
Fig. 4 (a) Designs of the wheel generated by the program. Comparison between the output light
flux for classical choppers used with different types of laser beams: (b) top-hat; (c) Gaussian;
(d) Bessel. The parameters of the setup are: wheels with n D 12 wings, for r D 5 mm, R D 30 mm,
with (1) m D 15/12, (2) m D 21/12; (3) m D 24/12
reveals several aspects: (i) regardless of the type of laser beam (top-hat, Gaussian,
or Bessel), the transmitted flux decreases with the increase of m—as it has been also
concluded from the study in Fig. 3; (ii) the shapes of the laser impulses generated
using the same chopper are different for each type of laser beam (for the same m,
therefore for the same wheel)—although for Gaussian (c) and Bessel (d) beams
there is a certain similarity of these shapes; (iii) the decrease in the transmission
with the increase of m is different for each type of laser beam—it is more severe for
Gaussian (c) with regard to top-hat (b) beams, and again more severe for Bessel (c)
with regard to Gaussian (c) beams.
4 Design Programs for Eclipse Choppers
The new type of wheels that we have proposed [6] has an interesting property that is
also demonstrated in Fig. 5, where an example of applying the program developed
for the different types of wheels is shown: more exact trapezoidal profiles can be
obtained with these eclipse choppers. While in our previous study [6, 7] we have
considered only semicircular edges, outward (Fig. 1b) or inward (Fig. 1c), in order to
obtain analytical expressions of the transmitted flux, using the developed program,
other types of edges, by example elliptical, can also be tested. Thus, in Fig. 5 the
way the program provides for elliptical edges is shown: (a) the shape of a window of
the wheel and (b) the graph of the transmitted flux for top-hat and Gaussian beams
with regard to the input constant flux † D ˆS r2 .
30
∑
Φg-A(τ)
20
B(τ)
(a1) (b1)
10
−20 −10 0 10 20
0 10 20 τ,τ,τ
−10
30
∑
20 Φg-A(τ)
B(τ)
10
(a2) (b2)
0 10 20 −10 0 10
τ,τ,τ
−10
Fig. 5 (a) Window with inward edges and section of the laser beam in the section of the beam—
for eclipse choppers; (b) output flux of the device. Parameters of the chopper: n D 11 windows,
r D 5 mm, R D 30 mm, Gaussian laser beams, with the following semi-axis: (1) a D 4 mm,
b D 3 mm (window with outward edges); (2) a D 4 mm, b D 3 mm (window with inward edges)
5 Conclusions
We presented the program we have developed for the design of optical choppers
with rotating wheels. We approached classical choppers (with linear edges) and
eclipse choppers (i.e., with elliptical edges). The program allows for the design of
the chopper wheel for certain parameters of the setup. It also provides the function
of the transmitted flux for different types of input laser beams: top-hat, Gaussian,
and Bessel. Future work includes other types of choppers and micro-choppers we
are currently developing, as well as applications as attenuators [8] or modulators.
Acknowledgments The work reported here was made possible by a Grant of the Romanian
National Authority for Scientific Research, CNDI–UEFISCDI project PN-II-PT-PCCA-2011-3.2-
1682 (http://3om-group-optomechatronics.ro/).
References
1. Bass, M. (ed.): Handbook of Optics. McGraw-Hill, New York (1995)

2. Benjamin, K., Armitage, A., South, R.: Harmonic errors associated with the use of choppers in
optical experiments. Measurement 39, 764–770 (2006)
3. Benmair, R.M.J., Kagan, J., Kalisky, Y., Noter, Y., Oron, M., Shimony, Y., Yogev, A.: Solar-
pumped Er, Tm, Ho: YAG laser. Opt. Lett. 15, 36–38 (1990)
4. Cira, O., Duma, V.-F.: Transmission functions of optical choppers for Gaussian beam distribu-
tions: Modeling and simulations. Proc. SPIE 8789, 87890E (2013)
5. Duma, V.-F.: Theoretical approach on optical choppers for top-hat light beam distributions.
J. Optic. A: Pure Appl. Optic. 10, 064008 (2008)
6. Duma, V.-F.: Optical choppers with circular-shaped windows: modulation functions. Comm.
Nonlinear Sci. Numer. Simulat. 16, 2218–2224 (2011)
7. Duma, V.-F.: Prototypes and modulation functions of classical and novel configurations of
optical chopper wheels. Lat. Am. J. Solids Struct. 10(1), 5–18 (2013)
8. Duma, V.-F., Nicolov, M.: Neutral density filters with Risley prisms: analysis and design. Appl.
Optic. 48, 2678–2685 (2009)
9. Gondal, M.A., Yamani, Z.H.: Highly sensitive electronically modulated photoacoustic spec-
trometer for ozone detection. Appl. Optic. 46, 7083–7090 (2007)
10. He, Y., Jin, W., Liu, G., Gao, Z., Wang, X., Wang, L.: Modulate chopper technique used in
pyroelectric uncooled focal plane thermal imager. Proc. SPIE 4919, 283–288 (2002)
11. McDermid, I.S., Walsh, T.D., Deslis, A., White, M.L.: Optical systems design for a strato-
spheric Lidar system. Appl. Optic. 34, 6201–6210 (1995)
12. Nag, A., De, A.K., Goswami, D.: Two-photon cross-section measurements using an optical
chopper: z-scan and two-photon fluorescence schemes. J. Phys. B-Atom. Mol. Opt. Phys.
42(6), 065103 (2009)
13. Olkonnen, H.: Chopper stabilized laser-Doppler skin blood velocimeter. Proc. SPIE 1922,
219–224 (1993)
14. Podoleanu, A.G., Dobre, G.M., Cucu, R.G.: Sequential optical coherence tomography and
confocal imaging. Optic. Lett. 29, 364–366 (2004)
15. Toshiyoshi, H., Fujita, H., Ueda, T.: A piezoelectrically operated optical chopper by quartz
micromachining. J. Microelectromech. Syst. 4(1), 3–9 (1995)
16. Wan, D.-P., Liu, H.-B., Wang, Y.-M., Hu, D.-J., Gui, Z.-X.: CO2 laser beam modulating for
surface texturing machining. Optic. Laser Technol. 40, 309–314 (2008)
A Delay Mathematical Model for HIV Dynamics
in HIV-Specific Helper Cells
Carla M.A. Pinto and Ana R.M. Carvalho
Abstract In this paper we study a delay mathematical model for the dynamics
of HIV in HIV-specific CD4C T helper cells. We modify the model presented
by Roy and Wodarz in 2012, where the HIV dynamics is studied, considering a
single CD4C T cell population. Non-specific helper cells are included as alternative
target cell population, to account for macrophages and dendritic cells. In this paper,
we include two types of delay: (1) a latent period for the interval of time for
cells, with contact with the virus, to be infected by the virions, released by them;
(2) a virion production period for the virions to be produced and released to the
bloodstream from the infected cells. We compute the reproduction number of the
model, R0 , and the stability of the disease-free equilibrium. We find that for values
of R0 < 1, the model approaches asymptotically the disease-free equilibrium. We
present numerical simulations of this fact. These results suggest that the model is
mathematically and epidemiologically well posed.
1 Introduction
HIV/AIDS has been a major global health problem of our time. There have been
some breakthroughs, some small conquests, but millions of dollars are spent every
year in the treatment of the disease and no cure is available yet. Approximately, 25
million HIV-infected individuals live in sub-Saharan Africa [1, 12].
HIV is the human immunodeficiency virus that acts in the human body by
destroying the immunological system. The disease has three stages: the acute
stage, characterized by extensive initial virus growth, the chronic or asymptomatic
phase, where virus loads are low and the patient appears healthy, and the AIDS
phase where there is a sharp increase of virus load and the destruction of the CD4C
C.M.A. Pinto ()

School of Engineering, Polytechnic of Porto, Department of Mathematics, Rua Dr António
Bernardino de Almeida, 431, 4200-072 Porto, Portugal
e-mail: cap@isep.ipp.pt
A.R.M. Carvalho
Faculty of Sciences, University of Porto, Rua do Campo Alegre s/n 4440-452 Porto, Portugal
e-mail: up080301038@alunos.fc.up.pt

466 C.M.A. Pinto and A.R.M. Carvalho
T helper cells, leading to the failure of the body. The fact that the HIV debilitates
not only the immune system but the HIV-specific responses [8], and its ability to
rapidly evolve in vivo, escaping the immune responses of the patient [3], makes the
HIV one of the most complex viruses to fight.
Many mathematical models have been proposed for the dynamics of HIV
infection [4, 7, 9–11]. Many models include delays. The most common delays in
the literature are constant delays and continuously distributed delays. Some authors
say that constant delays are not biologically realistic, the continuous delays being
the most adequate. Gamma distributions are frequently used to model continuous
delays [5, 6].
In this paper we study a delay mathematical model for the dynamics of HIV
in HIV-specific CD4C T helper cells. We modify the model presented by Roy
and Wodarz in 2012 [9], where the HIV dynamics is studied, considering a single
CD4C T cell population. Non-specific helper cells are included as alternative target
cell population, to account for macrophages and dendritic cells. In this paper, we
include two types of delay: (1) a latent period for the interval of time for cells, with
contact with the virus, to be infected by the virions, released by them; (2) a virion
production period for the virions to be produced and released to the bloodstream
from the infected cells. We consider that the delays can be approximated by a
gamma distribution [4].
The paper is organized as follows. The model is described in Sect. 2. In Sect. 2.2,
we compute the reproduction numbers and the stability of equilibria. Simulation
results of the full model are presented in Sect. 3. In Sect. 4, we state the main
conclusions of this work and discuss future research directions.
2 Model for HIV/AIDS Transmission
We introduce the model studied in this paper.
2.1 Description of the Model
We denote by x the uninfected specific helper cells and by y the infected specific
helper cells. Variable S models the uninfected non-specific helper cells and I the
infected non-specific helper cells. Variable v is the free virus population.
Disease progression is achieved by cells moving from one class to the other.
The HIV-specific helper cells, x, are produced at a constant rate r after being in
contact with the virus v. The T cell proliferation is proportional to the total virus
population, at low virus load, and saturates for higher values. Saturation of the
production of T cells may also occur for greater values of .x C y/. This regulates
the immune response against the virus. The degree of saturation is given by the
parameters and . Non-infected specific helper cells, x, become infected, after
A Delay Mathematical Model for HIV Dynamics in HIV-Specific Helper Cells 467
contact with the virus, at a rate ˇ and die at a rate d . The infected specific helper
cells, y, die at a rate a, are produced at a rate ˇ and produce virus at a rate k. The
non-specific non-infected target cells, S , are produced at a rate ˛, become infected,
after contact with the virus, at a rate ˇns , and die at a rate dns . Infected non-specific
target cells, I , die at a rate ans , are produced at a rate ˇns , and produce virus at a
rate kns . Finally, the free virus population, v, dies at a rate u.
The system of delay ordinary differential equations for the proposed model is
given by
P/D
x.t rx.t/v.t/.C/.C/
dx.t / ˇx.t /v.t /
.x.t/Cy.t/C/.v.t/C/
R1
P/Dˇ
y.t x.t /v.t /gn1 ;b1 . /e m1 d ay.t /
0
P / D ˛ dns S.t / ˇns S.t /v.t /
S.t
R1
P / D ˇns S.t /v.t /gn ;b . /e m2 d ans I.t /
I.t 2 2
0
R1 R1
v.tP / D y.t /gn3 ;b3 . /e m3 d C ns I.t /gn4 ;b4 . /e m4 d uv.t /
0 0
(1)
mi
The term e .i D 1; 2; 3; 4/ accounts for the cells that are infected at time
but die before becoming productively infected time units later. The distribution
function gni ;bi . / .i D 1; 2; 3; 4/ is defined in [5] as
ni 1
gni ;bi . / D e =bi (2)
.ni 1/Šbini
where the parameters ni and bi define the mean delay, Ni D ni bi , the variance, ni bi2 ,
and the peak, .ni 1/bi .i D 1; 2; 3; 4/, of the distribution. We consider ni 1,
i D 1; 2; 3; 4. If we substitute bi D Ni =ni , i D 1; 2; 3; 4 in equation (2), then we
obtain
nni i ni 1 ni =Ni

gni ;Ni . / D e (3)
.ni 1/ŠNi ni
where Ni 2 =ni is the variance of the distribution, i D 1; 2; 3; 4. The width of the
distribution is set by parameter ni , and the mean delay gives a way to set the location
of the delay.
The inclusion of a continuous delay implies the normalization of the delay kernel
R1
so that fi . /e mi d D 1, i D 1; 2; 3; 4, in order for the equilibria of the
0
non-delay model to remain equilibria of the delay one. In the case of a gamma
distribution fi . / D gni ;bi . /, we rescale the equations of the model to obtain [4]
P/D
x.t rx.t/v.t/.C/.C/
dx.t / ˇx.t /v.t /
.x.t/Cy.t/C/.v.t/C/
R1
P / D ˇN
y.t x.t /v.t /gn1 ;bO1 . /d ay.t /
0
P / D ˛ dns S.t / ˇns S.t /v.t /
S.t (4)
R1
P / D ˇNns S.t /v.t /g O . /d ans I.t /
I.t n 2 ;b 2
0
R1 R1
v.tP / D N y.t /gn3 ;bO3 . /d C N ns I.t /gn4 ;bO4 . /d uv.t /
0 0
where
ˇN D ˇ
.1Cm1 b1 /n1
ˇNns D .1Cmˇns n
2 b2 / 2
N D .1Cm 3 b3 /n3
R1 (5)
N ns D ns
.1Cm4 b4 /n4
bOi D 1Cm
bi
i bi 0 gni ;bOi . /d D 1
2.2 Reproduction Numbers and Stability

of Disease-Free Equilibria
In this subsection, we compute the reproduction number of model (1), R0 , using

the next-generation method [2]. The basic reproduction number is defined as the
number of secondary infections, due to a single infection in a completely susceptible
population. The disease-free equilibrium of model (1) is given by

P 0 D .x0 ; y0 ; S0 ; I0 ; v0 / D 0; 0; d˛ns ; 0; 0 (6)
Using the notation in [2] on system (4), matrices for the new infection terms, F , and
the other terms, V , are given by
2 3 2 3
00 0 a 0 0
F D 40 0 ˇNns S0 5 I V D 4 0 ans 05 (7)
00 0 N N ns u
The associative basic reproduction number is given by
N ns ˇNns ˛
R0 D .F V 1 / D (8)
dns ans u
where indicates the spectral radius of F V 1 . By Theorem 2 [2], we obtain the

following lemma.
Lemma 1. The disease-free equilibrium P 0 is locally asymptotically stable if
R0 < 1 and unstable if R0 > 1.
Proof. We perform the study of the stability of the disease-free equilibrium P 0 of

model (4). The eigenvalues of the jacobian matrix of system (4), around the disease-
free equilibrium P 0 , are computed through the determinant below:
2 3
d 0 0 0 0
6 0 a 7
6 0 0 0 7
6 7
M1 D 6 0 0 dns 0 ˇns S0 7 (9)
6 7
4 0 0 0 ans ˇNns S0 F2 ./5
0 N 3 ./
F 0 N ns F4 ./ u
where
Z 1
Fi ./ D gni ;bOi . /e d (10)
0
The following eigenvalues are easily obtained:
d; a; dns (11)
The remaining eigenvalues are the roots of the characteristic equation of a 2 2

matrix, calculated through the following determinant:

ans ˇNns F2 ./S0

M2 D (12)
N ns F4 ./ u
The determinant is equivalent to
.ans C /.u C / N ns ˇNns S0 F2 ./F4 ./ D 0 (13)
and after some algebra manipulation, we obtain
2 C G1 C G2 D G2 R0 F2 ./F4 ./ (14)
where
G1 D ans C u G2 D ans u (15)
In order for the characteristic equation (13) to have a purely imaginary root,
the eigenvalues must cross the imaginary axis, to become positive. By way of
contradiction, we assume that there is some ! > 0 such that D i ! is an eigenvalue
of (13), that is,
! 2 C G1 i ! C G2 D G2 R0 F2 .i !/F4 .i !/ (16)
Thus,
! 4 C G12 ! 2 C G22 D G22 R02 jF2 .i !/F4 .i !/j2 G22 R02 (17)
Therefore, if R0 < 1, the characteristic equation has no purely imaginary roots. We

conclude that the disease-free equilibrium is stable for R0 < 1.
3 Numerical Results
In this section, we present the numerical simulations of model (1). The parameter
values used in the simulations are ˇ D 0:01, d D 0:5, a D 1, r D 2, D 10,
D 10, D 1, D 1, u D 1, ˇns D 0:2, ˛ D 0:12, dns D 0:1, ans D 0:2, ns D 1,
m1 D 0:01, m2 D 0:01, b1 D 0:0625, b2 D 0:0625, n1 D 4, n2 D 4, m3 D 0:048,
m4 D 0:13, b3 D 0:25, b4 D 0:25, n3 D 5, and n4 D 6.
In Fig. 1, we plot the numbers of uninfected and infected non-specific target cells,
the numbers of uninfected and infected specific helper cells, and the population of
free virus. We observe that, for the given parameter values and initial conditions, the
model approaches the stable disease-free equilibrium asymptotically. Biologically,
this is explained as follows. The low initial virus load and the low initial number of
specific helper cells lead to the extinction of the specific helper cell response.
4 Conclusions
The model presented in this paper is a delay mathematical model for the dynamics
of HIV in HIV-specific CD4C T helper cells. It is a modification of the model
proposed in [9], where the HIV dynamics is studied, considering a single CD4C
T cell population. Non-specific helper cells are included as alternative target cell
population, to account for macrophages and dendritic cells. Two types of delay are
considered: (1) a latent period for the interval of time for cells, with contact with
the virus, to be infected by the virions, released by them; (2) a virion production
period for the virions to be produced and released to the bloodstream from the
infected cells. We compute the reproduction number of the model, R0 , and the
stability of the disease-free equilibrium. We find that for values of R0 < 1, the model
approaches asymptotically the disease-free equilibrium. We present some numerical
simulations. Future work is needed in order to compute the global stability of the
disease-free and of the endemic equilibrium.
Acknowledgements The authors wish to thank Fundação Gulbenkian, through Prémio Gul-
benkian de Apoio à Investigação 2003, and the Polytechnic of Porto, through the PAPRE Programa
de Apoio à Publicação em Revistas Científicas de Elevada Qualidade, for financial support. CP was
partially funded by the European Regional Development Fund through the program COMPETE
1.4 2.5
1.2
Uninfected specific helper cells

2
Non−specific target cells
1.5
0.8
0.6
1
0.4
0.5
0.2
0 0
0 100 200 300 400 500 0 100 200 300 400 500
Time Time
x 10−3
7 1
0.9
6
Infected specific helper cells
0.8
5 0.7
0.6
4
Virus
0.5
3
0.4
2 0.3
0.2
1
0.1
0 0
0 100 200 300 400 500 0 100 200 300 400 500
Time Time
Fig. 1 Uninfected (solid line, top left) and infected non-specific target cells (dashed, top left),
uninfected specific target cells (top right), infected specific helper cells (bottom left), and free virus
population (bottom right). Parameter values used in the simulations are as above (see text), and
initial conditions are x.0/ D 1, y.0/ D 0, S.0/ D 1:2, I.0/ D 0, and v.0/ D 1
and by the Portuguese Government through the FCT - Fundação para a Ciência e a Tecnologia
under the project PEst-C/MAT/UI0144/2013. The research of A.C. was supported by a FCT grant
with reference SFRH/BD/96816/2013.
References
1. CDC: Center for Disease Control (CDC).http://www.cdc.gov/HIV

