Fractional Calculus - Applications

MATHEMATICS RESEARCH DEVELOPMENTS
FRACTIONAL CALCULUS
APPLICATIONS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
Additional books in this series can be found on Nova’s website

under the Series tab.
Additional e-books in this series can be found on Nova’s website

under the e-book tab.
FRACTIONAL CALCULUS
APPLICATIONS
ROY ABI ZEID DAOU

AND
XAVIER MOREAU
EDITORS
New York
Copyright © 2015 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical
photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
nova.main@novapublishers.com
NOTICE TO THE READER

The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of
information contained in this book. The Publisher shall not be liable for any special,
consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or
reliance upon, this material. Any parts of this book based on government reports are so indicated
and copyright is claimed for those parts to the extent applicable to compilations of such works.
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise
contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter covered herein. It is sold with the clear understanding that the Publisher is not
engaged in rendering legal or any other professional services. If legal or any other expert
assistance is required, the services of a competent person should be sought. FROM A
DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
Additional color graphics may be available in the e-book version of this book.
LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA
Fractional calculus: applications/ editors: Roy Abi Zeid Daou (Lebanese German University, Sahel
Alma Campus, Keserwane, Lebanon) and Xavier Moreau (Université Bordeaux. 351, cours de la Libération,
33405 Talence, France).
pages cm. -- (Mathematics research developments)
Includes index.
ISBN: (eBook)
1. Fractional calculus. 2. Calculus. I. Abi Zeid Daou, Roy, editor. II. Moreau, Xavier, 1966- editor.
QA314.F725 2014
515'.83--dc23
2014038042
Published by Nova Science Publishers, Inc. † New York

CONTENTS
Foreword Alain Oustaloup vii

Preface ix
Chapter 1 The Heuristic Power of the Non Integer Differential Operator 1
in Physics: From Chaos to Emergence, Auto-Organisations
and Holistic Rules
Alain Le Méhauté
Chapter 2 Dynamics of Fractional Order Chaotic Systems 23

Sachin Bhalekar
Chapter 3 Pressure Control of CNG Engines by Non-integer Order 37

Controllers: A New Trend in Application of Fractional
Calculus to Automotive Systems
Paolo Lino and Guido Maione
Chapter 4 Linear Integer Order System Control by Fractional PI-State 65

Feedback
Rachid Mansouri, Maamar Bettayeb, Chahira Boussalem
and Ubaid M. Al-Saggaf
Chapter 5 From the Formal Concept Analysis to the Numerical Simulation 93

of the Thermal Diffusion Phenomena in a Finite Medium
Riad Assaf, Roy Abi Zeid Daou, Xavier Moreau
and Fady Christophy
Chapter 6 Temperature Control of a Diffusive Medium Using the CRONE 123

Approach
Fady Christophy, Xavier Moreau, Roy Abi Zeid Daou
and Riad Assaf
Chapter 7 Adaptive Second-Order Fractional Sliding Mode Control 149

with Application to Water Tanks Level Control
Danial Senejohnny, Mohammadreza Faieghi
and Hadi Delavari
vi Contents
Chapter 8 Features of Fractional Operators Involving Fractional Derivatives 165

and Their Applications to the Problems of Mechanics of Solids
Yury A. Rossikhin and Marina V. Shitikova
Chapter 9 Theory of Diffusive Stresses Based on the Fractional 227

Advection-Diffusion Equation
Yuriy Povstenko
Chapter 10 Modelling Drug Effect Using Fractional Calculus 243

Clara M. Ionescu
Chapter 11 Fuzzy Fractional PID Controllers: Analysis, Synthesis 259

and Implementation
Ramiro S. Barbosa and Isabel S. Jesus
Index 289
FOREWORD
Non-integer differentiation does not escape to the slogan “different operator, different
properties and performances”. This is indeed the concise formula that is likely to explain the
“why” of this operator, especially as most of its properties and performances favorably
distinguish, not only the operator itself, but also the models that use it.
It is true that we have established non-integer models that overcome the mass-damping
dilemma in mechanics and the stability-precision dilemma in automatic control, the
technological achievements associated with these models have been made possible thanks to
an adequate synthesis of the non-integer differentiation operator.
It is indeed the idea to synthesize non-integer differentiation (in a medium frequency
range) through a recursive distribution of passive components, of transitional frequencies or
of zeros and poles, which is at the origin of the non-integer differentiation operator real-time
use and, therefore, of both analogical and numerical applications that arise from it. As for the
corresponding dates, the synthesis as we have led it has been developed by stages and thus
proposed and experimented in the 70s for half-integer orders, in the 80s for real non-integer
orders and in the 90s for complex non-integer orders.
The first technological applications of this operator (henceforth usable in real-time) and
notably the first application in 1975 of a “half-integer order controller” to the frequency control
of a continuous dye laser, have widely contributed to take the non-integer differentiation out of
the mathematician drawers and to arouse new developments likely to enrich the theoretical
corpus of circuits and systems.
In this way France has been the first country to know a renewed interest on non-integer
differentiation, this renewal having been well relayed thanks to the dynamism of the foreign
scientific communities, at the European level as well as at the international level.
In this context, the French institutions have encouraged research in this field through the
acknowledgement of major scientific advances and the support of initiatives or actions aiming
to favor the synergies between the different themes and between the academic and industrial
components, the University-Industry partnership having indeed been nationally rewarded by the
AFCET’95 Trophy distinguishing the CRONE suspension as best technological innovation.
Concerning the acknowledgements, let us cite the selection of the CRONE control as a “striking
fact” of the Centre National de la Recherche Scientifique (CNRS) in 1997 and as “Flagship
Innovation” of Alstom in 2000 (Hanover and Baden Baden International Fairs, 2000), a Silver
Medal of the CNRS in 1997 and the Grand Prix Lazare Carnot 2011 of the Science Academy
(founded by the Ministry of Defense). Concerning the supports, let us cite the actions
viii Roy Abi Zeid Daou and Xavier Moreau
financially supported by the CNRS and the Ministry of Research: the edition of “La commande
CRONE” (Hermès, 1991) with an exceptional help of the ministry; the International Summer
School “Fractal and hyperbolic geometries, fractional and fractal derivatives in engineering,
applied physics and economics” (Bordeaux, 1994); the national project of the CNRS, “Non-
integer differentiation in vibratory insulation” (1997-1999); the colloquium “Fractional
differential systems” (Paris, 1998); the launching in 1999 of the thematic action of the Ministry
of Research “Systems with non-integer derivatives”; the launching in 2004 of the IFAC
Workshop “Fractional Differentiation and its Applications” through the first Workshop FDA’04
(Bordeaux, 2004) with S. Samko as chairman of the International Program Committee ; the
magisterial lecture “From diversity to unexpected dynamic performances” initiated by the
French Science Academy (Bordeaux, 5 January 2012).
But this academic support also found a guarantee in the industrial support brought by a
strong partnership with major companies as PSA Peugeot-Citroën, Bosch (Stuttgart) and
Alstom, such a partnership having indeed led to a high number of patents and technological
transfers that have widely proved the industrial interest of non-integer approaches.
Alongside the shared efforts to inscribe these approaches in the realist frame of the
University-Industry relations, our efforts have never stopped being shared with those of the
international scientific community to develop the best relations and collaborations within this
community. Without aiming for exhaustiveness, let us cite the involvement of European countries
in the diffusion, promotion and animation within the community: the research group “Fracalmo”,
which originates from Fractional calculus modelling, started with a round table discussion in
1996 during the 2nd International Conference “Transform methods and special functions” held in
Bulgaria; the journal “Fractional Calculus and Applied Analysis” (FCAA) started in 1998 with V.
Kiryakova as managing editor; the survey on the “Recent history of fractional calculus”
(Communications in Nonlinear Science and Numerical Simulation, 2010) at the initiative of J.T.
Machado who desired to make an inventory of the major documents and events in the area of
fractional calculus that had been produced or organized since 1974; the symposium “Fractional
Signals and Systems” (FSS) launched by M. Ortigueira in 2009 at Lisbon, then held in 2011 at
Coimbra (Portugal) and in 2013 at Ghent (Belgium). Both in and out of Europe, let us also recall
the various events of the Workshop FDA after its launching at Bordeaux in 2004 (under the aegis
of IFAC): Porto (Portugal) in 2006; Ankara (Turkey) in 2008; Badajoz (Spain) in 2010; Nanjing
(China) in 2012; Grenoble (France) in 2013; Catania (Italy) in 2014 (under the aegis of IEEE).
Founder of the CRONE team that counts today about ten permanent researchers, I
recognize this team to have always escorted me in the federative actions, launched within the
national or international scientific community, to energize and harmonize the researches on
both theoretical and applicative aspects of non-integer differentiation.
Xavier Moreau, member of the CRONE team, and Roy Abi Zeid Daou, associated member,
who are implied in this book as coordinators well attest this will to be at the initiative of collective
actions. It is then a pleasure for me, through this foreword, to warmly congratulate them for their
efforts of scientific coordination at the international level, especially as the number of proposed
contributions in this book well proves the dynamics of the fractional (or non-integer) community.
Alain Oustaloup
PREFACE
After presenting the first volume of this two-volume book, presenting a lot of
mathematical and theoretical studies and research related to non-integer calculus, the second
volume illustrates lots of applications related to this domain. In the following, we will present
the applications discussed in this book.
This volume is made up of 11 chapters. The first chapter presents the heuristic power of
the non-integer differential operators in physics starting from the chaos to the emergence, the
auto-organizations and the holistic rules. The second chapter shows the dynamics of the
fractional order chaotic systems along with some applications. The third chapter represents
the pressure control of gas engines by non-integer order controllers by showing a novel trend
in the application of the fractional calculus to automotive systems.
Chapter 4 shows the way to model fractional order equations using state space modeling
along with some applications. Another application related to this domain is the thermal
diffusive interface. Chapter five shows the analysis of a semi-infinite diffuse plane medium
along with the equations that model this medium, and some frequency and time domain
responses. However, chapter six treats this problem by controlling this plant using the well-
known CRONE controller.
Chapter eight presents the adaptive second-order fractional sliding mode control with an
application to a water tanks level system. Chapter nine treats the mechanical aspect by
showing the features of the fractional operators applied to this domain. Also, chapter nine
presents the theory of diffusive stresses based on the fractional advection-diffusion equation.
The modeling of drug diffusion during general anesthesia using Fractional Calculus is
shown in chapter ten and is considered as another application related to the biomedical field.
At the end, chapter eleven represents an overview of the fractional fuzzy controllers by
showing the analysis, the synthesis and the implementation of this module.
To sum up, this second volume presents lot of applications of fractional calculus in
several engineering domains as the thermal, the automotive, the mechanical, the biomedical
and much more. Note that this volume was preceded by a first volume that focuses on the
mathematical and theoretical aspects of fractional calculus.
In: Fractional Calculus: Applications ISBN: 978-1-63463-221-8
Editors: Roy Abi Zeid Daou and Xavier Moreau © 2015 Nova Science Publishers, Inc.
Chapter 1
THE HEURISTIC POWER OF THE NON INTEGER