2. Driessche, P., Watmough, J.: Reproduction numbers and sub-threshold endemic equilibria for
compartmental models of disease transmission. Math. Biosci. 180(1–2), 29–48 (2002)
3. Goulder, P.J., Walker, B.D.: The great escape – AIDS viruses and immune control. Nat. Med.
5, 1233–1235 (1999)
4. Lv, C., Yuan, Z.: Stability analysis of delay differential equation models of HIV-1 therapy for
fighting a virus with another virus. J. Math. Anal. Appl. 352, 672–683 (2009)
5. Mittler, J., Sulzer, B., Neumann A., Perelson A.: Influence of delayed viral production on viral
dynamics in HIV-1 infected patients. Math. Biosci. 152, 143–163 (1998).
6. Nelson, P., Perelson, A.S.: Mathematical analysis of delay differential equation models of HIV-
1 infection. Math. Biosci. 179, 73–94 (2002).
7. Perelson, A.S., Kirschner, D.E., Boer, R.: Dynamics of HIV infection of CD4C T cells. Math.
Biosci. 114, 81–125 (1993).
8. Rosenberg, E.S., Altfeld, M., Poon, S.H., Phillips, M.N., Wilkes, B.M., Eldridge, R.L.,
0
Robbins, G.K., D Aquila, R.T., Goulder, P.J., Walker, B.D.: Immunecontrol of HIV-1 after
early treatment acute infect. Nature 407, 523–526 (2000).
9. Roy, S.M., Wodarz, D.: Infection of HIVspecific CD4 T helper cells and the clonal composition
of the response. J. Theor. Biol. 304, 143–151 (2012).
10. Wodarz, D., Hamer, D.H.: Infection dynamics in HIV-specific CD4 T cells: Does a CD4 T cell
boost benefit the host or the virus? Math. Biosci. 209, 14–29 (2007).
11. Wodarz, D., Thomsen, A.R.: Effect of the CTL proliferation program on virus dynamics. Int.
Immunol. 17(9), 1269–1276 (2005).
12. WHO: Malaria and HIV Interaction and Their Implications for Public Health Police. The World
Health Organization (WHO) (2004).
The Application of Parametric Excitation
in Resonant MEMS Gyroscopes
Barry J. Gallacher, Zhongxu Hu, Kiran M. Harish, Stephen Bowles,

and Harry Grigg
Abstract Parametric excitation, via electrostatic stiffness modulation, can be

exploited in resonant MEMS gyroscopes. In the case of the rate gyroscope
parametric excitation may be used to amplify either the response to angular rate.
In the more complex mode of operation, known as “rate integrating,” the output of
the gyro is angled directly as opposed to angular velocity in the case of rate gyro. In
this rate-integrating mode of operation parametric excitation does offer an effective
energy control used to initiate and sustain the vibration and minimize damping
perturbations. A practical parametric excitation scheme implemented using digital
signal processing has been developed for both the rate and rate-integrating gyro.
Experimental results are presented demonstrating the benefit of using parametric
excitation. By taking advantage of the phase dependence of parametric amplification
and the orthogonality of the Coriolis force and quadrature forcing, the response to
the applied angular velocity may be parametrically amplified by applying excitation
of a particular phase directly to the sensing mode.
1 Introduction
The MEMS gyroscope has received significant attention from MEMS researchers
both in industry and academia [1, 14]. Applications in stability, guidance and
navigation have driven the technological development of state-of-the-art MEMS
gyroscopes to a high level; however, their performance in some key areas does not
approach that of other gyroscope technologies. All MEMS vibratory gyroscopes
operate by exploiting Coriolis coupling between two modes of vibration of a
suspended proof mass [1, 14]. However, due to the small mass of micromachined
gyroscopes, this Coriolis coupling is small. Parametric excitation offers several
approaches to mitigate against the small Coriolis coupling between the two modes
B.J. Gallacher () • Z. Hu • S. Bowles • H. Grigg

Newcastle University, Newcastle upon Tyne, UK
e-mail: b.j.gallacher@ncl.ac.uk; z.x.hu@ncl.ac.uk; s.r.bowles@ncl.ac.uk; harry.grigg@ncl.ac.uk
K.M. Harish
UTAS, Plymouth, UK
e-mail: kiran.shastri@gmail.com

474 B.J. Gallacher et al.
q2
0.3
0.2
0.1
1.0 0.5 0.5 1.0

0.1
0.2 q1
0.3
Fig. 1 Fixed elliptical trajectory of the rate gyro
and thus improve performance. The modes of vibration are often referred to as the
primary and secondary modes. By far the most common mode of operation for
MEMS gyroscopes is where the angular rate is measured. This type of gyro, known
as a “rate” gyro, has applications in the automotive, aerospace and consumer sectors.
The measurement of angle from a rate gyro requires integration of the angular rate
with respect to time. This process has several drawbacks. Noise and bias in the
measured angular rate, when integrated, will lead to a diverging error in the angle
measurement. In the rate gyro, one of the modes of vibration, known as the primary
mode and represented by the generalized coordinate q1 , is forced into resonance and
its amplitude is maintained constant. The measurement of the angular rate can be
shown to be proportional to the amplitude of a second mode, known as the secondary
mode and represented by the generalized coordinate q2 , excited via the Coriolis
force. If the vibration is plotted in mode space [q1, q2 ], then the line of oscillation
of the total vibration will have its gradient determined by the Coriolis force. The
line of oscillation is fixed in the mode space [q1, q2 ] for a given angular velocity
applied about the input axis of the gyroscope. Figure 1 shows the fixed elliptical
trajectory for the rate gyroscope. Note that the minor axis of the ellipse is related to
the structural imperfections and in the ideal case the trajectory becomes a straight
line.
A second way of utilizing the inertia of the modes of vibration provides a direct
measure of angle and hence avoids the issues associated with numerical integration.
This mode of operation known as “rate integrating” has been used successfully
since the mid-1980s in macroscopic hemispherical shell gyroscopes [8, 9]. The
hemispherical resonator gyroscope (HRG) has been developed to a high degree
and is capable of tactical and inertial grade performance. Micromachined “rate-
integrating” gyroscope designs are also discussed in the literature; however, these
have tended to be suitable for angular rates in the order of deg/s. In the “rate-
integrating” gyroscope Coriolis acceleration in a freely vibrating elastic structure
couples two modes of vibration such that energy is continuously transferred between
the two modes of vibration. The resulting beating modal vibrations have a beat
frequency related to the angular rate. If the displacements are plotted in the mode
space [q1, q2 ], as shown in Fig. 2, then the line of oscillation of the total vibration
will precess with an angular velocity related to the angular velocity applied about
The Application of Parametric Excitation in Resonant MEMS Gyroscopes 475
q2
4.x10−7
2.x10−7
−6.×10−7 −4.×10−7 −2.×10−7 2.×10−7 4.×10−7 6.×10−7
Ω
−2.x10−7
q1
−4.x10−7
Fig. 2 Precessing elliptical trajectory of the rate-integrating gyro
gyro’s input axis. The input axis of a gyroscope is the axis about which maximum
Coriolis coupling occurs between the two modes of vibration represented by the
generalized coordinates q1 and q2 . The precession angle will also be related to the
applied angle of rotation. This is in contrast to the rate gyro where the orientation of
the line of oscillation would remain fixed. For the special case where the two modes
share exactly the same natural frequency, the beat frequency and thus the precession
will be proportional to the angular rate and the angle of rotation.
2 Axisymmetric Structures and Modal Degeneracy
The property of wave inertia in elastic axisymmetric structures is well known. For
axisymmetric elastic structures modes of the same circumferential order share the
same natural frequency. This is known as degeneracy. The primary and secondary
modes therefore have the same natural frequency; thus excitation of the secondary
mode due to Coriolis acceleration will be at its natural frequency. Thus the response
is resonant and maximizes the effect of the Coriolis coupling. The theory of
structures exhibiting modal degeneracy is also well established [12].
3 What Parametric Excitation Offers
Due to the small mass of MEMS devices the Coriolis force is very small and
currently the performance of most micro-gyroscopes falls short of “tactical grade”
[14]. This is compounded by the relatively low level of structural precision
compared to their macroscopic counterparts and also “electrical feedthrough” of
the drive. The principal performance measurands for gyroscopes are bias drift,
angle random walk, scale factor and scale-factor stability. Scale factor defines the
ratio of the output response to the applied angular rate. In the case of the rate
gyroscope parametric excitation offers a way of improving the scale factor and
this is linked to the signal-to-noise performance of the gyroscope. This may be
achieved by amplifying either the primary or secondary modes of the gyroscope as
reported in [7, 16]. When applied to the primary mode of the gyroscope, parametric
amplification enables reduced forcing amplitudes and hence reduces electrical
feedthrough due to parasitic capacitances. However, the most beneficial method
employs parametric amplification of the secondary mode. Recall that the secondary
is the response to the Coriolis force. This approach is far more demanding on the
control system but it offers the greatest performance enhancements. Importantly,
parametric amplification of the Coriolis response occurs before any electrical
transduction or amplification. This is particularly advantageous for a range of
sensors, not just MEMS gyroscopes, where electronic noise is the major noise
contributor. In this case the signal-to-noise ratio and scale factor of the gyro will
be amplified. Parametric effects are also reported for other microsystems in [2, 11].
Other resonant devices that could benefit from parametric excitation include the
resonant magnetometer.
For the case of the rate-integrating gyro parametric excitation offers a way of
negating the effect of damping. By using the appropriate cyclic arrangement of
pumping electrodes it is possible to control the vibration energy associated with
each mode as well as controlling the total vibrational energy. This is important since
differences in the modal damping factors manifest themselves as angle or rate errors.
For gyroscopic applications, it is essential that parametric excitation must leave
the orientation of the normal modes unchanged. This imposes a particular cyclic
symmetric condition on the electrode configuration used to parametrically pump
the vibration. A single annular electrode provides a practical way of providing the
parametric excitation. However, use of a single annular electrode does not mitigate
against differences in the direct or cross damping terms.
4 Structural Form for the Gyro and Its Important Modes

of Vibration
Due to its planar structure and relatively simple MEMS fabrication process, the
ring structure is an ideal candidate for championing high-performance gyroscopic
attributes. Figure 3 shows the ring gyro with its lid removed and also a schematic
Fig. 3 The MEMS silicon ring gyro [Goodrich and silicon sensing]
a b
θ θ
45° X 45° X
Fig. 4 Radial displacement of the flexural modes of order n D 2. (a) Primary mode, (b) secondary
mode
highlighting the electrode arrangement for electrostatic actuation and sensing. The
mode shapes and natural frequencies of a perfect axisymmetric ring are well known
and available in the literature [5]. In the single axis ring gyroscope the flexural
modes of vibration in the plane of the ring are sensitive to angular rates applied
about the polar axis of the ring. The input axis of the ring gyroscope is therefore
its polar axis. Of practical interest are the flexural modes of circumferential order
n D 2, as shown in Fig. 4. The dynamics of the ring will be expressed relative to an
axis X passing through the centre of the ring and in its plane, as shown in Fig. 2.
With respect to this axis the modes of interest in the rate gyroscope are referred to
as the primary and secondary modes. The primary mode is characterized by radial
and tangential displacements having spatial variations proportional to cos(n
) and
sin(n
) and its contribution to the motion is described by the generalized coordinate
q1 . The secondary mode is characterized by radial and tangential displacements
Fig. 5 (a) Silicon ring and electrostatic actuation/sensing electrode. (b) Experimental setup
incorporating the packaged gyro, rate table and DSP control electronics
having spatial variations proportional to sin(n

) and cos(n
) and its contribution to
the motion is described by the generalized coordinate q2 .
Angular velocities about the polar axis of the ring result in Coriolis coupling
between the primary and secondary modes. Figure 4 illustrates the radial displace-
ment of the primary and secondary flexural modes of vibration.
Closed loop parametric amplification of the primary mode and open loop
parametric amplification of the secondary mode were demonstrated separately.
However, the two procedures may be implemented simultaneously without any
added difficulty. Figure 5 shows the experimental setup and includes the DSP
control system, the gyro and the rate table. Our work focuses on amplifying the
Coriolis response by directly applying a parametric excitation of particular phase to
the sense/secondary mode. Direct amplification of the Coriolis response results in
significant improvement in the signal-to-noise ratio. A theoretical description of the
parametrically amplified gyroscope is presented in order to gain an insight into the
principal factors pertinent to its exploitation. The excitation and control scheme have
been implemented digitally using a digital signal processing (DSP) development
board enabling very high sampling precision and execution speed necessary for
gyroscopic applications. Parametric excitation in both the rate and rate-integrating
gyro is achieved electrostatically.
5 Basic Gyrodynamics
The homogenous equation of motion of the ring gyroscope in the absence of any
parametric excitation is described by Eq. (1) [3]:
ŒM qR .t / C fŒC C ŒG g qP .t / C ŒK q .t / D 0 (1)

N N N N
where

q1 .t / 0 1 2n
q .t / D ; G D 2mo ŒJ ; ŒJ D ; D ':
N q2 .t / 1 0 n2 C 1
ŒM D mo M ; ŒK D ko K; ŒC D 2o !o mo C
and

1 Eo Izz 2 2"o ad
mo D 0 abd 1 C 2 ; ko D 1 n2 ; ˇ E D
n a 3
2ho 3
M D I C ı; K D I C ; C D I C :
The structural imperfections ı; may be approximated by assuming that the

mass density and elastic modulus vary in a periodic manner circumferentially in the
form
1
X 1
X
.
/ D 0 C mc cos .m
/ C ms sin .m
/
mD1 mD1
1
X 1
X
E .
/ D E0 C Ehc cos .h
/ C Ehs sin .h
/ :
hD1 hD1
It may be easily shown that only harmonics with m D h D 4, 8, 12, 16, etc.
contribute to the kinetic and strain energies. Single crystal silicon of orientation
<111> is used to form the ring structure. This crystal orientation possesses a
threefold rotational symmetry which does not perturb the degeneracy of the n D 2
in-plane flexural modes.
It has been assumed that the damping imperfections have the same structure as
the mass and stiffness imperfections; thus

ı1 ı2 1 2 1 2
ıD ; D ; D ;
ı 2 ı 1 2 1 2 1
2 2
n 1 4c n 1 4s
ı1 D ; ı2 D ;
2 .n C 1/ o
2 2 .n C 1/ o
2
E4c E4s
1 D ; D :
2Eo 2 4Eo
The Coriolis coupling between the two modes of vibration is represented by

the skew symmetric matrix G. For the case of typical vibratory gyroscopes the
imperfections may be adequately represented as perturbations.
6 Rate Gyro Parametric Excitation and Amplification
A brief overview of the linear model of parametric excitation will be presented since
a detailed description is available in [4, 6, 16].
6.1 Parametric Amplification of the Primary Mode
Excitation of the primary mode is achieved by applying the drive voltage Vd (t) to the
pair of primary drive electrodes (PD). The voltage VPD (t) across the PD electrodes is
VPD .t / D Vd .t / Vdc :
Harmonic forcing and parametric excitation is made possible by using a drive

voltage of the form
Vd .t / D VF .t / C VP .t /;
where

VP .t / D Vp sin 2!f t C 'p ;
VF .t / D Vf sin !f t;
and ®p represents the phase advance of the parametric excitation applied to the PD
electrodes with respect to the forcing.
In normal operation jVdc j > Vf , Vp . An approximate expression for the equation
of motion of the primary mode can be obtained by including only the excitation
terms which make a first-order contribution to the response at frequency ! f . The
linearized equation of motion of the primary mode under combined harmonic
forcing and parametric excitation is derived in detail in [4, 6, 7, 16] and has the
form
!1
qR 1 C qP C !1 2 p C 1p sin 2!f t C 'p q1 D 0p sin !f t D D1 .t /
Q1
(2)
with ! 1 as the natural frequency of the primary mode and

1 b̌ 2 1
p D Vp C Vf C Vdc
2 2
2 1 m1

1
1p D 2b̌1 Vp Vdc
m1

1
0p D 2b̌0 Vdc Vf
m1
where
b̌ D "o bd
0
2ho 2
b̌ D "o bd :
1
2ho 3
The steady-state solution of Eq. (2) may be written in the form

q1 .t / D A1 sin !f t C 1 :
By applying the harmonic balance method to Eq. (2), the magnitude of the
steady-state amplitude A1 and phase angle 1 of the primary mode with parametric
excitation are described by
q
&1 2 C 1p &1 cos 'p C b ! 1 4 C 1p b! 1 2 sin 'p C 14 1p 2
A1 Vp ; b! 1 ; 'p D 0p
! 1 4 C &1 2 14 1p 2
b
!
1
& C 12 1p cos 'p
! 1 ; 'p D tan
1 Vp ; b 2 1
b
! 1 C 2 1p sin 'p
(3)
where
! 1 2 D ! 1 2 !f 2
b
! 1 2 D !1 2 p
!1 !f
&1 D : (4)
Q1
It is clear from Eq. (3) that the linear model predicts that
1
!0 (5)
Gai n Vp ; b
! 1 ; 'p
when
1
! 1 4 C &1 2 1p 2 ! 0:
b
4

Here Gain Vp ; b ! 1 ; 'p denotes the parametric gain, which is defined as the
ratio of the mobility magnitude with parametric excitation “on” to the mobility
magnitude with parametric excitation “off” [3]. The DC components p present
in the electrostatic drive lowers the natural frequency of the primary mode to ! 1 .
From Eq. (5) describing the stability boundary of the primary mode it is clear that
the minimum value of 1p causing instability occurs when b ! 1 D 0: This means that
when the primary mode is excited at the softened resonant frequency, the parametric
amplitude needed to cause maximum amplification is a minimum. The excitation
parameters correspond to the nose or apex of the stability boundary.
6.2 Parametric Amplification of the Secondary Mode
Parametric amplification of the secondary mode is achieved by applying the

pumping voltage VP (t) to the pair of secondary drive electrodes (SD). The voltage
bias Vdc is still applied to the ring. The voltage VSD (t) across the SD electrodes is
VSD .t / D VP .t / Vdc
where

VP .t / D Vp sin 2!f t C 's
and ®s represents the phase advance of the parametric excitation applied to the SD
electrodes with respect to the Coriolis force.
The linearized equation of motion of the secondary mode under an applied
angular rate and parametric excitation has the form
!2
qR 2 C qP C !2 2 s C 2s sin 2!f t C 's q2 D F2 qP 1 C F12 q1 (6)
Q2
with ! 2 as the natural frequency of the secondary mode, and

1 1
s D b̌1 Vp 2 C Vdc 2
2 m2

1
2s D 2b̌1 Vp Vdc :
m2
Note that the phase advance ®s is measured with respect to the Coriolis force
F2 qP 1 .
The equation of motion of the secondary mode has two orthogonal forcing
terms: F2 qP 1 and F12 q1 corresponding to the Coriolis force and quadrature force,
respectively. With the primary mode excited at its softened natural frequency ! 1 ,
the steady-state solution of Eq. (6) may be written in the form
q2 .t / D A2 sin .! 1 t C 2/ A12 cos .! 1 t C 2/ :
By applying the harmonic balance method to Eq. (6), the magnitude of the
steady-state amplitudes A2 and A12 and phase angle 2 of the secondary mode with
parametric excitation are described by
q
! 2 4 C 2s b
&2 2 C 2s &2 cos .'s / C b ! 2 2 sin .'s / C 14 1 2
A2 Vp ; b! 2 ; 's D F2
! 2 4 C &2 2 14 2s 2
b
!
&2 C 12 2s cos .'s /
! 2 ; 's D arctan 2 1
2 Vp ; b
b
! 2 C 2 2s sin .'s /
(7)
q
! 2 4 2s b
&2 2 2s &2 cos .'s / C b ! 2 2 sin .'s / C 14 2s 2
! 2 ; 's D F12
A12 Vp ; b
! 2 4 C &2 2 14 2s 2
b
(8)
where
! 1 D !f
!22 D !22 !12

b
! 2 2 D !2 2 s
!2 !f
&2 D :
Q2
The /2 phase shift of the quadrature force relative to the Coriolis force means
that the quadrature force is shifted by relative to VP (t). Also, the DC components
s present in the electrostatic drive lower the natural frequency of the secondary
mode to ! 2 . From Eq. (7) the linear model predicts that
1
!0
Gain Vp ; b
! 2 ; 's
when
1
! 2 4 C &2 2 2s 2 ! 0:
b
4
The minimum value of 2s causing instability in the secondary mode occurs for
b
! 2 D 0: The phase advance ®s may be used to adjust the response amplitudes A2
and A12 . Maximum amplification of the response A2 and maximum deamplification
A2 2 A12 2
of the response A12 occur when dd's
D dd's
D 0 which gives the expression
!
!22
b
's D arctan : (9)
&2
For the ideal case where the mistuning b! 2 D 0, Eq. (9) shows that setting ®s D 0
will result in maximum amplification of the response to the Coriolis force. The phase
advance ®s for this ideal case of zero mistuning is therefore described by ®so with
®so D 0. The response to the quadrature forcing will be deamplified maximally. The
phase angle for the ideal case is described by 2 D 2o D 2 : Equation (9) shows
that when b ! 2 ¤ 0, corresponding to mistuning between the primary and secondary
modes, the phase advance resulting in maximum amplification (deamplification)
100n Vp=10V, Forward sweep

Vp=10V, Reverse sweep
F
Power spectral amplitude 10n
1n
A G B D E
100p
33020 33030 33040 33050 33060 33070
Parametric excitation frequency (Hz)
Fig. 6 Forward and reverse frequency sweeps for parametric excitation only
of the response A2 (A12 ) will be shifted from that corresponding to the tuned case
®so D 0. The phase advance will be described by
!
!22
b
's D 'so C arctan : (10)
&2
As demonstrated in [16], the rate of change of parametric amplification with

respect to the phase advance is substantial. Therefore, maximum amplification of
the response A2 will require the phase advance ®s to be calculated and set accurately.
Measurements of both the mistuning b ! 2 and the damping term −2 will be required.
6.3 Determining the Stability Boundary
The stability boundary can be determined experimentally either from the paramet-
rically amplified forced response or the response to parametric excitation only.
To determine the shape of the instability boundary the excitation frequency ¨
was swept, forwards and backwards, over values centred on 2!. A typical swept
response curve is shown in Fig. 6. For increasing values of frequency, starting from
point A, the response is, at first, indistinguishable from the measurement noise.
The interpretation of this is that the response corresponds to the trivial solution
of the equation of motion. At point B the response suddenly grows and becomes
easily seen above the noise. The interpretation now is that the trivial response has
become unstable and has developed into a limit cycle, with maximum amplitude
occurring at point C. The characteristics of this limit cycle are determined by the
33070
33068
33066
33064
33062
33060
33058
33056
0 1 2 3 4 5 6 7 8 9 10
Fig. 7 Measured stability boundary from frequency sweeps with parametric excitation only.
Parametric excitation frequency versus parametric excitation voltage
nonlinearity of the electrostatic actuation. Further increases in frequency produce

a reduction in the amplitude of the limit cycle until point D is reached. Here the
response is indistinguishable from the noise and corresponds to the trivial state. This
remains the situation as the excitation is swept up to point E and back again to point
D. When the excitation frequency is reduced further the response builds up. The
limit cycle is re-established, and its amplitude retraces the previously held values
between C and D. However, when the backward frequency sweep is continued, the
amplitude does not reduce along the original path from C to B but continues to
increase until a maximum is reached at point F. At F the limit cycle state becomes
unstable and the response thereafter collapses back to the trivial state at point G.
This type of behaviour is discussed in [10, 13, 15] and is typical of a Duffing-
type system (in this case a system with cubic electrostatic softening nonlinearity)
with parametric excitation. This shows that the stability boundary of the linear
parametric system is associated with the frequency range BD. By conducting a series
of such response tests at different values of excitation amplitude V a set of frequency
values corresponding to points B and D were obtained. These are plotted in Fig. 7
as a stability boundary and are compared with the estimated boundary using the
measured softened natural frequency and Q-factor and the measured geometry of
the capacitive actuator.
6.4 Parametric Amplification of the Primary Mode
The parametrically amplified forced response of the primary mode provides an

alternative way of determining the stability boundary. Figure 8 confirms the location
of the nose of the stability boundary using this method.
Fig. 8 1/parametric gain to determine the stability boundary
The objective for the amplitude control tests was to ramp down the harmonic
forcing voltage whilst sustaining the response amplitude at its set point through
using parametric excitation. Figure 9 shows the experimental test results with
the parametric gain set to 12. Ramping down of the harmonic forcing voltage
commenced at 30 s. At this time the parametric controller was switched on. The
response of the parametric controller to step changes in the desired response
amplitude was investigated. Figure 9 illustrates that with the gain set to 12, the
parametric controller can accommodate for 10 % step changes introduced at the
reference amplitude.
6.5 Parametric Amplification of the Coriolis Response
We propose to use parametric excitation to directly amplify the Coriolis response of

the sense mode. It is clear from Eq. (7) of the sense mode that if the frequency
mistuning is zero or tuned small enough, a parametric excitation with a phase
aligned with Coriolis force can be applied to the sense mode to significantly
amplify the Coriolis response. The test device already has a frequency mismatch
of 1.8 Hz. This frequency split is further reduced by electrostatic tuning to 0.1 Hz.
This level of mistuning is small enough to enable a practical demonstration of
parametric amplification of the Coriolis response. The phase advance can be set
to ®s D 0 without any measurable reduction in amplification. Frequency tracking
Forcing & Reponse (volts) Forcing Step changes of desired amplitude

0.1
Response
0.05
0
0 50 100 150 200
Time(sec)
Parametric voltage & Gain
16
15
Parametric voltage Transient process
12 Response gain
9
Parametric excitation start
6
3
0
0 50 100 150 200
Time(sec)
Fig. 9 Parametric gain set to 12. Response to step changes in desired amplitude
of the primary mode was under closed loop control using the PLL. The parametric
gain control circuit was not employed on the secondary mode. Figure 10 shows
the experimental results for the parametrically amplified Coriolis response and
demonstrates a gain by a factor of 11.
6.6 Signal-to-Noise Performance when Under Parametric

Amplification
For high Q-factor MEMS electronic noise dominates over mechanical noise. As a
result parametric amplification of the Coriolis response (the secondary mode of the
gyro) will also amplify the signal-to-noise ratio of the gyro, as described by
S S Nmech
D ; Nmech << Nelect and 0
N Nmech C Nelect S
Gain S
GainSN :
Nelect
a
9
non-parametric
8 Parametric
Rate measurement output (mv)

7
0
0.5 1 1.5 2 2.5 3
Rate table input(degree/second)
b
15
Parametric voltage: 5.5v
Parametric gain
10
0
0.5 1 1.5 2 2.5 3
Rate table input (degree/second)
Fig. 10 Open loop rate table tests. (a) Amplitude of rate output for sinusoidal rate input.
(b) Parametric gain
In order to confirm that mechanical noise contributions are small even when
parametrically amplified, measurements of the noise floor at the resonant frequency
of the gyro were made with parametric excitation and also without parametric
excitation for zero rate input.
Figure 11 shows the noise floor of the rate output with and without parametric
excitation for zero rate input. The parametric excitation was set to 5.5 V, which
corresponds to an amplification of 11 in the Coriolis response. The standard
deviation in the noise response increased from 6.35 to 7.35 V corresponding to
an amplification of 1.16. The improvement in the signal-to-noise ratio is therefore
given by GainSN D Gain/1.16. Since Gain D 11, the signal-to-noise ratio of the gyro
has increased by a factor GainSN D 9.48.
x 10-4
1
Without parametric
With 5.5v Parametric
Noise floor (volts) Standard deviation: 7.35E-6
Standard deviation: 6.35E-6
-1
0 50 100 150 200 250 300
Samples
Fig. 11 Measured noise floor with and without parametric amplification applied
7 Parametric Excitation Applied to the Rate-Integrating

Gyroscope
The dynamics of the rate-integrating gyro with parametric excitation provided by an

annular electrode may be conveniently expressed in terms of the eigenvectors of the
dynamical matrix D. Performing the coordinate transformation
q .t / D ŒX x .t /
N N
ŒX D X .1/ X .2/
N N
where
ŒD D ŒM 1 ŒK
yields
xR .t / C 2" .+o !o Œ ŒI C Œ C ' ŒJ / xP .t /

N N
C!o 2 ŒfŒI C " Œ& g "Û.t / ŒI x .t / D 0:
N
" << 1 and may be used as a small parameter:
The structural imperfections are represented by the matrices

& 0
Œ& D ; Œ D 1 2
0 & 2 1
and
!1 2 D !o 2 .1 C "& / D !o 2 .1 C /
!2 2 D !o 2 .1 "& / D !o 2 .1 /
define the normal mode frequencies.

The method of multiple scales provides an effective method for obtaining a useful
solution.
Thus,

x t ;b
t D x .0/ t ;b
t C "x .1/ t ;b
t C O "2
N N N
where t and b
t are the slow and fast timescales described by

t D "t and b
t D t C O "2 :
Assuming the voltage waveform is periodic and represented by the Fourier series
1
X 1
X
U.t / D uo C ur e ir!t C ur e ir!t with r! D 2b
! o C "ı:
rD1 rD1
gives the zeroth-order and first-order secular equations

x .0/ D Ao t e i!ob
t
C Bo t e i!ob
t
N N N
and
8 9
@Ao ˆ
< .' ŒJ C +o !o ŒŒI C Œ / Ao >
=
N
D N ;
@t :̂ i !2o Œ& Ao C i !2o ˆ uo Ao C ur Bo e i"ıb
t >
;
N N N
8 9
@Bo ˆ
< .' ŒJ C +o !o ŒŒI C Œ / Bo >
=
N
D N :
@t :̂ Ci !2o Œ& Bo i !2o ˆ uo Bo C ur Ao e i"ıb
t >
;
N N N
Assuming a solution of the form

ı ı
Ao t D A.o/ e .SCi 2 /t ; Bo D B .o/ e .Si 2 /t
N N N N
Φ
0
‡
0 5 ew Φu0 10 15 20 25 30
−10
−20 A B
Unstable
−30
Stability boundary
−40
−50
ed (rad/s)
−60
Fig. 12 Stability boundary for the parametrically excited rate-integrating gyroscope
leads to algebraic equation

2 3

A.o/
Œn c ŒI 4 N 5
D0
c ŒI Œm B .o/ N
N
and characteristic equation
a0 s 4 C a1 s 3 C a2 s 2 C a3 s C a4 D 0:
Purely imaginary solution is guaranteed provided
3 D a3 .a1 a2 a0 a3 / a1 2 a4 D 0:
The stability can therefore be plotted and is shown in Fig. 12. The parametric
excitation parameters (ı,˚) resulting in purely imaginary values of s may then be
found by solving the equation 3 D 0. This means that the amplitude and frequency
of the harmonic voltage necessary to make the net energy dissipation of both
normal modes zero are known. The normal mode amplitude envelope can therefore
be sustained parametrically without interfering with the precession caused by the
Coriolis acceleration. Recall that ı/2 is the detuning of the excitation frequency
from the average of the two normal mode frequencies and ˚ is the scaling factor for
the voltage amplitude. Initialization of the vibration can be achieved by choosing
excitation parameters (ı,˚) that reside within the unstable region. Point (B) in
Fig. 12 shows such an arbitrary point with the unstable region. Parametric excitation
will cause both modes to be excited. To avoid unbounded growth a control system
must adjust the excitation parameters such that (ı,˚) converge on the stability
boundary. Converging on point (A) of Fig. 12 is advantageous since it allows the
lowest value of ˚ to be used to stabilize the vibration.
p The square-wave voltage
amplitude corresponding to point (A) has the value 6:5 V.
8 Conclusions
Parametric excitation and amplification has been applied to the MEMS ring
gyroscope. The parametric stability boundary has been mapped; experimental,
parametric amplification approaches of two orders of magnitude have been demon-
strated in open loop. When applied to the Coriolis response of the rate gyro, an
amplification in the signal-to-noise ratio by a factor of 10 has been demonstrated.
Parametric excitation has been shown theoretically to offer a suitable excitation
method for the rate-integrating gyroscope.
Acknowledgments The authors would like to thank Goodrich Sensors and Integrated Systems
(Plymouth UK), the Scottish Microelectronics Centre and the EPSRC for their support in all
the projects that contributed to this work. The authors thank Jim Burdess for many useful and
interesting discussions in relation to this body of work.
References
1. Acar, C., Shkel, A.: MEMS Vibratory Gyroscopes: Structural Approaches to Improve Robust-
ness. MEMS Reference Shelf Series. Springer, New York (2009)
2. Carr, D.W., Evoy, S., Sekaric, L., Craighead, H.G., Parpia, J.M.: Parametric amplification in a
torsional microresonator. Appl. Phys. Lett. 77, 1545–1547 (2000)
3. Gallacher, B.J.: Principles of a micro-rate integrating ring gyroscope. IEEE Trans. Aero.
Electron. Syst. 48, 658–672 (2012)
4. Gallacher, B.J., Burdess, J.S., Harish, K.M.: A control scheme for a MEMS electrostatic
resonant gyroscope excited using combined parametric excitation and harmonic forcing.
J. Micromech. Microeng. 16, 1–12 (2006)
5. Gallacher, B.J., Burdess, J.S., Harris, A.J.: Principles of a three-axis vibrating gyroscope. IEEE
Trans. Aero. Electron. Syst. 37(4), 1333–1343 (2001)
6. Harish, K.M., Gallacher, B.J., Burdess, J.S., Neasham, J.A.: Experimental investigation of
parametric and externally forced motion in resonant MEMS sensors. J. Micromech. Microeng.
19(1), 015021 (2009)
7. Hu, Z.X., Gallacher, B.J., Burdess, J.S., Fell, C.P., Townsend, K.: A parametrically amplified
MEMS rate gyroscope. Sensor Actuat. A Phys. 167(2), 249–260 (2011)
8. Loper, E.J., Lynch, D.D.: The HRG: A new low-noise inertial rotation sensor. In: Sixteenth
Joint Services Data Exchange for Inertial Systems, CA, 16–18 Nov 1982
9. Lynch, D.D.: Hemispherical resonator gyroscope in inertial technology for the future. IEEE
Trans. Aero. Electron. Syst. 20(4), 414–444 (1984)
10. Nayfeh, A.H., Mook, D.T.: Nonlinear Oscillations. Wiley, New York (1970)
11. Rugar, D., Grutter, P.: Mechanical parametric amplification and thermomechanical noise
squeezing. Phys. Rev. Lett. 67, 699–702 (1991)
12. Thomas, D.L.: Dynamics of rotationally periodic structures. Int. J. Numer. Meth. Eng. 14,
81–102 (1979)
13. Thomsen, J.J.: Vibrations and Stability. Springer, Berlin (2003)
14. Yazdi, N., Ayazi, F., Najafi, K.: Micromachined inertial sensors. IEEE Proc. 86(8), 1640–1659
(1998)
15. Zhang, W., Baskaran, R., Turner, K.L.: Effect of cubic nonlinearity on auto-parametrically
amplified resonant MEMS mass sensor. Sensor Actuat. Phys. A 102(1–2), 139–150 (2002)
16. Zhongxu, H., Gallacher, B.J., Harish, K.M., Burdess, J.S.: An experimental study of high gain
parametric amplification in MEMS. Sensor Actuat. A Phys. 162(2), 145–154 (2010)
Influence of Smart Material on the Dynamical
Response of Mechanical Oscillator
Vinícius Piccirillo, Ângelo Marcelo Tusset, José Manoel Balthazar,

Davide Bernardini, and Guiseppe Rega
Abstract The dynamical response of systems with shape memory alloy (SMA)
or magnetorheological damper (MRD) presents a different behavior due to their
nonlinear characteristics. Both systems have a nonlinear behavior due to adaptive
dissipation related to their hysteretic behavior. This property is very attractive
in engineering field. This paper investigates the nonlinear dynamical behavior of
an SMA or MRD oscillator system. The LuGre mathematical model is used to
represent the MRD behavior. On the other hand, the SMA model is based on a
thermomechanical consistent model with four state variables.
Numerical simulations show different aspects about these two systems.
1 Introduction
In recent years, shape memory alloys (SMA) and magnetorheological dampers

(MRD) have attracted considerable attention due to the potential applications as
smart structures in innovative research and technological development. In this way,
SMA and MRD are being used in various dynamical applications associated with the
adaptive dissipation of energy related to their hysteretic behavior. These aspects can
be explored both in the adaptive passive and the active control [8, 10, 12, 13, 15].
It is known that SMA exhibit a high damping capacity. A physical reason for
that is the energy dissipation during the phase transformation (e.g., friction due to
V. Piccirillo () • Â.M. Tusset

Department of Mathematics, UTFPR - Federal Technological University of Paraná, Curitiba,
Brazil
e-mail: piccirillo@utfpr.edu.br; a.m.tusset@gmail.com
J.M. Balthazar
Department of Statistics, Applied Mathematical and Computation, UNESP – São Paulo State
University, Rio Claro, Brazil
e-mail: jmbaltha@rc.unesp.br
D. Bernardini • G. Rega
Dipartimento di Ingegneria Strutturale e Geotecnica, Sapienza Università di Roma, Rome, Italy
e-mail: davide.bernardini@uniroma1.it; giuseppe.rega@uniroma1.it

494 V. Piccirillo et al.
reorientation). The distinct thermomechanical behavior of SMAs is the result of a

transformation from the austenite (parent) phase to martensite (product) phase and
vice versa [8].
The asymptotic nonlinear dynamics of SMA oscillators subject to ideal (har-
monic) or nonideal (DC motor) mechanical excitation has been studied by various
authors employing different constitutive models for the restoring force. For exam-
ple, [3] used polynomial non-hysteretic models while [2, 5–7] used various types of
thermomechanical internal variable hysteretic models.
Magnetorheological (MR) damper is a promising semi-active device in areas of
vibration isolation for suspension systems and civil structures. The viscosity of MR
fluid is controllable depending on input voltage or current. Since, inherently, the MR
damper has uncertain nonlinear hysteresis dynamics, its modeling is an important
issue in the semi-active control of vibration isolation control [11].
Some mathematical models have been introduced in recent work [14] to describe
MR damper behavior, most of them based on friction phenomena descriptions
such as the Bingham and Bouc–Wen models [9]. Although some of these models,
e.g., the modified Bouc–Wen model [14], can accurately predict MR damper’s
dynamics, in most cases, they are complex and difficult to use in control design and
parameter adaptation schemes. In [1] an MR damper model based on LuGre model
is described. This model expresses the dynamic friction characteristics and the
hysteresis effect. Results demonstrate very good model performance and indicate
that the proposed model can be readily used in the design of semi-active control
algorithms. This paper deals with the nonlinear dynamics of an oscillator studied
in two versions where the restoring force is provided by the two kinds of smart
materials. The first one incorporates an SMA whose thermomechanical behavior is
described by the constitutive model proposed in [5]. The second one incorporates
an MRD modeled by means of a modified LuGre dynamical friction model [1]. A
comparison of these two oscillators is made in order to gain some information about
the influence of the two kinds of nonlinearity on the system dynamics.
2 Equation of Motion of the Physical Model
2.1 SMA Oscillator
A shape memory oscillator (SMO) is characterized by a shape memory device