DIFFERENTIAL OPERATOR IN PHYSICS:
FROM CHAOS TO EMERGENCE,
AUTO-ORGANISATIONS AND HOLISTIC RULES
Alain Le Méhauté
Physics Department, Kazan Federal University, Kazan, Russia
Abstract
The use of fractional differential equations raises a paradox due to the non-respect of the
space time noetherian axioms. In environments characterized by scaling laws (hyperbolic
geometry associated with fractional diff-integral) energy is no more the invariant of the
dynamics. Nevertheless the experimental action requiring the use of energy, the relevant
representation of the fractional-process, must be extended. The extension is carried out using
the canonical transfer functions in Fourier space and explained by their links with the
Riemann zeta function. Category theory informs the extension problem.
Ultimately the extension can be expressed by a simple change of referential. It leads to
embed the time in the complex space. This change unveils the presence of a time singularity at
infinity.
The paradox of the energy in the fractality illuminates the heuristic power of the
fractional differential equations. In this mathematical frame, it is shown that the dual
requirement of the frequency response to differential equations of non-integer order and of the
notherian constraints make gushing out a source of negentropique likely to formalize the
emergence of macroscopic correlations into self-organized structures as well as holistic rules
of behaviour.
1. Introduction
The heuristic power of the fractional derivation operator is based at least upon three
theoretical foundations

E-mail address: alm@materialsdesign.com
2 Alain Le Méhauté
 the canonical link between the fractional operators and fractal geometries. This type
of operator expresses the non-differentiability of the geometry (Le Méhauté, 1982a),
 the generalization of exponentiation operation in the frequency space. This
generalization is related to partial hyperbolic geodesics (Nivanen, 2005),
 the relation between this generalization and Riemann‟s (and Goldbach‟s) hypothesis
that gives meaning to the universal approximations of all analytic functions (Voronin,
1975) (Bagshi, 1987).
The assertion of heuristic power of fractional diff-integral operators finds its sources,
within the first generalization of Zipf laws in the form of so-called Zipf-Mandelbrot law (Le
Méhauté, 1974, 1977). These publications link the recursive structures of the statistical
supports (trees and hyperbolic groups) and the thermodynamic laws by merging geometries
and singularity distributions. By extending the concept of recursivity into geometry,
Mandelbrot published in 1975 his first works on „fractal geometries‟ (Mandelbrot 1975,
1982), a word forged for the purposes of the case. To describe the exchange of energy and
matter dynamics in heterogeneous environments (ie fractal structures) (here considered as a
mathematical distribution ground (Schwartz, 1950)) the author of this paper has developed at
the end of the seventies a model of transfer of energy on fractal structure embedded in a
physical environment (electrolyte, liquid, air, etc.) named TEISI model (Le Méhauté, 1980,
1982a, 1984). This model merges a class of convolution between fractal structures (geometric
factor) and traditional differential operators (dynamics factor). It gives a theoretical base to a
new type of irreversible dynamics. After the Oldham and Spanier (Oldham, 1974) and
Nigmatullin‟s senior formal works (Nigmatullin, 1975), the TEISI model and its fractional
operators feature the way for a theoretical analysis of processes characterized by an
irreversibility based upon geometry. Jointly, the relevance of these operators for the
optimization of electrochemical energy storage devices (Le Méhauté, 1982b, 1986) was
performed. There followed a „declension‟ of many issues about irreversibility that leads to the
establishment of a link between fractals and hyperbolic geometries (Nivanen, 2005), number
theory (Le Méhauté, 2010) and theory of categories (Riot, 2013).
Despite an undeniable prospective effectiveness and numerous coherent experimental
data (Le Méhauté, 1983, 1986, 1992), opinion still mainly given to this work, is the one
expressed by Pierre Gilles de Gennes referee to the Academy of Sciences of Paris, in the
defense of the first publications of the author (Le Méhauté 1982a). He wrote in footnote page
“L’introduction d’une hypothèse de self similarité sur la structure de certaines électrodes
parait extrêmement féconde. Par contre l’équation reliant directement la dépendance en
fréquence de l’impédance à un paramètre purement géométrique (…) semble très
conjecturale. Dans les quelques exemples connus de transport impliquant un fractal (amas de
percolation, réseaux de Sierpinsky,…) l’exposant de conductivité n’est pas lié directement à
la dimensionnalité”. Let us see why the assumption of the transport of species evoked by P.
G. de Gennes, twists the empirical point of view with respect to the hypothesis based on the
transfer of energy. Due to the range of frequency analysis (Hertz plus microwaves range) the
analysis of the efficiency of (i) the batteries and (ii) of the cable impedance served as an
excellent compass.
Let us see why despite the approval of the scientific community to de Gennes‟ reserves,
and after the epistemological choice of this community for considering the fractal interface
The Heuristic Power of the Non Integer Differential Operator in Physics 3
only as an external boundary conditions with regard to the transportation of physical species
(matter, ions, electrons, entropy, energy), the author continued his work in the original
direction by addressing the issue of transport in fractal media through the theory of
distributions as a simple limit case of transfer process.
2. A Nice Equation for an Heuristic Power

“Sole a beautiful equation may have a deep meaning!” This is the case of the equation
below. It establishes a merging between fractal geometry and the operator of the fractional
differentiation in the Fourier space, via the set of the integer numbers:
()d i  const , (1)
As shown in (Le Méhauté, 1982a), this equation does not a priori assume, (as one may
understand a posteriori) the presence of a link between space and time. In the frame of
distribution theory it just assumes the empirical constraint for a measurement of a recursive
ordered structure (interface), which canonical expression is given via a set of integers. The
emergence, a posteriori, of temporal variables is a simple mathematical consequence of the use
of this one dimensional set when it is plunged in the set of pure complex numbers. This
mathematical embedding comes from the interpretation of the Hausdorff-Mandelbrot equation
 d  const , (2)
(Mandelbrot, 1975, 1982) by using the number expressed by. In equation (1) the complex
number „i‟ expresses the independence of the local measure of the space with regards to the
computation required by a global normalization of the fractal interface. The reference „‟
expresses the requirement for time normalisation. The normalization keeps the scaling
invariance to the fractal geometry, when it is used as support of the exchange for the transfer
of energy.
Beyond all the outstanding issues about the definition of the operator in time (the whole
of this book), this equation expresses the issue open by the non-integer derivation, in its
recursive simplicity. For the reasons given above and developed below, we assert that the
Fourier space is the natural space of description for all recursive geometries:
   
1
. (3)
i1/ d
We maintain that, including in a majority of experimental cases, it exists a direct link

between the expression of the derivation of non-integer order in this space and the fractal
dimension of a substructure that is the support of the transfer of energy. It is expressed
physically through a generalization of the concept of capacity that we called „Fractance‟ (i.e.,
capacity on fractal support).
This concept founds the use of non-integer derivative in physics. Moreover the use of this
formulation in the frequency domain (Fourier or Laplace) postulates a „computational
foundation‟ for the emergence of well-ordered variable that we name the „irreversible time‟
which is practically useless in mechanics (Rovelli, 2004, 2006) (Connes 1994)
Such a variable must be considered like simple mathematical object in accordance with
Hausdorff-Mandelbrot rule. In practice, equation (1) interpolates an invariant factor of the
fractal dynamics between the concept of velocity (, d = 1) and concept the diffusion coefficient
(, d = 2). Probably intuitively perceived by de Gennes when assuming exclusively transport
processes, this interpolation points out an apparent physical „weakness‟. Indeed, due to its
dimensions, the new invariant cannot possess a simple physical meaning. Meanwhile at the
beginning of the eighties, the heuristic hypotheses of dynamic recursivity described via
diff-integral, was already well assured by the results obtained from the efficiency of the
batteries, and also supports preliminary development of the CRONE control (Oustaloup, 1983).
There was something of a contradiction. Including with the Mandelbrot‟s works the
engineering was at that time the spearhead of dynamics in fractality and advanced open
interrogations. The paradox raised by the contradiction between theory and experiment, (no
solved by the use of the concept of transport) unveiled an unquestionable opportunity for
making a conceptual breakthrough.
3. SWOT Method, Non Integer Diff-Integral and Co-Dimension

We recall that the TEISI model assumes a 2D symmetry of the macroscopic experimental
complex structure which geometry looks like a capacitor. In addition the symmetry of the
force-field locates the fractal dimension d in the range given by.
The relevant remark of P. G. de Gennes concerning the weakness of the TEISI model was
finally upset as follows: as a matter of fact, the relevant dimensional invariants required in
expression (1) are not part of the usual invariants used when energy is exchanged upon
integer dimensional interface. In that standard case, the dimensional equation for the
energy is:
E  ML2t 2 , (4)
(in Fourier space) and the dimension of the physical action is given by
A  ML2t 1 , (5)
(in Fourier space). These dimensions cannot be simply applied when environments is
characterized by scaling properties. It has to be really „adapted‟ to the problem. The standard
energy is expressed in space-time by the square of a velocity. This square factor cannot
appear naturally when dynamics is deployed in fractal geometry: indeed, the support is not
differentiable and therefore, cannot easily have any gradient operator at its disposal.
The second question addressed by the problematic concerns the global point of view
upon the interface: the space-time integral, expressed by the macroscopic law
xd  t , (6)
cannot be reduced to the diffusive formulation
x2  t , (7)
or even to
xd / 2  t , (8)
initial de Gennes proposal. Because, any experimental process is dissipative it might appear
simple to express the problem of fractality as an anomalous diffusive process
() 2 i 1/ d  const . (9)
Alas! There was worse to come: these formulations didn‟t solve, in any manner, the
controversial issue transport against transfer pointed out above. Are these threats, a
condemnation of the use of distribution theory and the non-integer order operators? The
answer is clearly negative. All attempts to simply mask the defect of fitting between the
concept of energy and the concept of fractal, is just a dead end. This question requires, as we
have shown very early (1978), a reference to the concept of fractal co-dimension (Tricot,
1999). If the formulation (1) infringes the principles of Emmy Noether (energy cannot more
be the expression of the homogeneity of fractal space-time), the difficulties are not
mathematically intrinsic but extrinsic. They are strongly related to the contradiction between
the Noether‟s axioms and the scaling properties of the fractal interface measurements. The
contradiction between standard theories and experiments, when fractal media is involved, is
due to the restrictive requirement imposed by the use of energy for checking the physical
properties. This difficulty must be addressed to an epistemological issue that not only
assumes an extrinsic meaning of energy (homogeneity of space-time) but uses of balls (in
maths) or/and energy (in physics) to perform all practical measurement even if far from
homogeneity (complex media). The non-differentiability and the related fractional statistics,
hidden behind the transfer TEISI model, unveil precisely the conceptual limitations that must
be addressed head on.
If the fractal dimension is d, let us observe that the Noetherian homogeneity can be
conceptually „restored‟ using the concept co-dimensional space by writing
  
E  MLd t 1 L2  d , (10)
or, in Fourier space, the geometric co-density of energy

E / L2 d  MLd t 1 ,  (11)
is obviously the Hausdorff-Mandelbrot dimensional content of the co-fractal space. Let us

also observe that the formulations can be reversed by writing
  
E  ML2  d t 1 Ld (12)
and
 
E / Ld  ML2  d t 1 . (13)
This inversion points out one of the major symmetry for the transfer of energy in fractal
media. The question raised by these formulas is the meaning of the co-fractality (). It is
obviously related to the embedding of the fractal interface in the two dimensional space. This
embedding is required to perform measurement and experiences as well. As exemplified for a
long time in energy storage industry (battery energy/power balance optimization), we can
confirm that the fractality, which raises the paradox Energy vs Fractal, also opens up new
opportunities for technical breakthrough: the design of the geometry is an extrinsic
optimization factor of the irreversibility of the process deployed in fractal media. Energy
storage requires 3D structures whereas Power requires 2D interface. Depending of the need,
the fractal dimension balance is obviously in between. Conversely, the entropy/negentropy
ratio must control the metric and therefore „curvature‟ of a fractal geometry.
We will see that the constraint of continuity along the fractal object is a central factor for
optimization principle. It is expressed by a geometrical phase. Then the metric of the
geometry determines the ratio of the entropy/negentropy production. To easily understand the
contents of the role of this ratio we need to consider briefly the fractional generalization of
exponentiation.
4. The Generalization of the Exponential Concept

The „exponential‟ and „logarithm‟ functions play a central role in physics. This role has
its origin in the morphism between addition and multiplication. As shown in category theory,
this morphism is essential in mathematics and physics because it establishes a strict
relationship between the constructivism performance, that is the addition of ordered structures
(adding is considered then as a co-product), and the partition or deconstruction of the same
structures into substructures (represented by the product). The question addressed by this
morphism concerns the content of the „differance‟ with „a‟ and not „e‟ (Derrida, 1973) which
controls the irreversibility. We shall find in non-extensive thermodynamic developments, a
useful analytic extension of exponential functions and logarithms within the time space
(Nivanen, 2005). However, these extensions can only reach a deep geometrical meaning in
Fourier space.
The exponential term in the Fourier space is always given through a semicircle
R
Z ()  . (14)
1  i
This is a transformation obtained by a geometrical reversing of a straight line with respect

to a point located outside of the line. This modular formulation (Ghys, 1990) is canonical and
it is rightly adapted to our purposes because it opens the matter of geodesics in hyperbolic
space. The impedance takes the dimension of a resistor:. For longtime and R defines a state in
the thermodynamic sense. The semicircle, inverse of a straight line, is also a geodesic because
its inverse Fourier transform is (i) the classical first order relaxation dynamics solution and
(ii) the merging of additive and multiplicative properties. Both properties give to the
exponential function a canonical characteristic of a closure (construction, -algebra/-
algebra, deconstruction).
Starting from that observation an heuristic approach built from this canonical form leads
to consider the TEISI model as a rather natural „declension‟ of exponential function because
the introduction of the fractal geometry as an interface for transfer of energy involves
geodesics as arcs of circle. We recall that TEISI model leads
R cos 
Z1 / d ()  . (15)
1  i 1 / d
Then, we can confirm that
u()
 ()  , (16)
v()
expresses the hyperbolic distance on . The scaling correlations conditions are deliver for
(respectively) where is the phase angle associated with the rotation given by in the complex
plan. This phase angle which requires information coming from the environment renders the
above Cole and Cole representation incomplete. The paradox is not only the absence of
standard thermodynamic equilibrium state function at infinity, (when this infinity is defined
by ), but also the requirement for a couple of referentials to define the whole representation.
Even if the origin of the phase at infinity comes from intrinsic recursive correlations
characterizing the fractal support of the process, it requires the embedding within a larger
environment. This requirement looks like Kan extension and virtualisation of sets in the frame
of the theory of categories (Riot, 2013). In spite of these difficulties Gemant analysis
(Gemant, 1935) and Cole-Cole synthetic formulation (Cole, 1942) has a real empirical
effectiveness. Let us observe that the fractional process is characterized by. Then the points at
infinity contain phase information which reduces relatively to R the entropy. Therefore,
beyond the questions raised by the temporal expression, the Fourier space representation
reveals a new core of problems concerning the balance between entropy and negentropy
production in complex media: the role of the boundary (environment). Then the question of
local causality in the traditional sense can be clearly addressed precisely from the incomplete
TEISI transfer formulation.
Hyperbolic distance
u / v   
1/ d
Z  ( ) 
v u
 0

 1  1 / d 
2
Figure 1. Typical Cole-Cole transfer function for a canonical fractional differential equation (Le
Méhauté, 1982).
To give a content to this remarks it is possible to refer to the impedance related to

diffusive TEISI degeneration when fractal interface fulfils the 2D space.
5. Diffusion Under Field