(SMD) that provides a restoring force against the relative displacements of a pair
of points of a main structure [2]. An SMD is composed of an arrangement of shape
memory materials (SMM) that may be designed to yield various kinds of behavior.
In this work the attention is focused on SMDs with pseudoelastic behavior (Fig. 1).
The SMO is considered within a thermomechanical environment character-
ized by a harmonic forcing F(t) D cos(˛t) and a convective heat exchange
Q(ª) D h(ªe ª) where and ˛ are the excitation amplitude and frequency, respec-
tively, ªe is the environment temperature, and h is the coefficient of convective heat
Influence of Smart Material on the Dynamical Response of Mechanical Oscillator 495
Fig. 1 Schematic representation of an SMD model
exchange [5]. At each time t the state of the SMO is described by displacement x(t),
velocity v(t), an internal variable (t) [0, 1] that models the internal state of the
SMD and by temperature ª(t). To model the complex hysteretic behavior of SMM
the state depends not only on the actual value (t) but also on the value 0 (t) of
at the beginning of the last process of change of that occurred before time t. As
discussed in [5] the time evolution of the state takes place according to a system of
five-dimensional differential equations
xP D v (1)
vP D x C .sgn.x// v C F (2)
P D Z Œsgn.x/v JQ (3)
ƒ
#P D ZL C # Œsgn.x/v JQ C Q (4)
J
in which Z and are constitutive functions that can take different expressions
depending on suitable state-dependent criteria [5]. The system response depends on
seven model parameters, J, , q1 , q2 , q3 , L, h, whose physical meaning is discussed
later (see also [5]) as well as on the damping coefficient of the viscous damper
(Fig. 1).
2.2 MRD Oscillator
The friction mechanism is a phenomenon in which two surfaces make contact at

a number of asperities at the microscopic level [4]. In the modified LuGre friction
model [4] [17], this mechanism is expressed by the average behavior of the bristles.
In [1], another MR damper model based on the LuGre model is described as
f D a z C 0 zV C 1 zP C 2 xP C b xV
P (5)
zP D xP 0 a0 jxj
P z (6)
where f is the damping force, V is the input voltage, and the parameters are defined
as follows: z(t) is the internal state variable (m); xP is the velocity of the oscillator
where MRD is attached (m/s); 0 is the stiffness of z(t) influenced by V (N/(m. V));
1 is the damping coefficient of z(t) (Ns/m); 2 is the viscous damping coefficient
(Ns/m); a is the stiffness of z(t) (N/m); b is the viscous damping coefficient
influenced by V (Ns/(m. V)); and a0 is a constant value (V/N).
We now consider a one-degree-of-freedom oscillator where the nonlinear restor-
ing damping force is provided by an MRD. We assume that stiffness is linear.
Balance of linear momentum is expressed by
mx 00 C cx 0 C f C kx D F (7)
where m is the mass, c a coefficient of dissipation, and F(t) D Acos(!t) an external

force. The damping force in MRD can reach approximately 1,500 N when the
voltage level in the MRD is 2.5 V [1]. Due to this strong force, the MRD is
normalized by the value of 400 corresponding to the input voltage V equal to zero.
Thus, the force developed by MRD isn’t bigger than the viscous linear damping. It
is convenient to work with dimensionless position and time, in such a way that Eq.
(7) is rewritten in the following form:
uR C uP C FMRD C u D cos .˛t / (8)
where
c f A !
D !0t , D , FMRD D 2
, D 2
,˛D
m!0 fV 0 mxst !0 mxst !0 !0
and xst means a static displacement of the system, is the dimensionless time,
and fV0 is the dimensional MRD damping force when the input voltage V D 0
(Fig. 2).
3 Numerical Simulation Results
The evolution of an SMO depends on seven parameters, which may be grouped as

follows: mechanical parameters, , q1 , q2 , q3 , that reflect the basic features of the
device (type and arrangement of the material) and determine the basic shape of
the pseudoelastic loop observed in isothermal conditions; thermal parameters, L, h,
that reflect the heat production, absorption, and exchange with the environment and
Fig. 2 Schematic representation of an MRD model
Table 1 SMA parameter values used in the

simulation
q1 q2 L h J
8.125 0.98 1.2 1.017 0.12 3.174
Table 2 MRD parameter values used in the simulation

0 1 2 a b a0
8.0 105 1.6 103 1.5 102 4.0 105 8.0 102 3.0 10 3
therefore determine the temperature variations of the device; and a thermomechan-

ical parameter J that determines the influence of the temperature variations on the
transformation forces [5].
Numerical SMA dimensionless parameters are shown in Table 1 which corre-
spond to the values taken from [5] and the parameter values of the MRD model
used in the simulation are shown in Table 2 which correspond to the values taken
from [1]. The parameters relative to SMA correspond to a typical device with a
pseudoelastic loop with medium–high level of hysteresis. On the other hand, the
MRD model used in this work has a bilinear form with respect to the input voltage
V. The model has the following properties: (a) all parameters are positive because of
the characteristic of the friction model; (b) there is no influence of v on the nonlinear
dynamics of z; and (c) a viscous damping coefficient is proportional to V [9].
Due to nonlinearities, the SMA and MRD oscillator responses are more complex
than the linear oscillator response. This section investigates the numerical dynamics
by establishing a comparison between the two different systems. In order to quantify
the structure associated with maximal amplitude, we construct the parameter space
diagram for ˛ (frequency excitation) versus (amplitude excitation), as shown in
Figs. 3, 4, 5, and 6. To obtain this diagram, we use a grid of 100 100 cells. For each
cell of the grid, we evaluate the largest amplitude response and plot using a color
scale (gray, black, red, green, and blue) as shown on the right side of the diagram.
Each simulation is carried out at fixed initial conditions, that is, u D [0, 0]. It can be
observed that the better performance of MRD depends on voltage applied in damper.
The parameter space plotted for MRD case is shown in Fig. 3 for the voltage applied
40.4767
1.2 36.4288
32.3811 50
Maximum Amplitude
Frequency excitation (α)
1
28.3335 40
24.2858 30
0.8
20.2382 20
16.1906
0.6 10
12.1429
0
8.09528 1.5
0.4
2
1
4.04764 1.5
0.5
0.2 1
0 Frequency excitation (α)
0.5 1 1.5 2 0 0.5 Amplitude excitation (γ)
Amplitude excitation (γ)
Fig. 3 Parameter space plots of the amplitude versus frequency excitation for the MRD oscillator
for V D 0
12.9276
1.2 11.6348 14
10.3421 12
Maximum Amplitude
1
9.0493 10
7.75655 8
0.8
6.46379 6
5.17103 4
0.6
2
3.87827
0
0.4 2.58552 1.5
1.29276 1 2
0.5 1.5
0.2 0 1
Frequency excitation (α) 0
0.5 1 1.5 2 0.5 Amplitude excitation (γ)
for V D 1
value V D 0. It should be noted here that the worst response is for high-frequency
excitation ˛ 1 and high-amplitude excitation 2 [1.3, 2]. Also, in this MRD case,
the space parameter region has several small amplitude vibrations, because the
damping force is significant to vibration reduction under these parameter conditions
(˛ vs ”). In order to show the influence of the voltage applied in MRD for vibration
attenuation of the system, other parameter space plots have been done and shown in
Figs. 4 and 5. It should be noted that the damping effect of the MR damper is very
significant in oscillator system. Figure 4 shows the response of the MRD oscillator
considered V D 1. Note that the amplitude of the response is significantly smaller
than that for the previous one. However, the critical situation occurs for the same
amplitude and frequency. Figure 5 shows that again the maximum amplitude occurs
at high-amplitude and high-frequency excitation, but this peak is smaller than the
previous cases, thus bettering the damping efficiency.
7.56686
1.2 6.81017
6.05349
8
Maximum Amplitude
5.2968
6
4.54011
0.8
3.78343 4
3.02674
0.6 2
2.27006
0
0.4 1.51337 1.5
2
1
0.756686 1.5
0.5
0.2 0 1
0.5 1 1.5 2 Frequency excitation (α) 0 0.5 Amplitude excitation (γ)
for V D 2
34.9198
1.2 31.4278
27.9359
1
24.4439 40
Maximum Amplitude
20.9519 30
0.8
17.4599
20
13.9679
0.6 10
10.4759
0
1.5 2
0.4 6.98396
1 1.5
3.49198
0.5 1
0.2 0 Frequency excitation (α)
0.5 1 1.5 2 0 0.5 Amplitude excitation (γ)
Fig. 6 Parameter space plots of the amplitude versus frequency excitation for the SMO
The parameter space plotted for SMA case is shown in Fig. 6. It is interest-
ing to note that the behavior of the SMA and MRD oscillators is different. It
should be observed here that the worst response is for low-frequency excitation
˛ 0.6 and high amplitude 2 [1.5, 2]. Figure 7 shows the parameter space
from Fig. 6 consideringotherperspectives. Note that qualitatively the SMA tends
to present better performance for high amplitude and frequency, but the MRD
presents smaller amplitude responses for high level of amplitude and low fre-
quency, and for small amplitude and frequency excitation the same response is
present.
Initially, numerical simulations were run to investigate the effects of varying two
parameters: amplitude and frequency excitation. The maximum force developed
during each combination of these parameters is shown in Table 3. The force was
evaluated at the amplitude/frequency range and the maximum force was collected
considering some amplitude response (see first column in Table 3). All of the results
35
40
Maximum Amplitude
30
30
Maximum Amplitude
25
20
20
10
15
0 10
2
5
1.5
0
1 0.5
0.4 0.2 1 1.5
0.8 0.6 1.5 1
0.5 1.2 1 0.5
1.4 2 0
1.6
Fig. 7 Three-dimensional representation of the dependence of the maximum amplitude upon

frequency excitation and amplitude excitation to SMO
listed are for the MR damper case in which 0, 1, and 2 V was applied in device
and SMO. As can be seen, a large variety of nondimensional force responses are
observed in both cases.
4 Conclusions
The purpose of this study was to analyze the influence of external parameters
on the oscillator dynamics, taking into account two different smart materials:
SMA and MRD. It is found that significantly different behaviors can be achieved
depending on amplitude and frequency external parameters applied in mechanical
oscillator. A nonparametric analysis is carried out considering the effects of system
dissipation. It is possible to conclude that the high dissipation capacity of both
materials due to the hysteresis loop is capable to improve the dynamical response
of the oscillator. However, depending on the external parameter combinations, one
material is better than the other. Independent of the voltage applied in MRD, the
comparison between MRD and SMA oscillators has been made from a qualitative
point of view. In cases of low levels of amplitude and frequency or low amplitude
and high frequency the response is similar. In MRD case, the passive damping
produces a damping force and therefore is suited for higher level of amplitude and
low level of frequency. On the other hand, the SMA dissipation is suited to improve
the response when the combination of high levels of amplitude and frequency
occurs.
Table 3 Nondimensional force developed during variation of the parameters ˛ and ˇ—MRD and SMA cases
Maximum
amplitude Maximum force for V D 0 (MRD) Maximum force for V D 1 (MRD) Maximum force for V D 2 (MRD) Maximum force—SMA
4 0.13 1.50 0.90 1.16
8 0.26 1.01 1.72 1.30
12 0.27 1.51 – 3.61
16 0.33 – – 8.13
20 0.44 – – 11.83
24 0.49 – – 15.85
28 0.55 – – 19.79
32 0.63 – – 23.95
36 0.69 – – –
40 0.73 – – –
Influence of Smart Material on the Dynamical Response of Mechanical Oscillator
501
References
1. Awrejcewicz, J., Olejnik, P.: Analysis of dynamic systems with various friction laws. Appl.
Mech. Rev. Trans. ASME 58(6), 389–411 (2005)
2. Bernardini, D., Rega, G.: Thermomechanical modeling, nonlinear dynamics and chaos in shape
memory oscillators. Math. Comput. Model. Dyn. Syst. 11, 291–314 (2005)
3. Bernardini, D., Vestroni, F.: Non-isothermal oscillations of pseudoelastic devices. Int. J. Non-
linear Mech. 38, 1297–1313 (2003)
4. Canudas, C., Olsson, H., Aström, K.J., Lischinsky, P.: A new model for control of systems with
friction. IEEE Trans. Autom. Control 40, 419–42 (1995)
5. Lacarbonara, W., Bernardini, D., Vestroni, F.: Nonlinear thermomechanical oscillations of
shape-memory devices. Int. J. Solids Struct. 41, 1209–1234 (2004)
6. Lagoudas, D.C., Khan, M.M., Mayes, J.J., Henderson, B.K.: Pseudoelastic SMA spring
elements for passive vibration isolation: Part II—simulations and experimental correlations.
J. Intell. Mater. Syst. Struct. 15, 443–470 (2004)
7. Nilkhamhang, I., Mori, T., Sano, A.: Direct and indirect stable adaptive control for suspension
systems with MR Damper. In: Proceedings of the 17th World Congress the International
Federation of Automatic Control, Seoul, Korea, 2008
8. Piccirillo, V., Balthazar, J.M., Pontes Jr., B.R., Felix, J.L.P.: Chaos control of a nonlinear
oscillator with shape memory alloy using an optimal linear control: Part I—ideal energy source.
Nonlinear Dyn. 55, 139–149 (2009)
9. Sakai, C., Ohmori, H., Sano, A.: Modeling of MR damper with hysteresis for adaptive vibration
control. In: Proceedings of the 42nd IEEE Conference on Decision and Control, Maui, Hawaii
USA, 2003
10. Savi, M.A., Paula, A.S., Lagoudas, D.C.: Numerical investigation of an adaptive vibration
absorber using shape memory alloys. J. Intell. Mater. Syst. Struct. 21, 67–80 (2011)
11. Spencer, B.F., Dyke, S.J., Sain, M.K., Carlson, J.D.: Phenomenological model of a magne-
torheological damper. ASCE J. Eng. Mech. 123(3), 230–238 (1996)
12. Tusset, A.M., Balthazar, J.M.: On the chaotic suppression of both ideal and non-ideal duffing
based vibrating systems, using a magneto rheological damper. Differ. Equat. Dyn. Syst. 21,
105–121 (2013)
13. Tusset, A.M., Rafikov, M., Balthazar, J.M.: An intelligent controller design for magnetorheo-
logical damper based on quarter-car model. J. Vib. Control 12, 1907–1920 (2009)
14. Wen, Y.K.: Method for random vibration of hysteretic systems. ASCE J. Eng. Mech.
102(EM2), 249–263 (1976)
15. Williams, K., Chiu, G., Bernhard, R.: Dynamic modeling of a shape memory alloy adaptive
tuned vibration absorber. J. Sound Vib. 280, 211–234 (2005)
Dynamical Pendulum-Like Nonconservative
Systems
Maxim V. Shamolin
Abstract We have elaborated the methods for the qualitative study of dissipative
systems and systems with anti-dissipation that allow us, for example, to obtain
conditions for bifurcation of birth of stable and unstable auto-oscillations. We
succeeded in generalizing the method for studying plane topographical Poincaré
systems to higher dimensions. In three-dimensional rigid body dynamics, we have
discovered complete lists of first integrals of dissipative systems and systems
with anti-dissipation that are transcendental (in the sense of classification of their
singularities) functions that are expressed through elementary functions in a number
of cases. We have discovered new qualitative analogs between the properties of
motion of free bodies in a resisting medium that is fixed at infinity and bodies in an
overrun medium flow.
1 Introduction
We study the nonconservative systems for which the methods for studying, for
example, Hamiltonian systems, is not applicable in general. Therefore, for such
systems, it is necessary, in some sense, to “directly” integrate the main equation of
dynamics. Herewith, we offer more universal interpretation of both obtained cases
and new ones of complete integrability in transcendental functions in two-, three-,
and four-dimensional rigid body dynamics in a nonconservative force field.
The results of the proposed work are a development of the previous studies,
including a certain applied problem from rigid body dynamics [4, 9, 11], where
complete lists of transcendental first integrals expressed through a finite combina-
tion of elementary functions were obtained. Later on, this circumstance allows us to
perform a complete analysis of all phase trajectories and show those their properties
which have a roughness and are preserved for systems of a more general form. The
complete integrability of such systems is related to symmetries of latent type.
M.V. Shamolin ()

Institute of Mechanics, Lomonosov Moscow State University,
1 Michurinskii Ave., 119192, Moscow, Russian Federation
e-mail: shamolin@imec.msu.ru; shamolin@rambler.ru

504 M.V. Shamolin
As is known, the concept of integrability is sufficiently fuzzy in general.

In its construction, it is necessary to take into account the meaning in which it
is understood (we mean a certain criterion with respect to which one makes a
conclusion that the structure of trajectories of the dynamical system considered is
especially simple and “attractive and simple,” in which class of functions we seek
for first integrals, etc. (see also [8]).
In this work, we accept the approach that as the class of functions for first
integrals, takes transcendental functions, and, moreover, elementary ones. Here,
the transcendence is understood not in the sense of elementary function theory
(for example, trigonometric functions), but in the sense of existence of essentially
singular points for them (according to the classification accepted in the theory of
function of one complex variable). In this case, we need to formally continue the
function considered in the complex domain (see also [13]).
2 Methods for Analyzing Zero Mean Variable Dissipation

Dynamical Systems in Spatial Dynamics of a Rigid Body
In this section, we consider the possibilities of extending the results of the plane
dynamics of a rigid body interacting with a medium to the spatial case. We analyze
the problems of a spherical pendulum placed in the overrunning medium flow and a
spatial body motion under the existence of a certain non-integrable constraint and
also show the mechanical and topological analogies of the latter two problems.
2.1 Statement of Problem of Spatial Body Motion in a

Resisting Medium Under Streamline Flow Around
The conjectures presented in [4], which concern the medium properties, are reflected
in constructing the spatial dynamical model of interaction of a medium with a body.
In this connection, there arises the possibility of formalizing the model assumptions
and obtaining the complete system of equations.
The whole interaction of the medium with the axe-symmetric body is concen-
trated on the part of the body surface, which has the form of a circular disk (Fig. 1).
Since the interaction is subjected to the streamline flow around laws, the force S
of this interaction is directed along the normal to the disk, and, moreover, the point
N of application of this force is determined by at least one parameter, the angle of
attack ˛, which is made by the velocity vector v D vD of the disk point D and the
exterior normal at this point (the line CD). The point D is the intersection point of
the middle perpendicular dropped from the center of masses C to the plane of disk.
Therefore, DN D R.˛; : : :/.
Dynamical Pendulum-Like Nonconservative Systems 505
Fig. 1 Spatial model of the dynamical action of the medium on the axe-symmetric body
Assume that the value of the resistance force S has the form S D s1 v2 ; where v
is the module of the velocity vector of the point D and the resistance coefficient s1
(as in the plane case; see [9]) is a function of only the angle of attack ˛: s1 D s1 .˛/.
Therefore, as previously, we consider such an “extension” of the problem, which
was mentioned in [1–4].
As in the plane case, along the line CD, an additional force T can act on the
body; as before, it is called the “following force.” The introduction of this force is
used for ensuring some given classes of motion (in this case, T is the reaction of
possible imposed constraints). In the case where there is no external force T, the
body executes a spatial free drag in the resisting medium.
Denote by Dxyz the coordinate system related to the body (Fig. 1). This
coordinate system related to the point D is chosen such that the tensor of inertia
in this system has the diagonal form: diagfI1 ; I2 ; I3 g. Assume that the mass
distribution is such that the longitudinal principal axis of inertia coincides with the
axis CD (this is the axis Dx), whereas the axes Dy and Dz lie in the plane of disk
and compose the right coordinate system. Moreover, as it was noted, we consider
the case of the dynamically symmetric rigid body, i.e., I2 D I3 .
In this case, to describe the body position in the space, we choose the Cartesian
coordinates .x0 ; y0 ; z0 / of the point D and three angles .
; ; /, which are defined
similar to the navigational angles as follows (compare with [5]).
Let us represent the turn from the inertial system Dx0 y0 z0 to the system Dxyz
as a composition of three turns T3 ./ ı T2 . / ı T1 .
/ under which, first, the frame
.ex0 ; ey0 ; ez0 / is turned around the vector ex0 by the angle
(T1 .
/ is executed):