The diffusive model can be written in the frame of the TEISI model as . Let us compare
this equation with the fractional expression of the diffusion equation that uses non integer
differential equations (Oldham, 1974) (Podlubny, 1999):
  2  1 / 2   1 / 2  
 D 2   1 / 2  D  1 / 2  D  . (17)
t x  t x  t x 
The second factor of that equation is a non-causal operator, whose physical meaning, to
our knowledge, has never been considered. Even hidden, this factor plays a role which is
crucial especially if some characteristics can fit the complementary set of
R cos  / 4
Z1 / 2 ()  , (18)
1  i 1 / 2
in the complex plan, that is to say, the role played by the co-dimension in the general TEISI
analysis. In order to fit this meaning, we have to observe that the thermodynamic „gradient‟
must be written
U
 U  (t )  U (t ) . (19)
x
It is expressed by the means of a pivot which is an energy in the process used for
measurement. The harmonic ratio
U ()
 Z () (20)
U  ()
is then precisely the impedance. As shown by J. P. Badiali in the frame of statistical

thermodynamics when based on path integrals (Badiali, 2013), this energy is used as a
reference point for energy balance. This pivot gives a generalization of the physical concept
of equilibrium and a reference for causality principle. The main question is the following:
when such a reference is used, is the entropy increasing or decreasing with respect to the
fractal correlations along the process? Indeed, the conditions of the existence of a flow of
„negentropy‟ remain to be clarified to address the issue of non-causal factor. The diffusive
example can serve as a guide to address the issue of the absence of energetical dimensional
fitting because this absence is clearly related to this non-causal factor.
1 / 2
 D
t1 / 2
Z1 / 2 ( ) 0
1 / 2
 D
t1 / 2
  45

t
0
Z 1 / d ( ) 0
Forward Backward

Figure 2. Analysis of diffusive process with field using the merging of forward and backward fractional
process in the Poincaré fundamental domain (Nivanen, 2005).
As evoked above, due to the use of energy as the essential experimental factor for defining
the equilibrium, the representation of empirical data by the means of a derivative operator of
non-integer order, over fit like the ball of measurement the fractal interface properties. So the
description of the fractal reality with non-integer operator is almost surely incomplete with
respect to the physical experiment. Nevertheless this assertion is paradoxical if coupled to the
Godel incompleteness theorems: the set of rational numbers is obviously larger than the set of
natural numbers. To precise the content of this paradox we have to analyse the response given
when half diff-integral is considered in the frame of non-integer operator application? Unlike
transport models and the use of temporal operators without precaution, the Fourier space used to
express the TEISI model opens into the origin of a natural incompleteness. We have already
shown that the model is related to the chaotic dynamics upon punctuated torus characterized by
an angle at boundary (Nivanen, 2005). In the diffusive frame, the angle at infinity is
   / 2/ 2 . This angle is critical: the canonical diffusive process (d=2) under field may be
considered as the merging schematized in the diagram given below. A couple of causal
(forward) and a-causal (backward) tiles of a generalized Poincaré fundamental domain mapping
(Figures 2 and 3) enters in confusion if   1/ 2 .
Because for d=2, the diagram merges a couple of components for stating a new meaning
of the resistance R, that is to say the role of local energy within the whole dissipative process.
The merging renders obvious the role played by the referential of the representation. The
passing from the non-integer order differential equation (irrelevance of the energy) to its
integer counterpart (relevance of the energy) requires a rotation of the external referential
with respect to the angle. The referential attached to the fractal interface (external with regard
to energy) is opposed to the overall-referential (qualified as internal). In the first referential
(limited to the harmonic analysis) the operational time is associated with the inverse Fourier
transform of omega. The process is linearly approximated. In the frame of the second
referential, the relevance of the time requires to take into account a singularity at boundary. In
the diffusive traditional equation, an initialization of a process in the Atlantic east coast of
United States, involved an immediate modification of the state of the sea in the west coast of
Britany. This non-physical singularity is fortunately a forgotten non-physical characteristic of
integer representation of a diffusive process. This paradox refers to internal equation
referential. In the first referential, nonobstant the singularity at infinity, the process states a
pseudo-equilibrium, as „Zeit Objekt‟. This object is given under the sole linear constraint. It is
qualified as a pseudo-equilibrium because it satisfies independence from the clock-time
variable, seen as an inverse transform of Fourier frequency. However this time is not the
relevant variable with respect to the energy balance. This balance requires a meaning for the
local concept of differentiation. It forgets the singularity on the time variable mainly ignores
on the way of fractional representation when taking into account the sole fractality. Somehow
paradoxically, the singularity involved with the complex time, restores the Noetherian-
homogeneity of lost space-time via the sole fractality.
We shall show that this singularity contributes to the emergence of a negentropique
factor. This factor balances the disequilibrium of the „energy‟ between the fractal interface
characterized by the forward process, using „something happening‟ in the co-fractal
environment. The theory of the categories may explain this emergence as a Kan extension of
the fractal space-time (Riot, 2013). At this stage the controversy opened in 1982 reaches its
heuristic pungency. The will stays focused on the TEISI transfer process supported by a huge
undeniable set of experimental data that ignores transportation. The angle at infinity appears
naturally as a key factor of the paradoxes concerning the energy. Since the incompleteness of
non-integer operators is pointed out, it also asks for understanding the precise choice of the
space time referential. From that point of view, the problem looks like the issue of the
Coriolis momentum within a rotational referential. The analogy is peculiarly relevant because
fractal geometry is strongly linked to the hyperbolic geometry. In this frame the fractal
dimension d is related to a curvature of the geodesics, i.e., their common acceleration. This
acceleration is characterized by the singularity of the dynamics that characterizes an extended
„time‟ at infinity. Then, the paradox of the energy, led by the role of the resistance R, that is to
say, the term of dissipation of energy, is related to the geometrical phase angle. According to
the canonical model TEISI this angle carries a factor related to the internal correlations upon
recursion which states, as for Coriolis momentum, the duality of the referential system when
fractal space is considered from inside or from outside. This is clearly unveiled when we
relate both Natural and Fourier spaces.
For example, the inverse Fourier transform of gives a measure of local Dirac (Angot,
1972), but also gives meaning to the ball for measurement upon the fractal object.
Conversely, the measurement of the gauge , does not have any analytic expression in
Fourier space. The gauge  is the only variable which may simply be expressed in physical
space via except if we forget the Fourier Transform limitations
1
1
 1
1 / i d 1 :  (t )  t d / 1   . (21)
 d
After analysis, the splitting between natural and Fourier space is normal, but may be
surprising especially since the changes associated with d = 2, lead the exact identity of the
diffusion operators involving a simple square roots:
1 1
: . (22)
i 1/ 2
t / 1/ 2
This simplicity is not general if. Therefore we cannot consider the diffusive process
(respectively the deterministic case) as a mere extension of the general fractal process. As
already pointed by Peano and Cantor for recursive geometries and in another area by Galois
and Abel with non-solvable groups, the diffusive process leads to face a degenerate form
which cannot easily be derived from the general fractal process. This explains the reported
link within others notes between our analysis and Gromov‟s hyperbolic groups: due to the
absence of sharpness on branching (small-triangle at infinity), the scaling process cannot be
completely defined from the hyperbolic tree seen as a set of branches. There is an uncertainty
at branching. The difficulty lies precisely in the degenerate nature of this issue if d = 2
(respectively if d = 1). In the case d = 1 the semi-circle fully explains the local exchange of
energy. In the case d = 2, we can consider the isometry between the arc and the euclidean
process by passing the nonlinear scaling. As shown in the scheme below, due to equivalence
of metric between hyperbolic and Euclidean (Nash theorem), we can use the fundamental
Poincaré‟s group to lead an isometry represented as a pseudo-linear application.
The illustration of the isomorphism gives a simple obvious form to the geometric content
of the Laplace operator d=2. This isomorphism, well understood from external algebra,
manages the relationship between Euclidean space, associated to the local exchange of energy
(gradient) and the hyperbolic geometry of the whole set of exchanges upon the interface
characterized by a Peano structure (divergence of the gradient vs. TEISI). Insofar d = 2, the
interface environment is none other than the interface itself. It prohibits any leakage of energy
out of the local entropy production zone.
Figure 3. Illustration of the Nash isomorphism in the case of diffusive process with field (d=2).
Although Nash‟s theorem ensures that, such an isometry exists almost surely, even for a
more complex hyperbolic structure than a Peano one, that is to say , intuition is much less
immediate and the associated application is in this case, no longer linear as shown in the
diagram below. In this diagram the tilling of the automorphism (related to the dynamics with )
is very distinct of the Poincaré‟s mapping.
0
Z 1 / d ( )
0
Z ( )
Figure 4. Representation in the frame of TEISI model of Nash‟s isomorphism in the general case.
The difficulties already pointed out can then be summarized as follows: merged in the
same „object‟, the sums of building (constructivist view and co-product) and of the products
of processes (leading to partitions or reverse engineering), the extension of the concept of
exponential in Fourier space, offers almost all the good properties of a generic functions.
These functions are strongly connected with fractal metrics. Its representation in the complex
plan is indeed a piece of hyperbolic geodesic. Nevertheless, this piece cannot possess one of
the crucial properties required to perform a rigorous reverse engineering of the process: the
distributivity. A „difference‟ remains after any proceeding along a loop of building and
deconstruction. This „difference‟ is expressed by a phase factor (time irreversibility). Once
again, for understanding this subtle issue we have to refer to the theory of categories. It
explains the paradox of the energy, by suggesting the Kan extension as an operation to be
considered for understanding the storage of information within the phase factor.
The use of any types of generalized exponential must be done with caution. It requires a
deep understanding of physical phenomena in its own and external space time structure. A
geometric understanding must guide the use of related fractional operators for the physical
representation of complex systems. Recall that what is behind the will of the representation of
the reality is almost always the desire to find a causal reason. Here, this must be carefully
considered only as a linear will. It may be a non-relevant extension of standard paradigms in
the case of complex environment. In spite of the non-linearity of power laws, the scientist can
easily fall into the trap of an artificial linearization. It is precisely to avoid this trap that it is
necessary to understand the link between non-integer dynamics and Riemann zeta function as
a topologic, or more precisely, homotopic parametrization of fractional dynamics.
6. Riemann Zeta Function and Non-Integer Differentiation