.ex0 ; ey0 ; ez0 / !.T1 .
// ex0 ; e.1/ .1/
y0 ; ez0 ;
506 M.V. Shamolin
.1/ .1/ .1/

then the frame ex0 ; ey0 ; ez0 is turned around the vector ey0 by the angle ( T2 . /
is executed):
.T2 . .2/ .1/
y0 ; ez0 !
ex0 ; e.1/ .1/ //
ex0 ; ey0 ; ez ;
.2/ .1/
and, finally, the frame ex0 ; ey0 ; ez is turned around the vector ez by the angle
(T3 ./ is executed):
.2/ .1/ .T3 .//
ex0 ; ey0 ; ez ! .ex ; ey ; ez /:
In this case, the vectors having the components in the frame .ex ; ey ; ez / obtain new
coordinates in the basis .ex0 ; ey0 ; ez0 /. In the basis .ex ; ey ; ez /, such a transformation
is given by certain matrix, and then the phase state of the system is characterized by
P P I I
twelve quantities .xP0 ; yP0 ; zP0 ;
I P x0 ; y0 ; z0 ;
; ; /.
Let us consider the spherical coordinates .v; ˛; ˇ/ of the velocity vector v D vD
endpoint of the point D in which the angle ˇ is measured from the axis Dy in the
disk plane up to the line .DN / which is the intersection of two planes, one of which
contains the vector v and the axis Dx and the other is the disk plane.
The latter spherical coordinates and also the components of the angular velocity
are expressed through the phase variables .xP0 ; yP0 ; zP0 ;
I P P I I
P x0 ; y0 ; z0 ;
; ; / via
non-integrable relations. Therefore, the phase state of the system is determined
by the functions .v; ˛; ˇ; ˝x ; ˝y ; ˝z ; x0 ; y0 ; z0 ;
; ; /, and the first six functions
are considered as quasi-velocities of the system. Here, the tuple .˝x ; ˝y ; ˝z / is
defined as
˝ D ˝x ex C ˝y ey C ˝z ez ;
where ˝ is the absolute angular velocity vector of the rigid body.

Since the generalized forces are independent of the body position in the space, the
coordinates .x0 ; y0 ; z0 ;
; ; / are cyclic. This allows us to consider the dynamical
part of the equations of motion as an independent subsystem.
Let us introduce the sign-alternating auxiliary function s.˛/ D s1 .˛/sign cos ˛:
By the theorem on the motion of the center of masses in the space in projections
to the related axes .x; y; z/ and the theorem on the change of the kinematic
moment with respect to these axes, we obtain the following complete system of
differential equations in the dynamical quasi-velocity space R1C fvg S2 f˛; ˇg
R3 f˝x ; ˝y ; ˝z g:
P sin ˛ C ˝y v sin ˛ sin ˇ ˝z v sin ˛ cos ˇ C .˝y2 C ˝z2 / D Fx =m;

vP cos ˛ ˛v
vP sin ˛ cos ˇ C ˛v P sin ˛ sin ˇ C ˝z v cos ˛

P cos ˛ cos ˇ ˇv
˝x v sin ˛ sin ˇ ˝x ˝y ˝P z D 0;

vP sin ˛ sin ˇ C ˛v P sin ˛ cos ˇ C ˝x v sin ˛ cos ˇ

P cos ˛ sin ˇ C ˇv (1)
˝y v cos ˛ ˝x ˝z C ˝Py D 0;
I1 ˝Px D 0;
I2 ˝Py C .I1 I2 /˝x ˝z D zN .˛; ˇ; ˝=v/ s.˛/v2 ;
I2 ˝P z C .I2 I1 /˝x ˝y D yN .˛; ˇ; ˝=v/ s.˛/v2 ;
where Fx D T s.˛/v2 .
The coordinates of the point N in the system .ex ; ey ; ez / take the form

0; yN .˛; ˇ; ˝=v/; zN .˛; ˇ; ˝=v/ ;
yN .˛; ˇ; ˝=v/ D R.˛/ cos ˇ h˝z =v, zN .˛; ˇ; ˝=v/ D R.˛/ sin ˇ C h˝y =v.
We can complement the system (1) by the kinematic relations. As before, we use
the following notation: is the distance DC and m is the mass of the body.
The dynamical system (1) contains the functions R.˛/ and s.˛/. To qualitatively
describe them, we use the existing experimental information about the properties of
the streamline flow around (see [3, 5]).
2.2 Case of Body Motion in a Medium Under Existence

of a Certain Non-integrable Constraint and Beginning
of Qualitative Analysis
As in the plane case, we consider the class of motions under which identity
v D jvj const; (2)
being a non-integrable constraint holds.
2.2.1 On Analytical Integral
In the case of the dynamically symmetric rigid body, system (1) has the analytic first
integral
˝x ˝x0 D const; (3)
i.e., the generalized forces admit a body self-rotation around the longitudinal
dynamical symmetry axis.
508 M.V. Shamolin
2.2.2 On Appearance of an Independent Subsystem
Let us use the methodological tool of reducing the system order, i.e., from the
first equation of (1), let us express the function T so that the total derivative of v
by the system (1) vanishes. As a result of this order reduction, system (1) has an
independent subsystem of the form
P sin ˛ sin ˇ C ˝z v cos ˛ ˝x0 v sin ˛ sin ˇ4

P cos ˛ cos ˇ ˇv
˛v
˝x0 ˝y ˝P z D 0;
˛v P sin ˛ cos ˇ ˝y v cos ˛ C ˝x0 v sin ˛ cos ˇC

C ˝x0 ˝z C ˝Py D 0;
I2 ˝Py D zN .˛; ˇ; ˝=v/ s.˛/v2 ;
I2 ˝P z D yN .˛; ˇ; ˝=v/ s.˛/v2 ;
in which the parameter v is added to constant parameters.
2.2.3 Case of Zero Projection of Angular Velocity on Longitudinal

Axis and Case of Analytic System
Let us consider the trajectories of motion of system (4) on the level of integral (3)
for
˝x0 D 0: (5)
In this case, it takes the form
P sin ˛ sin ˇ C ˝z v cos ˛ ˝P z D 0;

P cos ˛ cos ˇ ˇv
˛v
˛v P sin ˛ cos ˇ ˝y v cos ˛C

C ˝Py D 0;
I2 ˝Py D zN .˛; ˇ; ˝=v/ s.˛/v2 ; I2 ˝P z D yN .˛; ˇ; ˝=v/ s.˛/v2 : (7)
The first two equations (i.e., (6)) reduce to the form

P cos ˛ C v cos ˛ ˝z cos ˇ ˝y sin ˇ C
˛v

C ˝P z cos ˇ C ˝Py sin ˇ D 0; (8)

P sin ˛ v cos ˛ ˝y cos ˇ C ˝z sin ˇ C
ˇv

C ˝Py cos ˇ C ˝P z sin ˇ D 0; (9)
If, for example, S. A. Chaplygin conditions [1]
R.˛/ D A sin ˛; s.˛/ D B cos ˛I A; B > 0; (10)
hold, then, introducing the notations
AB Bh
n20 D ; b D n0 ; H 1 D ; Œb D ŒH1 D 1;
I2 I2 n0
we have the following analytic system:
˛P bn0 v sin ˛ C .1 C bH1 / Œ˝z cos ˇ ˝y sin ˇ D 0; (11)

ˇP sin ˛ .1 C bH1 / cos ˛Œ˝y cos ˇ C ˝z sin ˇ D 0: (12)
Let us complement it by the equations
˝Py D n20 v2 sin ˛ cos ˛ sin ˇ H1 n0 v˝y cos ˛; (13)

˝P z D n20 v2 sin ˛ cos ˛ cos ˇ H1 n0 v˝z cos ˛: (14)
The system (11)–(14) is closed. Note that system (11), (12) is equivalent
to (8), (9) only outside the manifold
O D f.˛; ˇ; ˝y ; ˝z / W cos ˛ D 0g: (15)
Let us project the angular velocity to the movable coordinate system Dz1 z2
(turning the system Dyz by the angle ˇ) such that
z1 D ˝y cos ˇ C ˝z sin ˇ; z2 D ˝z cos ˇ ˝y sin ˇ: (16)
P 2 , system (6), (7) is equivalent to the following system

In this case, since zP1 D ˇz
outside the manifold O 0 :

s.˛/
˛P D 1 C b I2hn0 cos ˛
z2 C b I2vn0 F .˛/
cos ˛
; (17)

v2 s.˛/ hv s.˛/
zP2 D I2
F .˛/ 1 C b I2hn0 cos ˛
z21 cos ˛
sin ˛
z
I2 cos ˛ 2
cos ˛; (18)
510 M.V. Shamolin

s.˛/ s.˛/
zP1 D 1 C b I2hn0 cos ˛
z1 z2 cos ˛
sin ˛
hv z cos ˛;
I2 cos ˛ 1
(19)

ˇP D 1 C b I2hn0 cos s.˛/
˛
z1 cos ˛
sin ˛
; (20)
where
O 0 D f.˛; ˇ; ˝y ; ˝z / W sin ˛ cos ˛ D 0g; F .˛/ D R.˛/s.˛/: (21)
Under S. A. Chaplygin condition (10) and the substitutions
< >7! n0 v <0 >; zk 7! n0 vzk ; k D 1; 2;
the system (17)–(20) takes the form of the analytic system
˛P D .1 C bH1 / z2 C b sin ˛; (22)

zP2 D sin ˛ cos ˛ .1 C bH1 / z21 cos ˛
sin ˛
H1 z2 cos ˛; (23)
zP1 D .1 C bH1 / z1 z2 cos ˛
sin ˛
H1 z1 cos ˛; (24)
ˇP D .1 C bH1 / z1 cos ˛
sin ˛
; (25)
Outside the manifold O 0 , from the dynamical system (17)–(20), the independent
third-order system (17)–(19) is separated. The system (17)–(20) is considered on
the tangle bundle to the two-dimensional sphere (see [13]). Since we have the
“separation” of the third-order system, the phase space of our system has a number
of bundles.
2.2.4 Symmetries of System Vector Field in Quasi-velocity Phase Space
The vector field of system (17)–(19) has the following three types of symmetries:
1. The central symmetry. Near .k; 0; 0/, k 2 Z, such a symmetry arises owing to
that the field in the coordinates .˛; z2 ; z1 / alternates its sign under the change
0 1 0 1
k ˛ k C ˛
@ z2 A ! @ z2 A
z1 z1
2. A certain mirror symmetry (CMS). With respect to the planes ƒi , i 2 Z, where

n o
ƒi D .˛; z2 ; z1 / W ˛ D C i ;
2
such a symmetry arises owing to that the ˛-component of the system vector fields
in the coordinates .˛; z2 ; z1 / is preserved under the change
0 1 0 1
=2 C i ˛ =2 C i C ˛
@ z2 A ! @ z2 A
z1 z1
whereas the z2 - and z1 -components alternate their signs.

3. The symmetry with respect to the plane
f.˛; z2 ; z1 / W z1 D 0gI
precisely, the z2 - and ˛-components of the system vector field are preserved
under the change
0 1 0 1
˛ ˛
@ z2 A ! @ z2 A
z1 z1
whereas the z1 -component alternates its sign.
2.3 On Transcendental Integrability of System
This subsection is devoted to studying the possibilities of complete integration of

the dynamical system considered. Here, we present first integrals of system (22)–
(25) expressed through elementary functions and also discuss the way of integrating
the general system (17)–(20).
2.3.1 Complete List of Invariant Relations
At the beginning we compare the third-order system (22)–(24) to the nonau-

tonomous second-order system
d z2 sin ˛ cos ˛ .1 C bH1 /z21 cos ˛= sin ˛ H1 z2 cos ˛

D ; (26)
d˛ .1 C bH1 /z2 C b sin ˛
d z1 .1 C bH1 /z1 z2 cos ˛= sin ˛ H1 z1 cos ˛
D : (27)
d˛ .1 C bH1 /z2 C b sin ˛
Let us rewrite the system (26), (26) on algebraic form using the substitution
D sin ˛
512 M.V. Shamolin
d z2 .1 C bH1 /z21 = H1 z2 d z1 .1 C bH1 /z1 z2 = H1 z1

D ; D : (28)
d .1 C bH1 /z2 C b d .1 C bH1 /z2 C b
Later on, if we introduce the uniform variables by the formulas zk D uk ; k D

1; 2; we shall reduce the system (28) to the following form:
d u2 .1 C bH1 /.u22 u21 / .b C H1 /u2 C 1

D ; (29)
d .1 C bH1 /u2 C b
d u1 2.1 C bH1 /u1 u2 .b C H1 /u1
D : (30)
d .1 C bH1 /u2 C b
Let us compare the second-order system (29), (30) to the nonautonomous first-
order
d u2 1 .1 C bH1 /.u21 u22 / .b C H1 /u2

D ; (31)
d u1 2.1 C bH1 /u1 u2 .b C H1 /u1
which is reduced uncomplicated to the complete differential:

.1 C bH1 /.u22 C u21 / .b C H1 /u2 C 1
d D 0: (32)
u1
And so, equation (31) has the following first integral:
.1 C bH1 /.u22 C u21 / .b C H1 /u2 C 1

D C1 D const; (33)
u1
which in former variables is looked like
.1 C bH1 /.z22 C z21 / .b C H1 /z2 sin ˛ C sin2 ˛

D C1 D const: (34)
z1 sin ˛
Later on, let us find the evident form of the additional first integral of the third-
order system (22)–(24). At the beginning for this we shall transform the invariant
relation (33) for u1 ¤ 0 as follows:
2 2
b C H1 C1 .b H1 /2 C C12 4
u2 C u1 D : (35)
2.1 C bH1 / 2.1 C bH1 / 4.1 C bH1 /2
As is seen, the parameters of the given invariant relation should satisfy the
condition
.b H1 /2 C C12 4 0; (36)
and the phase space of the system (22)–(24) is stratified on the family of the surfaces
which is assigned by the equality (35).
Thus, by virtue of the relation (33), equation (29) has the form
d u2 2.1 C bH1 /u22 2.b C H1 /u2 C 2 C1 U1 .C1 ; u2 /

D ; (37)
d b .1 C bH1 /u2
where
1
U1 .C1 ; u2 / D fC1 ˙ U2 .C1 ; u2 /g; (38)
2.1 C bH1 /
q
U2 .C1 ; u2 / D C12 4.1 C bH1 /.1 .b C H1 /u2 C .1 C bH1 /u22 /;
herewith, the constant of the integration C1 is chosen from the condition (36).
Therefore, the quadrature for the search of the additional first integral of the
system (22)–(24) has the form
Z
d
D

Z
.b .1 C bH1 /u2 /d u2
D :
2.1 .b C H1 /u2 C .1 C bH1 /u22 / C1 fC1 ˙ U2 .C1 ; u2 /g=.2.1 C bH1 //
(39)
The left-hand side (accurate to the additive constant), obviously, is equal to

ln j sin ˛j: If
b C H1
u2 D w1 ; b12 D .b H1 /2 C C12 4; (40)
2.1 C bH1 /
then the right-hand side of the equality (39) has the form
Z
1 d.b12 4.1 C bH1 /w21 /
q
4 .b12 4.1 C bH1 /w21 / ˙ C1 b12 4.1 C bH1 /w21
Z
d w1
.b H1 /.1 C bH1 / q D
.b12 4.1 C bH1 /w21 / ˙ C1 b12 4.1 C bH1 /w21
ˇq ˇ
ˇ ˇ
ˇ
1 ˇ b1 4.1 C bH1 /w1
2 2 ˇ bH
ˇ 1
D ln ˇ ˙ 1ˇ ˙ I1 ; (41)
2 ˇ C1 ˇ 2
ˇ ˇ
514 M.V. Shamolin
where
Z q
d w3
I1 D q ; w3 D b12 4.1 C bH1 /w21 : (42)
b12 w23 .w3 ˙ C1 /
Three cases are possible for the calculation of the integral (42):
I. jb H1 j > 2:
ˇp q ˇ
ˇ ˇ
ˇ .b H / 2 4 C b 2 w2 ˇ
1 ˇ 1 1 3 C1 ˇ
I1 D p ln ˇ ˙p ˇC
2 .b H1 /2 4 ˇˇ w3 ˙ C1 .b H1 /2 4 ˇˇ
ˇp q ˇ
ˇ ˇ
ˇ .b H1 /2 4 b 2 w2 ˇ
1 ˇ 1 3 C 1 ˇ
C p ln ˇ p ˇ C const:
2 .b H1 /2 4 ˇˇ w3 ˙ C1 .b H1 /2 4 ˇˇ
(43)
II. jb H1 j < 2:
1 ˙C1 w3 C b12
I1 D p arcsin C const: (44)
4 .b H1 /2 b1 .w3 ˙ C1 /
III. jb H1 j D 2:
q
b12 w23
I1 D C const: (45)
C1 .w3 ˙ C1 /
When we return to the variable
z2 b C H1
w1 D ; (46)
sin ˛ 2.1 C bH1 /
we shall have the final form for the value I1 :

I. jb H1 j > 2:
ˇ ˇ
ˇp ˇ
ˇ ˇ
1 ˇ .b H1 / 4 ˙ 2.1 C bH1 /w1
2 C1 ˇ
I1 D p ln ˇ q ˙p ˇC
2 .b H1 / 4 ˇ
2 ˇ .b H1 / 4 ˇˇ
2
b1 4.1 C bH1 /2 w1 ˙C1
2 2
ˇ ˇ
ˇp ˇ
ˇ ˇ
1 ˇ .b H1 / 42.1 C bH1 /w1
2 C1 ˇ
C p ln ˇ q p ˇ Cconst:
2 .b H1 /2 4 ˇˇ b1 4.1 C bH1 / w1 ˙C1
2 2 2 .b H 1 / 2 4ˇ
ˇ
(47)
II. jb H1 j < 2:
q
1 ˙C 1 b1 4.1 C bH1 / w1 C b1
2 2 2 2
I1 D p arcsin q C const: (48)
4 .b H1 /2 b1 . b12 4.1 C bH1 /2 w21 ˙ C1 /
III. jb H1 j D 2:
2.1 C bH1 /w1

I1 D q C const: (49)
C1 . b12 4.1 C bH1 /2 w21 ˙ C1 /
So, the additional first integral was found right before for the third-order
system (22)–(24), i.e., it was presented the complete tuple of the first integrals which
are the transcendental functions of its own phase variables.
It is necessary to substitute formally the left-hand side of the first integral (33)
instead of C1 in the expression of the found first integral.
Then the obtained additional first integral has the following structural form
(which is similar to the transcendental first integral from the plane-parallel dynam-
ics; see [4]):
z2 z1
ln j sin ˛j C G2 sin ˛; ; D C2 D const: (50)
sin ˛ sin ˛
Thus, there are already found two the independent first integrals for the inte-
gration of the fourth-order system (22)–(25). And to complete its integrability it is
sufficient to find one more (additional) first integral which “joins” equation (25).
Since
d u1 u1 .2.1 C bH1 /u2 .b C H1 // dˇ1 .1 C bH1 /u1

D ; D ; (51)
d .b .1 C bH1 /u2 / d .b .1 C bH1 /u2 /
then
d u1 b C H1
D 2u2 : (52)
dˇ1 1 C bH1
It is obvious that for u1 ¤ 0 the following equality is fulfilled:

q
1 2
u2 D .b C H1 / ˙ b1 .2.1 C bH1 /u1 C1 / ;
2
(53)
2.1 C bH1 /
516 M.V. Shamolin
b12 D .b H1 /2 C C12 4;
and then the integration of the following quadrature

Z
d u1
ˇ1 C const D ˙.1 C bH1 / q (54)
b12 .2.1 C bH1 /u1 C1 /2
will bring to the invariant relation
2.1 C bH1 /u1 C1

2.ˇ1 C C3 / D ˙ arcsin q ; C3 D const: (55)
.b H1 /2 C C12 4
In other words, the equality
2.1 C bH1 /u1 C1

sinŒ2.ˇ1 C C3 / D ˙ q (56)
.b H1 /2 C C12 4
is fulfilled and under the transition to the old variables
2.1 C bH1 /z1 C1 sin ˛

sinŒ2.ˇ1 C C3 / D ˙ q : (57)
.b H1 /2 C C12 4 sin ˛
In principle, it makes possible to stop on the latter equality to achieve the

additional invariant relation “joining” equation (25), if we add to this equality that it
is necessary to substitute formally the left-hand side of the first integral (33) instead
of C1 in the latter expression.
But we shall make the certain transformations which is reduced to obtaining the
following evident form of the additional first integral (herewith, the equality (33) is
used):
tg2 Œ2.ˇ1 C C3 / D
..1 C bH1 /u21 .1 C bH1 /u22 C .b C H1 /u2 1/2

D : (58)
u21 .2.1 C bH1 /u2 .b C H1 //2
Returning to the old coordinates, we shall obtain the additional invariant relation
as the form
tg2 Œ2.ˇ1 C C3 / D
..1 C bH1 /z21 .1 C bH1 /z22 C .b C H1 /z2 sin ˛ sin2 ˛/2

D ; (59)
z21 .2.1 C bH1 /z2 .b C H1 / sin ˛/2
or finally
1
ˇ1 ˙
2
.1 C bH1 /z21 .1 C bH1 /z22 C .b C H1 /z2 sin ˛ sin2 ˛

arctg D C3 D const:
z1 .2.1 C bH1 /z2 .b C H1 / sin ˛/
(60)
And so, the system of dynamic equations (1) under S. A. Chaplygin conditions (10)
has five invariant relations in considered case: there exist the analytical non-
integrable constraint (2), the cyclic first integral (3), (5), and the first integral (34),
and also there exists the first integral expressed by the relations (43)–(50) which
is the transcendental function of its phase variables (in sense of complex analysis
also) and expresses in terms of finite combination of the elementary functions, and
finally, there exists the transcendent first integral (60).
Theorem 1. The system (1) under the conditions (10), (2), (5) possesses five
invariant relations (the complete tuple), three of which are the transcendental
functions from the complex analysis view of point. Herewith, all the relations express
in terms of the finite combination of the elementary functions.
2.3.2 On Pendulum with Variable Dissipation
The system (17)–(20) is also a pendulum zero mean variable dissipation system.
The motion is executed under the action of the following two forces: the potential
force
v2
F .˛/
I2
and the linear-in-velocity force
v 0 d F .˛/
b ˛
I2 n0 d˛ cos ˛
with variable coefficient. In a certain subspace, this coefficient has the strictly
positive sign, and, therefore, the energy pumping from the is occurred here. In the
other subspace, this coefficient has the strictly negative sign. Therefore the force
scatters the energy forcing the body to damp its motion.
The latter remarks show that we deal with dissipative zero mean variable
dissipation system. The phase space is decomposed into a union of alternating
domain, in each of which there is a dissipation of only one sign.
518 M.V. Shamolin
Remark. In the case of the system (17)–(20) the search for integrals reduces to
integrating the Riccati equations whose solutions are not expressed in elementary
functions in the most general case.
2.4 Problem on Spatial Pendulum in Overrunning

Medium Flow
Analogously to the plane case [4], let us consider the problem of the spatial
pendulum placed in the overrunning medium flow.
Let a convex plane domain (circular disk, for simplicity) be clamped perpendicu-
larly to the holder OD by a spherical hinge, and let it be in the overrunning medium
flow, which moves with a constant velocity v1 ¤ 0. Assume that the holder does
not create any resistance (Fig. 2).
The total force S of the medium flow action on the body is directed parallel to
the holder, and the point N of application of this force is determined by only one
parameter, the angle of attack ˛ measured between the velocity vector vD of the
point D with respect to the flow and the holder. Therefore, the force S is directed
along the normal to the side opposite to the direction of the velocity vD and passes
through a certain point N of the plane domain displaced from the point D forward
with respect to the direction of vD . Such conditions arise in using the model of
streamline flow around spatial bodies [4].
y0 S
N
x
D
v0
Ψ
v∞
O θ
x0
z0
Fig. 2 Spherical pendulum in the homogeneous overrunning medium flow

The vector e determines the orientation of the holder. Then S D s1 .˛/v2D e; where
the resistance coefficient has the form s1 D s1 .˛/ D s.˛/sign cos ˛: Let Ox0 y0 z0 be
the immovable coordinate system. The direction of the overrunning flow coincides
with the direction of the axis x0 . Let us relate the coordinate system Dxyz with
the body, where the axis Dx is directed along the holder and axes Dy and Dz are
rigidly related to the plane domain.
The coordinates of the point N in the system Dxyz have the form .0; yN ; zN /.
Analogously to the problem of the free body motion, we introduce the function
R.˛/ and also the angle ˇ measured in the plane Dyz. In thus case for simplicity,
let S. A. Chaplygin properties (10) hold. For any admissible function R.˛/, the
analysis is performed analogously.
If the body is dynamically symmetric (I1 and I2 D I3 are principal moments of
inertia in the system Dxyz) and .˝x ; ˝y , and ˝z / are projections of the angular
velocity in the system Dxyz, then the equations of motion take the following form,
which is analogous to (13), (14) (H1 D 0, for simplicity):
˝y0 D n20 v2D sin ˛ cos ˛ sin ˇ; ˝z0 D n20 v2D sin ˛ cos ˛ cos ˇ: (61)
The resistance force admits the existence of the first integral (3), and, in this case,
the condition ˝x0 D 0 is taken into account in equations (61).
Let us introduce the angles .
; / determining the orientation of the pendulum
(Fig. 2). The angle
is measured from the axis x0 to the holder, whereas is
measured from the projection of the holder on the plane Oy0 z0 to the axis y0 . Then
cos
D cos cos ; sin
cos D cos sin ; sin
sin D sin : (62)
2.4.1 Complete System of Equations
The relations connecting .vD ; ˛; ˇ/ and .

; ; ˝y ; ˝z / (l is the holder length) have
the form
vD cos ˛ D v1 cos

;
vD sin ˛ cos ˇ D l˝z C v1 sin
cos ; (63)
vD sin ˛ sin ˇ D l˝y v1 sin
sin :
Later on, we have

P D ˝y cos
sin
;
P D ˝z C ˝y sin cos
sin
; (64)
P D ˝y cos ;
520 M.V. Shamolin
whence we easily deduce that
P sin sin
˝y D ; ˝z D P P : (65)
cos cos cos
Using properties (62) and (65), we have the identities
sin
sin
˝y D
P sin C P cos ; ˝z D
P cos P sin : (66)
cos
cos
Equations from (61), (63), and (66) compose a complete system for determining
the pendulum motion on the level of the integral ˝x0 D 0.
2.4.2 Systems of Differential Equations and Topological Analogy
Starting from three groups of equations two of which are differential and the third
is algebraic, it is easy to prove the following proposition.
Theorem 2. The complete system of the pendulum motion on the tangent bundle to
the two-dimensional sphere has the following form:

R C ln20 v1
P cos
C n20 v21 sin
cos
P 2 cos
sin
D 0; (67)

R C
P P 1Ccos2
C ln2 v1 P cos
D 0: (68)
cos
sin
0
As in the case of a free body, system (67), (68) has symmetries. It also has a
complete tuple of first integrals, and the angle is a cyclic coordinate.
Theorem 3. System (67), (68) is topologically equivalent to the system (17)–(20).
Therefore, as in the plane case, there is a mechanical analogy between the pendulum
in the medium flow and the free body under the presence of a certain non-integrable
constraint.
Remark. The angle ˛ for a free body is equivalent to the angle
, whereas the angle
ˇ is equivalent to the angle . Moreover, for systems (22)–(25) and (67), (68) to be
identical, it is necessary to set l D and v1 D v. The constant velocity of the
characteristic point of the circular disk for a free body corresponds to the constant
velocity of the overrunning flow on the pendulum. The relation l D tells us that
for a free body, the stationary motion ˛ 0 is exponentially unstable, and for the
pendulum, the stationary motion
0 is exponentially stable.
2.5 Topological Structure of the Phase Portrait

of the System Studied
In this subsection, we present a scheme of global qualitative analysis of the

dynamical system (17)–(19) on the whole phase space f˛; z2 ; z1 g. For any function
F , the phase portrait of system (17)–(19) has the same topological type.
System (17)–(19) has no trajectories having infinitely distant points of the phase
space as its ˛- and !-limit sets. Moreover, the system has no simple and complicated
limit cycles (see also [10]).
2.5.1 Reducing System to the Form Studied
Let us consider the case of the absence of an additional dependence of the functions
yN ; zN on angular velocity (i.e., for simplicity, H1 D 0). For convenience of
drawing the three-dimensional phase portrait and preserving the right-oriented
coordinate system f˛; z12 ; z1 g, let us make the formal change ˛ ! ˛. Moreover,
because of the existence of symmetries, we study the domain
f.˛; z2 ; z1 / W < ˛ < 0; z1 > 0g: (69)
In this case, the system takes the form
˛P D z2 C b I2vn0 F .˛/
cos ˛
;
2
zP2 D Iv2 F .˛/ C z21 cos ˛
sin ˛
; (70)
zP1 D z1 z2 cos ˛
sin ˛
;
and under S. A. Chaplygin condition (10), the analytic system has the form
˛P D z2 C b sin ˛;
zP2 D sin ˛ cos ˛ C z21 cos ˛
sin ˛
; (71)
zP1 D z1 z2 cos ˛
sin ˛
:
For simplicity, let us study the system (71), and let us “rectify” the field along the
cylinders f.˛; z2 ; z1 / W z2 C b sin ˛ D 0gI precisely, making the change of the phase
variables u D z2 C b sin ˛; we pass from system (71) to the system
˛P D u;
uP D sin ˛ cos ˛ C z21 cos ˛
sin ˛
C bu cos ˛; (72)
zP1 D z1 Œu b sin ˛ cos ˛
sin ˛
:
522 M.V. Shamolin
Fig. 3 Phase portrait of the system (72) for z1 0 and ˝ ! u formally
For b D 0, system (72) (denoted by (720 )) has two analytical integrals. The
following but important proposition is obvious.
Theorem 4. The plane
f.˛; u; z1 / W z1 D 0g (73)
is integral for the system (70).

Theorem 5. Plane (73) “contains” the portrait of the system from the plane
dynamics (see Fig. 3 if we extend the system field to the lines f.˛; u/ W si n˛ D 0g by
continuity).
Let us introduce the family of (three-dimensional) layers
….˛1 ;˛2 / D f.˛; x1 ; x2 / 2 R3 W ˛1 < ˛ < ˛2 g: (74)
2.5.2 Conservative Third-Order Comparison System
We have already presented many assertions concerning Poincaré topographical

systems (PTS) and more general comparison systems on two-dimensional manifolds
[6,7]. To study third-order systems, we need PTS and comparison systems of higher
order. We do not dwell on the general theory of PTS and comparison systems of
higher dimension and restrict ourselves to its application to the system studied.
Since system (720 ) has two analytical integrals
ˆ1 D z21 C u2 C sin2 ˛ D C10 ; ˆ2 D z1 sin ˛ D C20 ; C10 ; C20 D constI (75)

the latter fiber the phase space at each point of which we can draw two surfaces
given by relations (75) that intersect along the phase characteristic of system (720 ).
For each point of the phase space of system (72), let us define two pairs of subspaces
in each of which the characteristic of system (72) enters or emanates from. The first
integrals (75) “help us” to study the behavior of phase trajectories of system (72)
(compare with [14, 15]).
If g1 D fsin ˛ cos ˛; u; z1 g, g2 D fz1 cos ˛; 0; sin ˛g are the gradients of
surfaces ˆ1 ; ˆ2 , then the inner products
%k D .gk ; vN /; k D 1; 2
(Nv is the vector field of system (72)) have all the properties of the characteristic
functions (see also [12]).
Theorem 6. The characteristic functions %k have the form
%1 .˛; u; z1 / D b cos ˛Œu2 C z21 ; %2 .˛; z1 / D bz1 sin ˛ cos ˛:
2.5.3 Equilibrium Points of System Studied
The system (72) under study has the following equilibrium states:
1. The repelling point .0; 0; 0/
2. The attracting point .; 0; 0/
3. The saddles .=2; 0; Q1 / and Q1 2 Œ0; n0 / in each small area parallel to the
plane O˛z2
4. The centers .=2; 0; Q2 / and Q2 2 .n0 ; C1/ in each small area parallel to the
plane O˛z2
In cases 3 and 4, the singular points are not isolated.
Through the whole section, we have performed the qualitative analysis in
sufficient detail. The next subsection is a consequence of the previous material.
2.5.4 Phase Portrait Structure
Theorem 6 implies the following assertions:

1. Point 1 is the ˛-limit set of the separatrices entering points 3 in the layer
….=2;0/ .
2. Point 2 is the !-limit set of the separatrices emanating from points 3 in the strip
….;=2/ .
3. The !-limit (˛-limit) sets of the separatrices emanating from (entering) points 3
in the layer ….=2;0/ (in the layer ….;=2/ ) are the same points.
4. The part of the phase space containing points 4 entirely filled with closed
trajectories.
524 M.V. Shamolin
2.6 Trajectories of Spherical Pendulum Motion and Case

of Its Nonzero Twist Near Longitudinal Axis
2.6.1 Pendulum Trajectories on Sphere
In accordance with the properties of the phase space partition into trajectories, the
typical trajectories of the point D of the plane domain fall into classes:
1. The trajectories corresponding to the oscillatory domain. Such trajectories are
curves on the sphere that unboundedly approach the poles of the sphere (along
the flow) as t ! ˙1.
2. The trajectories corresponding to the rotational domain. Such trajectories are
curves on the sphere that almost always fill annulus-like domains on the sphere
symmetric with respect to the equator.
2.6.2 Spherical Pendulum Under Nonzero Proper Twist
We immediately present the equations of the pendulum motion under the condition
˝x0 ¤ 0. These equations have the following form:

R C ln20 v1
P cos
C n20 v21 sin
cos
P 2 cos
sin
II12 ˝x0 P cos

sin
D 0; (76)

R C
P P 1Ccos2
C ln2 v1 P cos
C I1 ˝x0
P cos
D 0: (77)
cos
sin
0 I2 sin
Now let us immediately pass to the classification of possible pendulum trajecto-

ries on the sphere:
1. Trajectories analogous to trajectories 1 for the case ˝x0 D 0. The asymptotics of
behavior of such trajectories is the same as above.
2. The trajectories analogous to trajectories 2 for the case ˝x0 D 0. Such
trajectories are everywhere dense on the whole sphere [15].
Acknowledgements This work was supported by Russian Fund of Basic Research, project 12-
01-00020-a.
References
1. Chaplygin, S.A.: On motion of heavy bodies in an incompressible fluid. In: A Complete

Collection of Works [in Russian]. Vol. 1, Izd. Akad. Nauk SSSR, Leningrad 133–135 (1933)
2. Chaplygin, S.A.: Selected Works [in Russian]. Nauka, Moscow (1976)
3. Gurevich, M.I.: Jet Theory of Ideal Fluid [in Russian]. Nauka, Moscow (1979)
4. Samsonov, V.A., Shamolin, M.V.: Body motion in a resisting medium. Moscow Univ. Mech.
Bull. 44(3), 16–20 (1989)
5. Sedov, L.I.: Continuous Medium Mechanics [in Russian]. vol. 1, 2. Nauka, Moscow (1983,
1984)
6. Shamolin, M.V.: Applications of Poincaré topographical system methods and comparison
systems in some concrete systems of differential equations. Vestn. MGU, Ser. 1, Mat., Mekh.
2, 66–70 (1993)
7. Shamolin, M.V.: Spatial Poincaré topographical systems and comparison systems. Usp. Mat.
Nauk. 52(3), 177–178 (1997)
8. Shamolin, M.V.: On integrability in transcendental functions. Usp. Mat.Nauk. 53(3), 209–210
(1998)
9. Shamolin, M.V.: New Jacobi integrable cases in dynamics of a rigid body interacting with a
medium. Dokl. Ross. Akad. Nauk. 364(5), 627–629 (1999)
10. Shamolin, M.V.: A new family of phase portraits in spatial dynamics of a rigid body interacting
with a medium. Dokl. Ross. Akad. Nauk. 371(4), 480–483 (2000)
11. Shamolin, M.V.: New integrable cases and families of portraits in the plane and spatial
dynamics of a rigid body interacting with a medium. J. Math. Sci. 114(1), 919–975 (2003)
12. Shamolin, M.V.: Comparison of Jacobi integrable cases of plane and spatial body motions in a
medium under streamline flow around. Prikl. Mat. Mekh. 69(6), 1003–1010 (2005)
13. Shamolin, M.V.: Integrability of Some Classes of Dynamic Systems in Terms of Elementary
Functions. Vestn MGU, Set. 1, Mat., Mekh. 3, 43–49 (2008)
14. Shamolin, M.V.: On integrability in elementary functions of certain classes of nonconservative
dynamical systems. Contemporary Mathematics and Its Applications. Geometry Mech. 62,
131–171 (2009)
15. Trofimov, V.V., Shamolin, M.V.: Geometrical and dynamical invariants of integrable Hamilto-
nian and dissipative systems. Fund. Prikl. Mat. 16(4), 3–229 (2010)
Index
A boundary shape perturbation, 91–92