We recall that the expressions of the Zeta functions and are arithmetically related. We do
not enter below in the subtilities of this relationship insofar it is well known for the experts in
number theory. We considered only the function assuming that generalizations in the frame
TEISI model are obvious. We recall that this function is expressed in the set of integers by (i)
a power law under a form of a co-product:
n
1
 ( s)  s
, (23)
n 1
with and also (ii) as a product based on the prime numbers (Tenenbaum2011):
 
 1 
 (s)     . (24)
pi  
1
1
 ps 
 i 
The Riemann function therefore carries the combination of sums and products merged
together. This merging recalls all the problems mentioned above. As consequence, the
relationship between zeta function and fractality seems rather natural and the author has
developed this question in details (Le Méhauté, 2010). The theory of categories makes echoes
to this proximity in relation with the punctuated torus. In addition we have already derived a
physical demonstration of the Riemann hypothesis based on the TEISI model by using a
transition phase geometrical model. The basic diagram of the proof is given below (Figure 5).
According to the parametrizations of the hyperbolic distance v / u  1/  this diagram
1/ d
suggests to introduce and then for building a topological morphism between each terms of the
sum  (  i ) and each integer point upon Z (i) . Each couple of points is obviously
parametrized using integer frequency.
 (1  s )  (s)
 (1  s)

 

 (s )
 s  1 / 2      
Figure 5. Key representation of the Riemann Zeta Function hearth in the frame of the TEISI model (Le
Méhauté, 2010). Also representation of macro KMS condition versus Fractional Differentiation
Riemann Approach through exchange of axis.
More generally, the zeta function (respectively its inverse) is part of a larger set, taking
into account the sign of the rotation in the complex plan especially and according to the new
set  ( s),  (1  s),  ( s ),  (1  s ) . This set is obviously centered upon   1 / 2 . This set may be
associated through a morphism to a new set of transfer functions which appear clearly in the
diagram above. Both sets authorize the construction of a new morphism upon the sets .... .
The phase characterizing the rotations, and coupling the components of the sets .... , it
constrains the morphism between the sets, and this constraint allows a thorough
understanding of Riemann as well as Goldbach‟s hypothesis. The trapping of a geometric
phase for defining the thermodynamic states is the key factor of the fractal approach of
Riemann‟s hypothesis (Le Méhauté, 2010).
The Riemann hypothesis establishes that there are zeros for the zeta function in the only
event . This constraint establishes a direct link between the diffusive process analysed via
fractional operators and Riemann zeros. The Goldbach hypothesis asserts in this case that any
even number is a sum of two primes. Zeros being related to a static state, the first condition
implicitly says that there is no assignable stable equilibrium from, except for .The Zeit objekt
character is therefore irreducible in the general case. Fractal system is always dynamic. The
second conjecture leads to more subtle conclusions that we examine elsewhere.
At this stage a crucial property of zeta functions amplifies the heuristic power of the
derivation of non-integer order. This property is the functional relation:
s  1 s 
s 1 s 
 2    ( s)   2    (1  s ) . (25)
2  2 
The morphism between transfer functions and zeta functions leads one to say that the
functional relationship on zeta gives credit to a homologous relation upon the impedances. In
other respects, due to morphism in the couple of sets, another functional relation exists
between them. A brief analysis of its properties shows that this relation relates from the
disjoint sum of the dual transfer functions and in the complex plan. As shown schematically
above (Figure 5), by coupling both impedances, this relationship resets the invariance of the
energy and raises the time singularity at infinity. This statement requires a theoretical and
mathematical work currently in progress (Riot, 2013). Restoring the notherian consistency by
extension (Kan extension), the understanding of the Riemann hypothesis through the theory
of categories solves the paradox of a dynamic responses for which the energy is not an
invariant. The sum hidden behind Kan extension closes the fractional process not upon a
transport process as assumed usually, but upon a deterministic process that constitutes the
underlying (or horizon) of the process when it is performed by an experimentalist
(respectively via the test function in distribution theory). This assertion does not mean that the
diffusive process could not constitute an underlying; quite the contrary. We may not to forget
that determinism and diffusive process are strongly linked together. The issue that arises here
is the question of the functional content of this Kan‟s sum and its links with the un-causal
component of the diffusion process, which is given through component even if it is not
accessible just via a physical measurement. It is here that the prime numbers enter the lists for
giving a statistical meaning to the Kan extension, and therefore for stating the concept of Kan
entropy. Each integer number, being associated with point upon the set (seen as hyperbolic
distance), is indeed the product of prime numbers. It is written as a product in a set of primes
each of them having a position upon. Each point can be seen as both (i) the time constant of
the process which is a freedom factor of the experimentalist and (ii) the reference distance as
the hyperbolic metric upon.
Surjection
Z   Z 1

n  2
  N
  p 
ki
i
pi 
Figure 6. Simplified representation of Kan extension in the frame of the TEISI model.
This operation is a canonical surjection, that is to say an epimorphism. A part of

information contained in is then lost within the product decomposition, whereas another data
are supplied through. Clearly mixed upon the information coming from through the Kan
extension requires a statistic entropic calculus. Only merging all data coming from and its
complement provides the whole physical information. This was evident since frequency
measurement assumes, notwithstanding the choice of the exchange interface, the choice of the
reaction developed on the interface; therefore an a priori choice of a time constant (chosen
within a discrete set of numbers). This freedom limited by the natural set of integer number
must be taken into account. Even constrained experimentally the choice for observing the
geometry using a dedicated process is an epistemological choice. The operation of surjection
of the time constant upon forces the representation of the process to break his pure
thermodynamic state, and to plunge into a disordered set of primes, that, due to the mixing,
has a higher cardinal than the one of the primes; irreversible time appears at this step. From
surjection both complex structure and its complement merges pure and partitioned states,
multiplication and addition. The linear reference being always the pure states is solely given
by a prime number. One can indeed observe that certain values n (respectively may
themselves be primes. In this case the value  is pinned upon the unit (respectively n). The
surjection becomes a simple application. Then, by inversion, the application is an injection.
This shows that if is an ordinate distribution of frequencies, the correlations are only
controlled via. may then be associated with a prime numbers distribution, rightly ordinated
according to its own structure, but mixed with respect the distribution of the frequencies upon
. This function of mixing via the surjection leads the negentropic/entropic balance. It is not
only the origin of the closure of the fractional processes upon the Noether axiomes but the
source of the auto organization and of the emergence. These new phenomena is required to
close experimentally „non causal components‟ and to regularize the singularity at infinity.
Due to the use of energy to do that, it must be related to a competition between hyperbolic
distances upon and. Let us notice that the second term, strongly related to zeta function,
provides the pure stable states used as references after Kan extension. This observation
requires more comments.
7. Auto Organization and Emergence

The origin of emergence in Complex Systems (Holland, 1998) is an open question that
always baffles the scientists. Like phase transition, the phenomenon of emergence is indeed a
highly nonlinear process led by generally unknown criterions. Nevertheless we know that
emergence, characterized by holistic properties, must be sourced within local chaotic
behaviour. The Riemann zeta function remarkably illustrates this feature. The local chaos is
associated with the stochastic occurrence of the prime numbers whereas, the global behaviour
is ordered according to the mathematical well known distribution
log n
 (n)  . (26)
n
Similarly (i) the ideal gas rules PV=NRT that comes from the molecular chaos, (ii) the
WLF law of renormalization for polymers comes from the molecular properties of the
monomers, (iii) the Peukert law (Peukert, 1897) comes from the surface disorder of the
electrodes of batteries, (iv) the processes of aggregation with or without fields raise dendritic
fractal structures, etc. From the considerations given above let us seek to understand why
macroscopic rules emerge so naturally, from chaotic local process. Especially, from above
analysis, we would like to know why the competition between internal and external
referential plays a so crucial role in the emergence of auto organization. As noticed
previously the relation
R1 / 2  R cos / 2  R  R cos   R (27)

means that the fractional process represented by is less dissipative than the deterministic
process that serves as the underlying. Nevertheless it is more dissipative than the diffusive
process which bounds via d=2 the geometrical embedding. However, since the energy is not
the invariant of the fractional process, it seems impossible at this general stage to evoke a
variable that we could call entropy. For giving a meaning to his concept the analysis requires
a precise analysis of the Kan extension, that is to say the embedment of the fractal structure of
dissipation into a larger infinite set. The surjection using the set of prime number then gives a
statistical status of the concept of entropy. Moreover, the properties of scaling of allows to
define two distinct types of statistical entropy namely which has a thermodynamic and
statistical meaning of the ball of measurement, and which has only a statistical signification
evoked in the frame of the interpretation of the Kan extension. Within the standard statistical
meaning, the relation ordering the impedances leads the inequality. Because it is empirically
based on the performing of energy, the fractional process involves only local dissipative
properties (for all integer point of the transfer function) but it involves also global correlations
that can be pointed via by means of a distribution of prime adequate time constants. The
categorical analysis of Riemann‟s conjecture (Riot, 2013) then suggests designing a set of
infinite orthogonal of primes, in which natural numbers can find their locations. Such a
referential is natural for fractional process because any scaling is simply a change of unit. It
follows that, the deterministic process Z1 may use as Euclidean dynamic external referential
to source the fractional processes Z , whereas diffusive process Z1 / 2 defines in the same
referential an hyperbolic distance able to minimize the geometrical entropy. Along the
browsing upon process Z from high to low frequencies, the partition function associated
with the distribution of prime decreases dually.
FRACTAL
 CO-FRACTAL
Diffusive dissipation
Excess vs 2D Fractal dissipation
Fractional Diff-Integral
(Kan mixing extension )
Dynamic Operators
Emergence
Balance of d-Kan mixing
(scaling correlations)
Pseudo Entropic
Determinist Dissipation Fractal Production
Figure 7. Schematic representation of the problematic of „entropy‟ production on self-similar structure,