Acceleration Bubnov–Galerin, 33, 50
curves, 108 Floquet–Bloch, 96–99
gravitational, 385 Galois, 307
gravity, 61, 70, 124, 214 lubrication theory, 91–94
Accelerometer, 372 Newton’s, 369
Actuation, 198 Approximation
electrostatic, 478, 485, 4477 neutral network, 71
Airflow, 373, 417 subsequent, 365, 366, 368
Airfoil, 417, 418, 424 Area, 11, 17, 21, 89, 99, 107, 109, 128, 129,
Algorithm 152, 180, 249, 250, 253, 283, 374,
FASTSIM, 262 392, 417, 420, 425, 433, 434, 460,
Newton–Leipnik, 141 461, 473, 494, 523
Amplification wing, 417
closed-loop, 478 Attractor
parametric, 476, 478, 480–489, 492 chaotic, 82, 203, 228, 384, 387, 388,
Amplitude, 1–10, 20, 21, 53, 55, 61, 62, 392–394
66, 72, 78, 95, 97, 104, 107, 108, rare, 206, 208–210, 337
110–112, 168, 170–172, 182, 183, stable, 390
227, 262, 263, 265–267, 298–302, typical, 399
333, 383, 393, 411, 416, 421–423, Axes/axis, 25, 62, 108, 178, 226, 227,
425, 426, 436, 439, 474, 476, 271–276, 296, 306, 309–311, 313,
481–488, 491, 494, 497–501 339, 342, 385, 394, 415–417,
modulation, 169–172 423, 424, 450, 452, 459, 463,
Analysis 469, 474, 475, 477, 478, 505–508,
Fourier, 50, 55–59, 61 519, 524
spectral, 109, 110, 371
stability, 232, 402 B
wavelet, 50, 53, 55–59, 61, 109 Beam
Angle Bessel, 458, 462, 464
heading, 449, 450 Gaussian, 458, 460–464
of attack, 416–418, 424, 504, 505, 518 laser, 458–460, 462–464
roll, 449, 450, 452 Behavior
Anisotropy, 96, 109 asymptotic, 247
Approach chaotic, 104, 146, 245–253

in Mathematics & Statistics 93, DOI 10.1007/978-3-319-08266-0
528 Index
Behavior (cont.) Characteristic, 15, 24, 55, 57, 61–63, 66,

complex, 77, 83, 108 68, 69, 72, 73, 104, 115, 125,
non-linear, 162 129, 168, 191, 208, 226, 227,
quasi-linear, 162 247, 256, 259, 260, 297, 344–347,
quasi-regular, 113 373, 375, 376, 402, 416, 418–424,
stochastic, 342 459, 469, 470, 484, 491, 494, 497,
turbulent, 50, 51, 62–69, 75 520, 523
undesired, 77, 80 Coulomb’s, 259
Belt, 408 Chopper
driving, 407 eclipse, 458–459, 463, 464
Bifurcation optical, 457, 458, 464
closed-loop, 440 Coefficient
diagram attenuation, 95
stable branch, 321, 436 binomial, 140
two-dimensional, 81 damping, 27, 52, 56, 61, 65, 133, 134,
unstable branch, 204, 207, 321 179, 195, 385, 418, 424,
map, 205–206, 210 495–497
non-smooth, 204, 208–210 friction, 283
path, 206 Poisson’s, 52, 53, 56, 61, 65
pattern, 203, 204, 210 stiffness, 60, 129, 133, 162
period-doubling, 206, 209 transmission, 461
point, 205, 208 Composite
reverse, 437 theory, 93
scenario, 79, 80 viscoelastic, 95
smooth, 204, 206 Condensation, 337
smooth saddle-node, 209 Bose–Einstein, 337
transcritical, 320, 326, 329, 330 Condition
Bisectrix, 32, 33 boundary, 27, 29, 31, 51, 52, 61–69, 71, 74,
Blade, 458–460 89, 90, 153, 155, 237, 287, 363–365,
rotor, 175 367, 369, 491
Body Chaplygin, 509, 510, 517, 521
rigid, 11–22, 181, 251, 503–524 constraint, 11
sway, 103, 108, 109, 111, 113 initial, 1, 6, 8, 9, 19, 27, 30, 52, 56, 61, 65,
Bone 71, 141, 143, 145, 164, 165, 167,
cancellous, 86, 99 189, 215–218, 222, 228, 230, 234,
trabe, 86 238, 248, 251, 262, 287, 302, 322,
Border, 153, 250, 262, 434 323, 366, 368, 385, 412, 413, 436,
stability, 436, 437 439, 470, 471, 497
Buckling, 24, 25 homogeneous, 164
jump-type, 32 perfect bonding, 88
matching, 31, 154, 155
C no-slip, 443–449, 454
Cell sufficient, 357, 358, 360, 419
dendritic, 466, 470 Constraint
helper, 465–471 holonomic, 187, 273, 278, 279, 305
periodic, 391, 392, 394 kinematic, 270–272, 276
sperm, 295, 296, 301, 302 nonholonomic, 273, 443–454
transient, 392 reaction, 116
Chaos servo, 185–200
monitoring, 50 Construction, 23–25, 40–41, 205–207, 209,
spatial, 51, 59, 62, 67, 74, 75 210, 247–249, 289, 338, 353, 354,
temporal, 51, 59, 62, 67, 74, 75 504
Index 529
Control viscoelastic, 86, 87, 96, 97, 99

adaptive, 38, 39, 47 viscous, 162, 495, 496
ramp, 78–83 Definition
amplitude, 486 Grünwald–Letnikov, 140
digital, 384 Leibniz, 140
energy, 225, 235–242, 250 Riemman–Liouville, 140
input, 185, 463 Deflection, 23–25, 32, 33, 52, 56, 60–63,
passive, 430, 438, 439, 441 65–67, 74, 109, 179, 260
torque, 189, 384 Degeneracy, 246, 475, 479
tracking, 213–243 Degree
voltage-mode, 78, 205 of freedom (DOF), 116, 169, 177, 180,
Controller 235–241
design, 82 of saturation, 466
Lipschitz, 218, 224 Delay, 70, 108, 149, 319, 329, 333, 371–381,
continuous, 218, 224 383, 465–471
optimal Lyapunov-stable, 242 continuously distributed, 466
Convergence, 53, 62, 222, 232, 234, 243, 286, Delta, 366
288–290, 292, 367–369 Kronecker, 366
exponential, 232 Denominator, 18, 227, 328, 332, 333
Converter Density
power, 77, 78, 203–210 air, 417
PWM controlled, 77–79, 83 mass, 87, 91, 287, 479
Coordinates power spectral (PSD), 110, 111
curvilinear, 25, 62 Derivative, 40, 58, 88, 119, 126, 139–149,
dimensionless, 339, 343 166, 188, 194, 197–199, 232, 233,
generalized, 3, 16, 122, 124, 133, 162, 172, 339, 366, 367, 418, 508
179, 309, 313, 452, 474, 475, 477, normal, 88, 89
478 Diagram, 79, 81, 98, 205–207, 497
vector, 116, 122, 126 bifurcation, 77–83, 143–145, 147, 204–210,
polar, 12, 29, 91, 152 398–401, 404
Coupling, 71, 162, 169, 180, 374, 439, 473 Difference, 30, 33, 58, 79, 131, 193, 194, 204,
Coriolis, 473, 475, 476, 479 228, 230, 231, 316, 317, 321, 343,
Crack, 12, 50, 282 344, 363, 369, 381, 421, 425, 461,
damping, 50 476
Crane, 116, 131, 132, 136, 185, 198 phase, 421
Criterion, 24, 25, 44, 218, 250, 262, 285, 387, Differentiation
402, 419, 420, 433, 504 Euler, 194
Schur–Cohn, 402 backward, 194
Curvatures, 25, 52–54, 61, 62, 67 Dimension
Curve correlation, 372
dispersion, 98, 99 fractal, 339, 384, 394
plane, 364 Minkowski, 394
response, 484 Dirac
fermions, 337, 343–347
points, 337
D Discontinuity, 33
Damage, 11, 203 Disk
Damper falling, 449
magneto rheological (MRD), 494–501 rolling, 450–452
viscous, 495 vertical, 449
Damping rolling, 449–450
frequency, 96 Dislocation, 338, 340
structural, 418, 424 fractal, 337–340, 342, 347, 349
530 Index
Disorder, 103–105, 108, 109 Electroencephalogram (EEG), 372, 379

neuromuscular, 105 Electromyogram (EMG), 372, 380
Dispersion, 85–100 Electrooculogram (EOG), 372, 380
Bloch, 86, 87, 99 Element
Displacement massless, 115, 261
axial, 129, 283 spring-damping, 42, 115, 132, 189
longitudinal, 70, 87, 256, 258, 265 thin-walled, 175
virtual, 219, 222, 223 Ellipse
Dissipation, 70, 425, 491, 493, 496, 501, concentric, 311, 313
504–524 confocal, 311
energy, 493 Elongation, 41, 163
Distribution, 71, 109, 186, 248, 286–288, Embryo, 295–302
343–349, 466, 467, 505 EMG. See Electromyogram (EMG)
gamma, 466, 467 Energy
Domain, 12, 14, 21, 55, 89, 91, 110, 322, 353, contribution, 250
360, 363, 364, 372, 388, 390, 416, control, 225, 235–242
418, 420, 421, 426, 430, 457, 504, deformation
517–519, 521, 524 spring, 125
spectral, 372, 457 exchange, 21, 161, 162, 168, 170, 439
Drift, 321 external, 250, 251
bias, 476 flow, 415
Drilling internal, 250
directional, 280, 282 kinetic, 3, 5, 10, 117–124, 274, 277
horizontal, 281–292 mechanical, 86, 88
Dynamics potential, 3, 124–125, 295
chaotic, 50, 51, 53, 59, 72, 74, 103–113, EOG. See Electrooculogram (EOG)
143 Equation
spatio-temporal, 50, 51, 59, 72, 74 algebraic, 72, 252, 491
internal, 186, 189–191, 195–199, 416, 417 nonlinear, 12, 24, 127, 133
inverse, 185–200 compatibility, 25, 129, 136
irregular, 74, 108 conjugate, 165
nonlinear, 11–22, 25, 50, 51, 57, 61, 62, constraint, 116, 190, 192, 195, 276, 443,
64, 70–72, 74, 75, 205, 209, 337, 444, 448, 449, 454
494 differential, 3, 11, 12, 14, 15, 19, 28, 33,
quasi-periodic, 51, 74 50, 51, 59, 62, 72, 141, 172, 186,
system, 32, 140, 143, 162, 187, 193, 287, 193, 227, 239, 252, 258, 261, 287,
494, 509 298, 316, 318–320, 363, 367, 369,
underactuated, 185–200 371–381, 385, 467, 495, 506, 520
turbulent, 51 exponential, 165
shell, 69, 74 first order differential, 369, 385
flow, 158
generalized inverse, 223
E Gibbs–Appell, 223
EEG. See Electroencephalogram (EEG) homogenous, 478
Effect Kármán, 51
digital, 383 Lagrange, 5, 31, 162, 223, 274, 277
unsteady, 417 linearized, 480, 482
Eigenfunction, 339, 340 logistic, 326–330
Eigenvalues, 284, 321, 330, 339, 404, 469 motion, 151, 417–419, 422
Eigenvectors, 284, 393, 489 nonautonomous, 217, 220
Electrode nonholonomic, 444
annular, 476, 489 nonlinear, 194, 338, 339, 347, 397
pumping, 476 delay differential (DDE), 371–381
Index 531
ordinary differential (ODE), 24, 28, 33, 50, Fluid, 151, 154, 242, 494
56, 59, 65, 71, 72, 74, 75, 123, 186, viscous, 151–158
188, 199, 251, 298, 320, 363, 364, Flux, 459–464
366, 369, 391, 416, 467 light, 462
partial difference (PDE), 24, 50, 51, 56, 61, Force
64, 65, 72, 74, 75 actuator, 37–39, 41, 42, 44, 45, 47, 384,
Poincare, 223 485
Riccati, 518 Coriolis, 186, 473, 474, 476, 482, 483,
second order differential, 316 486
secular, 490 cutting, 177, 178
unconstrained, 238, 239 damping, 38, 43, 496, 497, 501
uncontrolled, 215, 228 Duffing type, 2, 3, 10
variational, 308, 315, 316 elastic, 13, 416, 418
Equilibria, 388, 390, 466–470 excitation, 296, 299, 302
unstable, 388 external, 125, 126, 299, 302, 408, 411,
Equilibrium, 24, 105, 108, 109, 163, 295, 434, 439, 505
296, 340, 387, 415, 416, 418–421, friction, 42, 135, 259, 261, 283, 285, 408,
424–426, 434, 435, 437, 440, 409, 411
468–470, 523 generalized, 126, 128, 129, 506,
asymptotically stable, 326, 420, 424, 425, 507
468 gravity, 124, 133, 135
Excitation harmonic, 283, 407–414
frequency, 53, 57, 74, 141, 145, 263, 264, external, 125, 126
268, 290, 299, 301, 484, 485, 491 horizontal, 132
parametric, 473–492 interaction, 283
Expansion, 21, 87, 88, 94, 96–99, 152–154, quadrature, 482, 483, 513
247, 284, 320, 321 reaction, 104–106, 187, 367
multiple scale, 21 restoring, 1–3, 10, 13, 17, 108, 260, 265,
Exponent 267, 494
Lyapunov, 51, 53, 70, 72, 143, 228, 232, spike, 39, 41–45, 47
372, 387 spring, 38, 236, 237
maximal, 15, 51, 53, 70–72, 75, 143, static, 283
225, 228, 232–234, 242, 246, 326, tangential, 417, 418
372, 387 Forcing
harmonic, 283, 407–414, 480,
F 494
Feedback orthogonal, 482
acceleration, 221 quadrature, 482, 483
control, 199, 221 dimensionless, 385
negative, 221, 224 Formula
signal, 205, 206, 208 iteration, 366
Field Klingel’s, 264
deformation, 338, 340, 343 Lagrange, 31
gravitational, 215, 397, 398 Frame, 194, 217, 271–273, 276, 423, 425,
stress, 93, 340, 343, 344, 349 443–451, 505, 506
temperature, 23 reference, 271–273, 276, 443–449,
vector, 322, 510–511, 523 451
Flexibility Frequency
bending, 116, 127–129, 131 bands, 96
longitudinal, 116, 124, 136 beat, 474, 475
Flow natural, 176, 177, 179–181, 184, 290, 475,
linearized, 321 477, 480–483, 485
streamline, 154, 504–507, 518 dominant, 176, 177
532 Index
Friction I
dry, 255–268, 407–414 Impact, 11–17, 21, 199, 296, 302, 353,
slider, 259 397–401
Function Imperfection, 356, 479
Airy’s, 56 structural, 474, 479, 489
autocorrelation, 53, 72, 143, 282, 284 Impulse, 24, 457–462
correlation, 343, 344, 346 Index
deflection, 52, 56, 60 arousal, 375
dimensionless complex, 340 sleep quality, 372, 376–381
distribution, 343–349, 467 Inertia, 179, 180, 189, 226, 251, 264, 274,
exponential, 165 276, 297, 385, 418, 419, 474, 475,
Hamiltonian, 308–310, 312, 313 505, 519
linear, 367 Instability, 57, 190, 265, 320, 420, 481,
meromorphic, 307 483
periodic, 97, 327, 331 boundary, 484
spatially, 97 Integral
scalar, 151 elliptic, 2, 338
sensitivity, 364, 366, 369 first, 4–6, 8, 10, 252, 307, 310, 313, 332,
stream, 151, 152 484, 503, 504, 507, 511–513,
stress, 52, 53, 56, 60, 61, 65 515–517, 519, 520, 523
transcendental, 503, 504, 515, 517 Integration
trigonometric, 20, 163, 504 Newmark, 287
vortex, 151 numerical, 233, 234, 247, 249, 474
Interaction, 23, 59, 62, 74, 162, 196, 209,
281–285, 291, 292, 305, 326, 330,
G
347, 416, 418, 424, 426, 504
Gait
Investigation
frequency, 42, 43
analytical, 398, 436, 507, 522
optimal, 38, 41, 42, 45
numerical, 205, 398
shift, 38, 41–47
Isomorphism, 17, 443
Gradient, 187, 250, 251, 423, 474, 523
Graph, 32, 131, 169, 191, 198, 247, 297, 299,
302, 356–358, 459–461, 463
J
Growth, 143, 320, 326–328, 330–334, 465,
Jump, 32, 62, 74, 208, 302, 319, 329,
491
332, 436
unbounded, 491
Gyroscope
hemispherical resonator (HRG), 474
L
MEMS, 473–492
Lagrangian, 2, 3, 162, 223, 227
synchronization, 226–233, 242
Law
conservation, 2
H energy-conservation, 5, 10
Hamiltonian, 245–253, 305–310, 312–316, friction, 407
340, 503 Limit
Heterogeneity, 87, 99 long-wave, 96
Holder, 175–184, 417, 421, 518, 519 stability, 81
weightless, 417 stress, 282
Human immunodeficiency virus (HIV), Limiter, 397–399, 405
465–471 Linearization, 241, 242, 363, 444
Hypnogram, 372, 374–381 Load
Hypothesis, 104, 353 critical, 24, 25, 28, 32, 33
Kirchhoff–Love, 64, 70 longitudinal, 52, 53, 59, 61
Hysteresis, 264–266, 494, 497, 501 non-homogeneous, 28
nonlinear, 494 resistance, 79, 82
Index 533
shear, 51, 56, 74 Method