within integer environment (Fractal merged with a co Fractal). In order to balance the energy between
deterministic and diffusive process, the irreversible process leads the creation of holistic emerging
properties.
On one hand it decreases because the number of primes at stake decreases with frequency.
Therefore the configurationl entropy goes down. On the other hand, because the only entropy
that may be involved in the process, the 2d-entropy, is lower than the statistical entropy that
would result from backward (Kan extension) and forward approximation of the „Zeit Objekt‟
characterized by Z (  0) , a Kan entropy appears. It must lead holistic characteristics. In
other words, the lowering of the dissipation within the competition between two types of
metric (Euclidean and 2d-hyperbolic) contributes to the building of an auto organization
(and/or Holistic rules) during the frequency deployment of the dissipative process. This
would be a general property of all dissipative processes except for the diffusive one. The
emergence is clearly a constructive-dissipation. Up to now this issue was hidden by implicit
hypothesis of limitation of dissipative processes by diffusive transportation.
Conclusion
Far from inconsistency, the fractional diff-integral operator enlarges the epistemological
perspectives, open by integer analysis. Due to the fact that the set of rational numbers is larger
than the set of integers and according to the couple of Godel‟s theorems about
incompleteness, we cannot reduce the fractional analysis to the paradigmatic point of view of
integer dynamics. In particular, the energy is no more an invariant of space time dynamics. It
has to take into account the behaviour at infinity, that is to say the structure of the
environment. This assertion is clearly a key point of what is often called „inconsistency‟ of
fractional analysis. This apparent „inconsistency‟ is due to the fact that energy stays the main
factor of experimental performance as well as the application of the theory of fractal
distribution requires regularization via a test function. The paradoxes are due to the
incompleteness of the paradigmatic analysis with respect to the fractional one. This
incompleteness of the standard paradigms, explains the divergence characterizing the
behaviour of complex systems analysis with respect to the system able to accept a reverse
engineering. Far from the very nice artefact of renormalization technics, how to overcome in
physics the divergence between Euclidean and hyperbolic systems? The physical systems do
that naturally through what we call „emergence‟, a response of the system to bypass the
paradox led by non notherian internal structure. The theory of categories highlights this
question through the concepts of Kan extension and the related virtual sets which close
precisely the systems upon fractality as shown by Bagshi and recently Riot.
This analysis shows why the expression of a dissipative process by means of a derivative
operator of non-integer order is relevant within the meaning of fractality (recursive
structures), but also why it requires a tricky physical analysis involving the introduction of
Kan entropy. In fact, the energy being not a natural invariant of the fractional operator, the
practical use of this operator for the representation of physical systems and its extensions in
engineering (for example as in the CRONE control devices) requires to understand a subtle
problem of closure at infinity. This question can be expressed in various ways. For his part,
the author chooses to express this problem in Fourier space by considering the link between
the dynamic transfer functions, the hyperbolic spaces with angle at infinity (or equivalently
the Gromov hyperbolic groups) and finally the Riemann zeta properties in theory of numbers.
P. Riot chooses another way. He has shown recently that incompleteness can be analysed
using the category theory. In all cases, the complex systems require, for being analysed, an
embeddement, that is to say at least two points of view. The paradox of compatibility between
the experience that requires the use of energy and the theory ends into the emergence of self
similarity and vice versa. New holistic hierarchies then appear and contribute to the
emergence of new properties or geometrical features such as aggregation, dendritic growth,
the geographical contours, etc... The model TEISI contributes for his part to show why the
concept of complex time is the hearth of the problematic.
Alongside the traditional time (seen as inverse transform of the Fourier frequency) the
concept of complex time contributes to the emergence of a singular time which is precisely
the time of the holistic emergence, self-organization and innovation. The emergence is
expressed by splitting between internal and external referential during the frequencial
deployment of the process.
We have seen that all the considerations that are needed to understand the physical
meaning of the fractional differential equations naturally open upon heuristics issues. Recall
that these opportunities were very early noticed. It concerned in particular (i) the physical
sense of the wave function in quantum mechanics (d = 2) now illuminated by zeta functions
and Kan extension, (ii) the topological superconductivity; (iii) the question of the topological
gravity by inversion of space-time variable. All these questions and associated opportunities
remain yet open upon future developments.
Acknowledgment
The author thanks D. Tayurski and J. P. Badiali for deep exchanges in quantum and
complex systems physics, as well as P. Riot and I. Polubny for mathematical partnership.
References
(Angot, 1972) Angot A., Compléments de mathématiques à l’usage des ingénieurs de
l’électrotechnique et des télécommunications, Masson, Paris, 1972.
(Badiali, 2013) Badiali J.P., Toward a new fundations of statistical mechanics, arXiv:
1311.4995, V1, cond-math.stat-mech, 20 nov., 2013.
(Bagshi, 1987) Bagshi B., Recurence into topological dynamics and Riemann Hypothesis,
Acta Math Hungar, 50, p. 227-240, 1987.
(Cole, 1942) Cole K.S. and Cole R.H., Dispersion and absorption in dielectric. Direct current
characteristics, J. Phys. Chem 10, p. 98-105, 1942.
(Connes, 1994) Connes A. Non Commutative Geometry, Acad. Press. San Diego C.A. 1994.
(Derrida, 1973) Derrida Jacques, ‘Différance’ in speech and phenomena, Northwest
University Press Ilinois, 1973.
(Gemant, 1935) Gemant A. The conception of concept viscosity and application to dielectric,
Transaction of the faraday society, V.31, pp. 1582-1590, 1935.
(Ghys, 1990) Ghys E and de La Harpe P : sur les groupes hyperboliques d‟après Gromov,
(Bern 1988) 1-25, Prog. Math., 83, Birkhäuser Boston, MA, 1990.
(Holland, 1998) Holland J. H., Emergence: from chaos to order, Helix books Addison-
Wesley MA, 1998.
(Le Méhauté, 1974) Le Méhauté A, Application à la hiérarchie départementale française d‟un

modèle d‟adaptation des individus à une structure de désutilité en arbre, Annales du
Centre de Recherche de l’urbanisme Presses du CRU, Paris, 1974.
(Le Méhauté, 1977) Le Méhauté A. and Apelby J., A parametric analysis of the structure of
international energy consumption, Energy, 2, p 105-114, 1977.
(Le Méhauté, 1980) Le Méhauté A., de Guibert A., Delaye M. et Filippi C., Phénoménologie
du transfert de matière et d‟énergie en milieu présentant une similitude interne. Rapport
de Recherche, Laboratoire de Marcoussis, Division Electrochimie, 458, Décembre 1980.
(Le Méhauté, 1982a) Le Méhauté A., de Guibert A., Delaye M. et Filippi C., Note
d‟introduction à la cinétique des échanges d‟énergies et de matières sur des interfaces
fractales, C.R. Acad. Sci., Paris, t. 294, Paris, pp.835-837, 1982.
(Le Méhauté, 1982b) Le Méhauté A. et Crepy G., Sur quelques propriétés de transferts
électrochimiques en géométrie fractales, C.R. Acad. Sci., T. 294, Paris, pp.685-688, 1982.
(Le Méhauté, 1983) Le Méhauté A., Dugast A., Introduction of scaling properties in
electrochemistry: from TEISI model to lithium layered structure, Solid State Ionics, 9/10,
P 17-30, 1983.
(Le Méhauté, 1984) Le Méhauté A, Transfer process in fractal media, J. Stat. Phys. Vol. 36
N°5/6 N.Y. 665-676, 1984.
(Le Méhauté, 1986) Le Méhauté A, Frushter L., Crépry G., le Méhauté A., Batteries,
Identified Fractal objects, J. power Sources, 18, 51-62, 1986.
(Le Méhauté, 1992) Le Méhauté A., Heliodore F, Cottevieille D., Propagation des ondes en
milieux hétérogènes, Revue scientifique et technique de la défense, p 22-33, 1992.
(Le Méhauté, 2010) Le Méhauté A., El Kaabouchi A., Nivanen N., Riemann‟s conjecture and
fractal derivative, Computer and mathematics with application Chaos Solitons and
Fractals, Vol 59 (6), pp. 1610-1613, 2010.
(Mandelbrot, 1975) Mandelbrot B., Les géométries fractales, Flammarion Paris 1975.
(Mandelbrot, 1982) Mandelbrot B., The fractal geometry of nature, Freeman, San Francisco,
1982.
(Nigmatullin, 1975) Nigmatullin R. Sh., Kader B.A., Krylov V.S., Sokolov L.A.,
Electrochemical methods of transfer processes research in liquids. //Success in Chemistry
Uspekhi Khimii, v. XLIV, issue.11, pp.2008-2034, 1975.
(Nivanen, 2005) Nivanen L, Wang Q.A., le Méhauté A. „from complexe systems to
generalized statistics and algebra In fractional differentiation and application Eds A. le
Méhauté, J.A. Tenreiro-Machado, J.C. Trigeassou, and J. Sabatier U-Book, 2005
(Oldham, 1974) Oldham K.B. and Spanier J., The fractional calculus, Academic Press NY,
1974.
(Oustaloup, 1983) Oustaloup A., Systèmes asservis linéaires d‟ordre fractionnaire: théorie
et pratique. Masson, Paris, 1983.
(Peukert, 1897) Peukert W., Über die Abhängigkeit der Kapazität von der Entladestromstärke
bei Bleiakkumulatoren, Electrotech. Z. V.18, pp.287-292, 1897.
(Podlubny, 1999)Podlubny I, Fractional differential Equations, Academic press NY, 1999.
(Riot, 2013) Riot P. http://repmus.ircam.fr/mamux/saisons/saison13-2013-2014/2013-12-06
(Rovelli, 2004) Rovelli C., Quantum Gravity, Cambridge University Press, 2004.
(Rovelli, 2006) Rovelli C., Qu’est ce que le temps? Qu’est ce que l’espace?, Bernard Gilson,
2006.
(Schwartz, 1950) Schwartz L., Théorie des distributions, Herman, Paris, 1950.
(Tenenbaum, 2011) Tenenbaum G. et Mendes France M., Les nombres premiers entre ordre
et chaos, Dunod, Paris, 2011.
(Tricot, 1999) Tricot Claude, Courbes et dimensions fractales, Springer Verlag, 1999.
(Voronin, 1975) Voronin S., Theorem of the universality of Riemann zeta function, Izv. Akad.
Nauk. SSSR Ser. Matem., Vol. 39, p. 476-486, 1975, Reprinted: Math SSSR. Izv, Vol.9,
pp.443-445, 1975.
Chapter 2
DYNAMICS OF F RACTIONAL O RDER