transversal, 24, 27, 28, 51, 61, 62, 65, 66, Bubnov–Galerkin, 24, 28, 62
74 computational artificial intelligence, 245
finite difference (FDM), 24, 30–33, 50, 53,
56, 61, 71, 74, 75, 363
M finite element (FEM), 50, 71, 75, 86, 282
Manifold rigid (RFEM), 115–136
hyperbolic, 321, 326 modified, 116
invariant, 315, 319, 320, 433, 434, 436, harmonic balance, 421, 481, 482
437, 439 hybrid, 245
smooth, 321 multiple time scale (MTS), 172
Manner, 25, 51, 67, 217–219, 260, 261, 263, Newton–Raphson, 367
296, 355, 358, 438, 479 numerical, 21, 155, 158, 245, 384,
asynchronous, 358 387–389
Map, 59, 67, 72, 176, 205–206, 210, 289, 291, plane-wave expansion, 87, 96–99
292, 384–389, 391, 392, 395, 398, relaxation, 33
404 Ritz, 24
Poincaré, 53, 67, 398 Runge–Kutta, 28, 32, 53, 56, 61, 62, 65, 71,
Mapping, 12, 39, 205, 444, 452 74, 75, 134, 141, 252
simple cell, 384, 391–395 shooting, 363–369
Margin, 460 statistical forecasting, 245
hesitation, 355, 356 Micro-chaos, 383–395
Marrow, 86–88, 91, 92, 95–97, 99 Micro-fragmentation, 376, 378, 380
Mass Milling, 175–184
body, 108, 110, 112, 264 process, 176, 178, 179, 182
shift, 107–109 non-stationary, 178
centre, 117, 118, 385 Mistuning, 483, 484, 486
time-dependent, 430, 439 zero, 483, 486
Material Model
elastic-plastic, 23 Bolotin–Novikov, 59, 74
homogenous, 64 chaotic, 337
isotropic, 61, 64, 65 Coulomb friction, 283
Matrix cylinder assemblage, 92
angular natural frequencies, 179, 180 discrete-time, 204, 205
constraints, 179 kinematic, 444, 454
damping, 284 LuGre friction, 495
distribution, 186 mathematical, 3, 51–52, 56, 62, 105, 172,
dynamical, 489 219, 246, 251, 256–261, 266, 416,
equation, 277 465–471, 494
Jacobian, 187, 190, 366, 369, 469 multi-body, 261
mass, 136, 186, 241, 284 phenomenological, 416, 426
Moore–Penrose, 220, 224 planar, 116
inverse, 220, 224 rheological, 258, 259
symmetric, 17, 479 spatial, 116, 505
transformation, 506 two-point, 338, 340–344, 349
Member, 59, 153, 154, 224, 225 Modeling, 11, 38–39, 199, 219, 246, 247, 282,
structural, 24, 50, 51, 59, 62, 70 283, 373, 494
Memory, 140, 141, 143, 145, 149 Modulus
shape, 494 elasticity, 61, 283
Mesh elliptic, 2
three-dimensional, 86 shear, 88, 91, 94, 95
two-dimensional, 86 complex, 87
534 Index
Moment, 58, 295, 307, 342, 416, 506, 519 Lagrange, 119
inertial, 129 Laplace, 151
central, 117 non-Hermitian displacements, 340
Momentum, 10, 309, 313, 496 Reimman–Liouville, 139
conservation, 5, 10 Optics, 337, 343
Motion quantum, 337, 343
chaotic, 143, 145, 149, 227, 231, 251, 393, Optimization, 214, 251, 282, 353
395 Optomechatronics, 457–464
doubling, 227 Orbit
overshooting, 410–414 doubling, 79
periodic, 145, 227, 397, 398, 407–414 periodic, 79
sinusoidal, 397–405 Oscillation
slip, 409–413 chaotic, 203, 383
stick, 409–411 periodic, 110
subharmonic, 78, 203
Oscillator
N chaotic, 325
Nanosystem, 337–349 conservative, 10
Network coupled, 323, 349, 407–414
Bayesian, 352, 354 Duffing, 1–10
dynamic, 459 isochronous, 2
statement, 351–360 nonlinear, 1, 3, 10, 17, 20, 337, 429, 494
neural, 70–72, 75, 372 fractal, 337–349
static, 357 phenomenon, 299
Neuron spin, 71 biomechanical, 295
Nodes, 31, 338, 340, 342, 354 self-excited, 240
deviations, 342 spring-mass, 194
Noise state, 15, 21
electronic, 476, 487 stick-slip, 407, 409
floor, 488, 489 with hardening, 10
signal, 476, 478, 487–489, 492 with softening, 1, 2, 10
Nondetermination, 355 Output, 39, 44, 55–57, 71, 72, 79, 185–190,
Nonlinearity, 6, 17, 24, 59, 62, 136, 236, 302, 192, 194, 195, 197–199, 204, 205,
373, 418, 485, 494, 497 282, 285, 287, 288, 381, 386–389,
Number 392–393, 395, 462, 463, 476, 488
complex error, 39, 79
elliptic, 14, 17
hyperbolic, 16
hyperbolic, 12, 14, 15, 18, 22 P
real, 71 Parameter
Reynolds, 151, 156–158, 423, 424 initial, 341
Numerator, 190, 328, 332, 333 lumped, 61
unknown, 250, 363–365, 367
O Pattern, 37–47, 104, 112, 203, 204, 210, 246,
Object, 24, 25, 53, 62, 72, 80, 176, 246, 250, 249, 253, 390
251, 337, 351–354, 415–417, 426 topological, 389–392, 395
dynamic, 358, 360 Pendulum
Operator aerodynamic, 415–426
differential, 89, 166 inverted, 383–386
fractional, 139, 140 spatial, 518–520
Grünwald–Letnikov, 139, 140 spherical, 214, 216, 218, 219, 504, 518, 524
Hamilton, 338, 349 spring, 161–172
Index 535
Period, 1–10, 15, 40, 47, 55, 77–80, 141, 143, Power, 15, 21, 41, 42, 44, 45, 47, 51, 53, 57,
204–209, 227, 248, 251, 320, 330, 62, 63, 67, 72, 77–79, 111, 163, 166,
331, 388, 391, 392, 397–399, 410, 203–210, 234, 282, 285, 287, 288
411, 413, 466 electric, 415
latent, 466, 470 Precession, 227, 475, 491
Periodicity, 20, 21, 89, 90, 104, 324, 328, Precision, 163, 282, 284, 392, 478
331–333 structural, 476
Permutation, 447 Principle
Perturbation d’Alembert, 218–220, 222–224, 241
boundary shape, 91–92 Gauss, 218–221, 225
damping, 145 Probability, 282, 284–292, 354
initial, 145, 251, 295 Problem
Phase anisotropic Kepler, 305–307
slip, 408, 410 Cauchy, 24, 28, 32, 61, 62, 71
stick, 408, 410, 411 inverse simulation, 199
Phenomenon, 13, 59, 157, 295, 299, 301, 319, stability, 15, 24, 189
320, 323, 337, 394, 408, 413, 416, static, 24, 28, 127, 132, 133
424, 429, 495 Procedure, 14, 18–19, 38, 50, 57, 72, 82,
energy pumping, 429 86–91, 136, 176, 247, 278, 308,
Photons, 337 363–367, 392, 478
Plane wavelet-based Galerkin, 50
complex Process
elliptic, 16 fertilization, 295
hyperbolic, 16 unharmonic, 12
frontal, 104, 107, 108, 110, 111, 113 Property, 269, 296, 413, 446, 463, 475
sagittal, 104, 107, 108, 110, 111, 113 monotonic, 359
Plate, 24, 32, 49–75, 86, 106, 178, 180,
261
Platform, 104–106, 112, 113 Q
drilling, 131 Quantifiers, 372
Point recurrence plots, 372
attracting, 523 Quasilinearization, 363
contact, 178, 271, 443–448, 450, 452
critical, 32, 246, 248, 320, 326
equilibrium, 105, 109, 434, 437–438, 523 R
fold, 320 Ramp
inflexion, 320 simple, 78
mass, 37–40, 42, 44, 311–313 waveform, 80
material, 398 Realization
periodic, 388 orthogonal, 187–191, 193, 196, 198, 199
Poincare, 141 orthogonal-tangential, 187, 191, 193, 194,
repelling, 523 198, 199
Polarization, 337, 346 pure tangential, 187, 191, 198, 199
Polynomial skew-orthogonal, 188, 189, 195
characteristic, 402, 419, 420 Reasoning, 351
cubic, 398 diagnostic, 351
Portrait, 53, 72, 86, 246, 422 Recording, 337, 353, 372
phase, 67, 72, 142, 148, 149, 247–250, 411, polysomnographic, 371
412, 414, 434, 436, 521, 524 Regime
Posturography, 104 chaotic, 66, 77, 205, 206, 208, 210, 252
Potential non-periodic, 143
algebraic, 308 Regression, 394
non-smooth, 430 linear, 394
536 Index
Relation, 14, 24–26, 28, 51, 60, 71, 74, 90, 98, Shape, 11, 12, 16, 37, 91–92, 127–129, 131,
121, 124–126, 133, 141, 215–217, 171, 172, 176, 250, 283, 329, 368,
219–221, 231, 232, 234, 259, 260, 462, 463, 477, 484, 497
270–273, 277, 301, 338–340, 346, irregular, 461
354, 357, 359, 418–421, 445, 506, Shell
507, 513, 517, 519, 520, 523 isotropic, 51, 56, 74
invariant, 511–517 rectangular, 29, 51, 56, 59
Resonance, 21, 161, 162, 166–168, 172, Signal
184, 263, 265, 268, 432, 439, periodic, 78
441, 474 PWM, 77–79, 83
Response ramp, 78
beating, 437, 439–441 sawtooth, 79, 80
dynamic, 86, 139–149, 493–501 Simulation, 41–43, 45–47, 57, 74, 80, 83, 86,
strongly modulated (SMR), 439–441 129, 135, 140, 172, 181–184, 191,
Reynolds number, 151, 156–158, 423, 424 195–199, 234, 238, 247, 248, 256,
Rigid-body, 11–22, 181, 251–253, 503–524 260–266, 278, 279, 282, 286, 287,
Robot 289, 292, 349, 388, 416, 424, 425,
multi-wheeled, 444 466, 470, 471, 497–501
tricycle, 452–454 Monet Carlo, 282, 286–288, 292
Rod, 86, 423 Singularity
Root, 297, 307, 419, 420, 469 explicit, 187, 240, 320
imaginary, 469, 470 fold, 434, 435, 440
Rotation, 13, 17, 107, 108, 184, 270, 272–275, Sink, 391, 392
278, 290, 385, 416, 417, 423, 424, energy, 429–441
443–446, 449, 452, 453, 460, 461, Sleep
475, 507 rapid eyes movement (REM), 372, 374,
Rounding, 383, 385–389, 392–395 378–380
Rule, 11, 44, 46, 270, 371, 372, 416, 430, 459 slow wave, 372, 376, 378
Runge, 28, 32, 56, 65 Slippage, 131
Solid, 50, 51, 72, 86, 94, 96, 97, 233, 397, 424,
425
S mechanics, 305
Saddle, 208, 209, 404, 434, 437, 523 Solution
Sampling, 144, 146, 147, 250, 282, 286, 288, canard, 319
292, 383–385, 478 chaotic, 141, 145, 248, 389
strategy, 292 periodic, 6, 20, 80, 81, 143, 149, 326,
Scoring, 371, 372, 377, 381380 328–333, 399, 408, 411–412
sleep, 371–381 quasi-periodic, 141, 149, 248
Sensitivity, 364, 366, 369 steady state, 481, 482
Sensor, 79, 103, 104, 372, 476 Space
pressure, 104 mode, 474, 477
Series of translation, 445
covariance, 143 state, 204, 246, 387–389, 391, 392
Fourier, 20, 247, 431, 490 Speed, 17, 42–44, 46, 47, 176–178, 181, 182,
infinite, 97 184, 256, 262, 264–266, 268, 285,
power, 17, 21, 163 287, 291, 408, 416, 417, 423–425
Taylor’s, 365 flow, 417, 418, 423, 424, 458, 478
saw-tooth, 16 Spike, 37–47, 112, 325
Set Spring, 38, 42, 43, 115, 125, 161–172, 189,
edges connecting vertices, 357 191, 194, 195, 198, 236, 237, 256,
fuzzy, 357 258, 259, 262, 264, 265, 416
of vertices, 357, 358 massless, 132
Index 537
Stability neurological, 103

boundary, 481, 484–486, 491, 492 nonautonomous, 223, 232, 233, 242
loss, 24, 25, 51, 57, 62, 74, 319, 426 nonlinear, 223, 232, 233, 242
question, 330 Poincaré, 522
Stance topographical (PTS), 522
steady, 103 Poisson, 307
unstable, 104, 107 slender, 115, 116
State stochastic, 282, 284–285
initial, 190, 247, 387, 388 third-order, 510, 511, 515, 522–523
of equilibrium, 340 tissue regulation, 104
Stiffness, 51, 59, 60, 62, 72, 90, 96, 116, 129, unconstrained, 215–217, 220, 225, 236–239
130, 133, 175–184, 258, 261, 264, Van der Pol-Duffing, 139–149
284, 408, 430, 479, 496 vibrating, 17, 407
coefficient, 129, 134, 162 visual, 103
Stress wheeled, 270
local, 93 robotic, 443–454
pre-critical, 25 worm, 43
Subnetworks, 360
Symmetry T
central, 510 Tension, 133, 236, 282
mirror, 510 Term
System damping, 145, 476
aeroelastic, 415, 416, 426 singular, 19
balance, 103, 113 Theorem. See Theory
biomedical, 457 Theory
cart-mass, 198 central limit, 286
conservative, 238, 503–524 chaos, 372
constrained, 187, 215, 216, 219–221, 223, ergodic, 245
224, 238, 240 Fenichel geometric singular perturbation,
controlled, 217, 222, 224, 225, 233, 234, 319
239, 240, 384 Kalker’s rolling contact, 261
deterministic, 282–284 KAM, 245, 246, 248
dissipative, 517 Lyapunov’s stability, 245
Duffing-type, 2, 10 Morales–Ramis, 308
dynamical, 11, 25, 139, 140, 145, 149, of fractional calculus, 338, 347
161, 162, 222, 225, 228, 231–234, Poincare–Bendiksson, 155
245–253, 282, 324, 351–360, 389, Tikhonov, 320
391, 397, 399, 410, 504–524 Toda
electronic, 209 chain, 236, 238–242
hybrid, 203 potential, 236
expert, 351, 352, 354, 355, 360 spring, 236, 237
Hamiltonian, 245–253, 307, 308, 503 Torque, 104, 189, 191, 274, 275, 384
hybrid, 179, 204, 210 Tracking, 38, 39, 198–200, 213–243
linearized, 145, 433 frequency, 486
locomotion Trajectory
terrestrial, 37–47 closed, 248, 523
undulatory, 37 exact, 223–225, 242, 463
worm-like, 37 phase, 4–7, 9, 71, 109, 133, 155–157, 247,
mechanical 249, 503, 523
nonlinear, 213–243 limiting (LPT), 161–172
underactuated, 185–200 pseudo-periodic, 247
musculoskeletal, 103, 104, 108 sway, 107
538 Index
Transform 386, 389, 391, 445–447, 504–506,

Fourier, 143, 264 510–511, 518
short-time(DSTFT), 109–111 Cartesian, 87
Hough, 249, 250, 253 Velocity
Transformation angular, 227, 251, 271, 385, 386, 448, 473,
coordinate, 192, 489 474, 478, 506, 508–510, 519, 521
Euclidean, 13, 21 initial, 18, 215, 295, 296, 367
rotation, 21, 453 phase, 95
translation, 445, 453 Verification, 4, 10, 359, 444
Transition, 11, 16, 21, 25, 29, 51, 53, 54, 67, experimental, 47, 279
74, 112, 141, 152, 162, 172, 206, Vibration
208, 209, 240, 295–302, 322, 329, axial, 283
331, 333, 376, 459, 516 chaotic, 24, 57, 63, 66, 383
Type free, 132, 133, 295
Boolean, 354 modal, 474
Bose–Einstein, 343–348 parametric, 51–55, 72
Fermi–Dirac, 343–347 spectrum, 182
torsional, 116, 287, 290
Vibro-impact, 12, 429
Viscoelasticity, 88
V Volume, 56, 86, 91, 92, 94–96, 340
Validation, 72, 127–131, 136, 444, 449–454,
459
Validity, 56, 66, 71, 75, 352, 355, W
357–360 Wave
Value acoustic, 86, 88, 99
boundary, 19, 20, 56, 71, 75, 88, 89, dynamics, 50
363–369 harmonic, 17, 88, 95, 97
initial, 18, 19, 46, 324, 364, 366, 368, local, 86, 96
369 shear, 85–100
qualitative, 353 Wavelet
trial, 365 analysis, 50, 53, 56, 61, 109
Variability, 372 Daubechies, 50, 58
inter-rater, 372 Gauss
Variable complex, 59
complex, 431, 504 real, 59
spatial, 61 Meyer, 58
uniform, 512 Morlet, 59
Variation, 17, 24, 78, 86, 88, 93, 103, 108, complex, 59
110, 111, 113, 196, 308, 315, 316
spatial, 477, 478, 497, 500 Z
Vector, 12, 21, 71, 72, 89, 95, 98, 116, 122, Zona pellucida (ZP), 295–302
124, 126, 127, 179, 186, 187, 192, Zone
215–219, 222, 223, 234, 242, 250, subharmonic, 53, 74
251, 271, 277, 284, 285,322, 342, transitional, 53, 74

Applied Non-Linear Dynamical Systems PDF

Uploaded by

Copyright:

Available Formats

Applied Non-Linear Dynamical Systems PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Applied Non-Linear Dynamical Systems PDF

Uploaded by

Copyright:

Available Formats

Springer Proceedings in Mathematics & Statistics

More information about this series at http://www.springer.com/series/10533

ISSN 2194-1009 ISSN 2194-1017 (electronic)

© Springer International Publishing Switzerland 2014

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The 12th International Conference “Dynamical Systems-Theory and Applications”

conservation laws, equations of motions, and solutions of the governing equations

determination of critical values of the parameters responsible for change of the

Verhulst applies in Chap. 23 slow–fast timescales in the framework of Fenichel

Łódź and Warsaw, Poland Jan Awrejcewicz

Duffing-Type Oscillators with Amplitude-Independent Period . . . . . . . . . . . . . 1

Modelling of Ropes with Consideration of Large Deformations

Constrained n-Body Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

The Application of Parametric Excitation in Resonant MEMS

Ivana N. Kovacic and Richard H. Rand

Abstract Nonlinear oscillators with hardening and softening cubic Duffing

Classical Duffing oscillators are governed by

Their restoring force F D x ˙ x 3 includes a linear geometric term as well as a

x.0/ D A; xP .0/ D 0; (2)

I.N. Kovacic ()

© Springer International Publishing Switzerland 2014 1

THDO D p ; TSDO D q : (3)

finding also the corresponding Lagrangians, conservation laws, as well as solutions

The derivation of the mathematical model (4) is based on the transformation

Equation (7) now yields the kinetic energy

so that Eq. (8) becomes

Based on X 2 =2 C XP 2 =2 D const:, the system exhibits the first integral

The other firstRintegral is related to the principle of conservation of momentum for

that dI =dt D x, we obtain

a sin .t C ˛/ D x f; a cos .t C ˛/ D xP f C x 2 f 0 : (18)

and the equation of motion takes the form

Two first integrals (16) and (17) are

with initial conditions being [see Eq. (2)]

x1 .0/ D 0; x2 .0/ D xP 1 .0/ D A; x3 .0/ D xR 1 .0/ D 0: (24)

Equation (21) is plotted in Fig. 4a for h1 D 1. To analyze phase trajectories in more

By using (18) and (19) one can derive

xP sin .t C ˛/ C x 2 sin .t C ˛/ x cos .t C ˛/ D 0: (26)

Its solution satisfying Eq. (2) is

This equation of motion is now given by

x3 cos x1 x22 sin x1 C sin x1 D h2 : (33)

These two integrals can be manipulated to exclude x1 and to derive

.x22 1/2 .h1 C .1 h1 C h22 /x22 C x24 /2 C 2.1 h1 h22

Combining equations in (18) and using a D A and ˛ D =2, we derive

Its solution satisfying Eq. (2) is

1. Calogero, F.: Isochronous Systems. Oxford University Press, Oxford (2008)

© Springer International Publishing Switzerland 2014 11

2 Physical and Mathematical Principles

2.1 Linear and Elementary Nonlinear Phenomena

Although the notion of linearity is quite clear in terms of mathematical formulations,

q D A1 exp .i !1 t C '1 / C A2 exp .i !2 t C '2 / (1)

perfectly stiff plane (mirror)

2.2 Impact Models and Non-smooth Temporal Substitutions

Based on the class of linear generating models, whose typical representative is

Mechanical model Basic function Generated

Fig. 3 The “oscillating time” x

smooth functions (http://en.wikipedia.org/wiki/Split-complex_number). Moreover,

2.3 Different Asymptotic Approaches to the Vibration Theory

Harmonic limit Impact limit

Here we take D 2 for a closed cylindrical shell. In addition, the following

P D "R ŒG1 C1 C G1 D1 A B G1 HA C G1 Qq.t /;

@2 ./ 1 @./ 1 @2 ./

@4 ./ 2 @3 ./ 1 @./ 1 @./

2 @4 ./ 2 @3 ./ 4 @2 ./ 1 @4 ./

f0;j D 3 f1;j 3 f2;j C f3;j ; (23)

where fi;j D f .ri /j , ri D i h .i D 0; 1; 2; 3/, 0 j M 1, and h is the

f1;j D f1;j for 0 j M 1: (24)

wi;j D wi;M Cj ; Fi;j D Fi;M Cj for j D 0I 1; 0 i N 1: (25)