C HAOTIC S YSTEMS
Sachin Bhalekar∗
Department of Mathematics, Shivaji University, Kolhapur, India
Abstract
This chapter deals with the fractional order generalization of the chaotic system pro-
posed by Li et al [Nonlinear Analysis: Modelling & Control, 18 (2013) 66-77]. We
discuss the dynamical properties such as symmetry, dissipativity, stability of equilib-
rium points and chaos. Further we control the chaos in proposed system and present
the synchronization phenomenon.
Keywords: Chaos, fractional order, stability

AMS Subject Classification: 26A33, 65P20, 34L30
1. Introduction
Differential equation is one of the basic tools to model the motion of a physical system.
When the solutions of these differential equations are bounded, they either settle down to
an equilibrium state or oscillates in a periodic or quasi-periodic state. If the order of the
system is three or more and the system is nonlinear then there is a possibility of another
type of solutions. Such solutions exhibit aperiodic motion for all time and never settles.
These solutions are highly sensitive to initial conditions also. Thus, it is not possible to
predict the behavior of the solution for a long time. Such systems are termed as chaotic
dynamical systems [1].
If the order of the derivative in the system is not an integer then it is called as fractional
order system [2, 3, 4]. There are two very important advantages of the fractional order
derivative over a conventional integer order derivative. The first advantage is the nonlocal
nature of fractional order derivative. The integer order derivative of a function at a point
can be approximated using the nearby values whereas the fractional derivative needs the
∗
E-mail addresses: sbb maths@unishivaji.ac.in, sachin.math@yahoo.co.in
24 Sachin Bhalekar
total history from an initial point. This nonlocality of fractional derivative (FD) is playing a
vital role in modelling memory and hereditary properties in the system. Hence the models
involving FD are more realistic than those involving integer order ones. Second advantage
is the FD is capable of modelling intermediate processes. In some physical problems such
as the diffusion of fluid in porous media, the integer order derivative cannot model the real
phenomena because the processes are intermediate. In such case, one has to consider a
model involving the intermediate (fractional) order derivative.
The generalized dynamical systems involving fractional order derivative are extensively
studied by researchers. In the seminal paper, Grigorenko and Grigorenko [5] proposed a
fractional order generalization of Lorenz system. Further, various fractional order systems
such as Chen system [6], Lü system [7], Rossler system [8]and Liu [9] system are studied
in the literature.
In this chapter, we propose a fractional order generalization of the chaotic system pro-
posed by Li et al. [17] and discuss its properties.
2. Preliminaries
2.1. Stability Analysis
We discuss the stability results of the following fractional order system
Dα1 x1 = f1 (x1 , x2 , · · · , xn ),
Dα2 x2 = f2 (x1 , x2 , · · · , xn ),
..
.
Dαn xn = fn (x1 , x2 , · · · , xn ), (2.1)
where 0 < αi < 1 are fractional orders. The system (2.1) is called as a commensurate order
if α1 = α2 = · · · = αn otherwise an incommensurate order.
A point p = (x∗1 , x∗2 , · · · , x∗n ) is called an equilibrium point of system (2.1) if fi (p) = 0
for each i = 1, 2, · · · , n.
Theorem 2.1. [10, 11] Consider α = α1 = α2 = · · · = αn in (2.1). An equilibrium point
p of the system (2.1) is locally asymptotically stable if all the eigenvalues of the Jacobian
matrix
 
∂1 f1 (p) ∂2 f1 (p) · · · ∂n f1 (p)
 ∂1 f2 (p) ∂2 f2 (p) · · · ∂n f2 (p) 
J = . (2.2)
 
.. .. .. ..
 . . . . 
∂1 fn (p) ∂2 fn (p) · · · ∂n fn (p)
evaluated at p satisfy the following condition
|arg(Eig(J))| > απ/2. (2.3)
Theorem 2.2. [12, 13] Consider the incommensurate fractional ordered dynamical system
given by (2.1). Let αi = vi /ui , (ui , vi ) = 1, ui , vi be positive integers. Define M to be the
least common multiple of ui ’s.
Dynamics of Fractional Order Chaotic Systems 25
Define
λM α1 , λM α2 , · · · , λM αn

∆(λ) = diag −J (2.4)
where J is the Jacobian matrix as defined in (2.2) evaluated at point p. Then p is locally
asymptotically stable if all the roots of the equation det (∆(λ)) = 0 satisfy the condition
|arg(λ)| > π/(2M ).
This condition is equivalent to the following inequality
π
− mini {|arg(λi )|} < 0. (2.5)
2M
Thus an equilibrium point p of the system (2.1) is asymptotically stable if the condition
π
(2.5) is satisfied. The term 2M − mini {|arg(λi )|} is called as the instability measure for
equilibrium points in fractional order systems (IMFOS). Hence, a necessary condition for
fractional order system (2.1) to exhibit chaotic attractor is [12]
IMFOS ≥ 0. (2.6)
Theorem 2.3. [18] Consider the polynomial
P (λ) = λ3 + a1 λ2 + a2 λ + a3 = 0. (2.7)
Define the descriminant for equation (2.7) as
D(P ) = 18a1 a2 a3 + (a1 a2 )2 − 4a3 (a1 )3 − 4(a2 )3 − 27(a3 )2 . (2.8)
1. If D(P ) > 0 then all the roots of P (λ) satisfy the condition
|arg(λ)| > απ/2 (2.9)
where 0 ≤ α ≤ 1.
2. If D(P ) < 0, a1 ≥ 0, a2 ≥ 0, a3 > 0, α < 2/3 then (2.9) is satisfied.
3. If D(P ) < 0, a1 > 0, a2 > 0, a1 a2 = a3 then (2.9) is satisfied for all 0 ≤ α < 1.
2.2. Numerical Method for Solving Fractional Differential Equations

The numerical methods developed for solving equations involving integer order deriva-
tives are not useful to solve those involving fractional derivatives. A modification of Adams-
Bashforth-Moulton algorithm is proposed by Diethelm et al. in [14, 15, 16] to solve FDEs.
Consider for α ∈ (m − 1, m] the initial value problem (IVP)
Dα y(t) = f (t, y(t)) , 0 ≤ t ≤ T, (2.10)

(k) (k)
y (0) = y0 , k = 0, 1, · · · , m − 1. (2.11)
The IVP (2.10)–(2.11) is equivalent to the Volterra integral equation

m−1 k t
(k) t 1
X Z
y(t) = y0 + (t − τ )α−1 f (τ, y(τ )) dτ. (2.12)
k! Γ(α) 0
k=0
26 Sachin Bhalekar
Consider the uniform grid {tn = nh/n = 0, 1, · · · , N } for some integer N and h := T /N .
Let yh (tn ) be approximation to y(tn ). Assume that we have already calculated approxima-
tions yh (tj ), j = 1, 2, · · · , n and we want to obtain yh (tn+1 ) by means of the equation
m−1
X tkn+1 (k) n
hα hα X
f tn+1 , yhP (tn+1 ) +

yh (tn+1 ) = y0 + aj,n+1 f (tj , yn (tj ))
k! Γ(α + 2) Γ(α + 2)
k=0 j=0
(2.13)
where
 α+1
 n − (n − α)(n + 1)α , if j = 0,
aj,n+1 = (n − j + 2)α+1 + (n − j)α+1 − 2(n − j + 1)α+1 , if 1 ≤ j ≤ n,
1, if j = n + 1.

The preliminary approximation yhP (tn+1 ) is called predictor and is given by

m−1
X tkn+1 (k) n
1 X
yhP (tn+1 ) = y + bj,n+1 f (tj , yn (tj )), (2.14)
k! 0 Γ(α)
k=0 j=0
where
hα
bj,n+1 = ((n + 1 − j)α − (n − j)α ) . (2.15)
α
Error in this method is
maxj=0,1,··· ,N |y(tj ) − yh (tj )| = O(hp ), (2.16)
where p = min (2, 1 + α).
3. The Model
A new 3D chaotic system proposed by Li et al. in [17] is
ẋ = a(y − x) + yz,
ẏ = (c − a)x − xz + cy,
ż = ey 2 − bz, (3.17)
where a, b, c and e are constants. We propose a modification in system (3.17) involving

fractional order derivatives as below:
Dα x = a(y − x),
Dα y = (c − a)x − xz + cy,
Dα z = ey 2 − bz, (3.18)
where 0 < α ≤ 1. Now we discuss some dynamical properties of the system (3.18).
1. Symmetry. The system is symmetric about z-axis because the transformation

(x, y, z) −→ (−x, −y, z) does not change the system.
2. Dissipativity. Consider
∂ ẋ ∂ ẏ ∂ ż
div(V ) = + +
∂x ∂y ∂z
= c − (a + b). (3.19)
The system is dissipative if c < a + b.
3. Non-generalized Lorenz system. The generalized Lorenz system (GLS) discussed

in [20] is of the form
 
0 0 0
A 0
Ẋ = + x  0 0 −1  X, (3.20)
0 λ3
0 1 0
where X = [x, y, z]T , λ3 ∈ ℜ and A is a 2 × 2 matrix with eigenvalues λ1 , λ2 ∈ ℜ

such that −λ2 > λ1 > −λ3 > 0. Since our system
     
−a a 0 0 0 0 yz
Ẋ =  c − a c 0  + x  0 0 −1  X +  0  (3.21)
0 0 −b 0 1 0 ey 2
is having different structure than the system (3), it is not a GLS.
4. Equilibrium points and their stability.

p The system (3.18) has three equilibrium

p
points viz. E0 = (0, 0, 0), E1,2 = ∓ −b(a − 2c)/e, ∓ −b(a − 2c)/e, −a + 2c .
The Jacobian matrix evaluated at an arbitrary point X = (x, y, z) is given by
 
−a a 0
J(X) =  c − a − z c −x  . (3.22)
0 2ey −b
The characteristic equation of J(E0 ) is therefore,
P = λ3 + (a + b − c)λ2 + (a2 + ab − 2ac − bc)λ + (a2 b − 2abc) = 0. (3.23)
The following expression defines the descriminant of the equation (3.23).
D(P ) = −(3a2 − 6ac − c2 )(a2 + b2 + bc − ab − 2ac)2 . (3.24)

√
The sign√of D(P ) is same as the sign of −(3a2 −6ac−c2 ) = −(a−(1+2/ 3)c(a−
(1 − 2/ 3)c)). The sufficient conditions for the stability of E0 are discussed in the
following theorem which are consequences of Theorem 2.3.
p
Theorem 3.1. (a) If 2c < a < 13/3c then the equilibrium point E0 is asymptot-
ically stable.
(b) If c = 0, b > 0 and a > 0 then E0 is asymptotically stable for α < 2/3.
28 Sachin Bhalekar
4. Numerical Simulations
We consider the system(3.18) with a = 41, c = 31, e = 3.2 and b = 4. The equilib-
rium points are now E0 = (0, 0, 0) and E1,2 = (±5.123475, ±5.123475, 21). The saddle
equilibrium E0 is useful in connecting chaotic scrolls and the scrolls will be around E1,2 .
The eigenvalues of Jacobian at E1,2 are −20.3789, 3.18947 ± 18.1059ı and the minimum
effective dimension is turned out to be 0.888994. Numerical observations also show that
the system is chaotic for α > 0.88. The stable orbits for α = 0.88 are shown in Figs. 1
whereas the chaotic attractors for α = 0.89 are shown in Figs. 2.
Figure 1a. Stable time series x(t) for α = 0.88.
Figure 1b. Stable time series y(t) for α = 0.88.

Figure 1c. Stable time series z(t) for α = 0.88.
Figure 2a. Chaotic xy-phase portrait for α = 0.89.
5. Synchronization
Though the chaotic trajectories are aperiodic in nature, two different chaotic trajectories
can oscillate together if the appropriate controller is applied [21, 22].
The chaos synchronization technique has been explored in secure communications [23]
30 Sachin Bhalekar
Figure 2b. Chaotic yz-phase portrait for α = 0.89.
and cryptographic systems [24].

Deng and Li introduced the synchronization of fractional order chaotic systems in [25].
Since then various researchers presented different techniques of synchronization of chaos
in fractional order systems. Now we utilize the nonlinear feedback control technique to
synchronize the proposed fractional order chaotic system.
Consider the drive system as
Dα x = 41(y − x),
Dα y = −10x − xz + 31y,
Dα z = 3.2y 2 − 4z (5.25)
and response system as
Dα x1 = 41(y1 − x1 ),
Dα y1 = −10x1 − x1 z1 + 31y1 + u2 ,
Dα z1 = 3.2y12 − 4z1 + u3 . (5.26)
The nonlinear feedback controllers ui ’s are chosen as u2 = x1 z1 − xz + 22y − 22y1 and

u3 = 3.2(y 2 − y12 ). If the errors in synchronization are termed as e1 = x − x1 , e2 = y − y1
and e3 = z − z1 then the error system can be obtained by subtracting the system (5.26)
Figure 3c. Chaotic xz-phase portrait for α = 0.89.
from the system (5.25) as below
Dα e1 = 41(−e1 + e2 ),
Dα e2 = −10e1 + 9e2 ,
Dα e3 = −4e3 . (5.27)
The eigenvalues of the coefficient matrix of the error system (5.27) are
−30.6629, −1.33712, −4. This shows that the errors will tend to zero as t tends to
infinity. In the figs. 3 and 4 we show the synchronization of the system for fractional
orders α = 0.80 and α = 0.98 respectively. The dashed lines in synchronized signals show
the response signals whereas the solid lines show the drive signals. The solid lines, the
dotted lines and the dot-dashed lines in Figs. 3(b) and 4(b) show the errors e1 , e2 and e3
respectively.
6. Conclusion
In this chapter, we discussed the dynamics of fractional order chaotic system. It is
observed that the nonlinear system with fractional system order less than three can exhibit
32 Sachin Bhalekar
Figure 3a. Synchronized signals x(t) and x1 (t) for α = 0.80.
Figure 3b. Errors in synchronization for α = 0.80.
chaos. Further, the synchronization of chaos is presented using nonlinear feedback control
technique.
Figure 4a. Synchronized signals y(t) and y1 (t) for α = 0.98.
Figure 4b. Errors in synchronization for α = 0.98.
Acknowledgments
Author acknowledges Shivaji University, Kolhapur for the minor research project grant
provided under the “Innovative Research Activities”.
34 Sachin Bhalekar
References
[1] K.T. Alligood, T.D. Sauer, J. A. Yorke, Chaos: An Introduction to Dynamical Systems,
Springer, New York, 2008.
[2] Podlubny, I., Fractional Differential Equations, Academic Press, San Diego (1999).
[3] Samko, S. G., A. A. Kilbas and O. I. Marichev, Fractional Integrals and Derivatives:
Theory and Applications, Gordon and Breach, Yverdon (1993).
[4] Kilbas, A. A., H. M. Srivastava and J. J. Trujillo, Theory and Applications of Frac-
tional Differential Equations, Elsevier, Amsterdam (2006).
[5] I. Grigorenko and E. Grigorenko, Chaotic dynamics of the fractional Lorenz system,
Phys. Rev. Lett. 91 (2003) 034101.
[6] C. Li, G. Peng, Chaos in Chen’s system with a fractional order, Chaos Solitons Frac-
tals 22 (2004) 443–450.
[7] J. G. Lü, Chaotic dynamics of the fractional order Lü system and its synchronization,
Phys. Lett. A 354(4) (2006) 305–311.
[8] Li, C. G. and G. Chen, Chaos and hyperchaos in the fractional order Rossler equations,
Phys. A: Stat. Mech. Appl., 341, (2004) 55–61.
[9] Daftardar-Gejji, V. and S. Bhalekar, Chaos in fractional ordered Liu system, Comput.
Math. Appl., 59, (2010) 1117–1127.
[10] M. S. Tavazoei and M. Haeri, Regular oscillations or chaos in a fractional order system
with any effective dimension, Nonlinear Dynamics, 54(3) (2008) 213–222.
[11] D. Matignon, Stability results for fractional differential equations with applications
to control processing. In: Computational Engineering in Systems and Application
multiconference, vol. 2, pp. 963-968, IMACS, IEEE-SMC Proceedings, Lille, France,
July (1996).
[12] M. S. Tavazoei and M. Haeri, Chaotic attractors in incommensurate fractional order

systems, Physica D, 237 (2008) 2628-2637.
[13] W. Deng, C. Li and J. Lu, Stability analysis of linear fractional differential system
with multiple time delays, Nonlinear Dynamics, 48 (2007) 409-416.
[14] K. Diethelm, N. J. Ford and A. D. Freed, A predictor-corrector approach for the nu-
merical solution of fractional differential equations, Nonlinear Dynamics 29 (2002)
3–22.
[15] K. Diethelm, An algorithm for the numerical solution of differential equations of frac-
tional order, Elec. Trans. Numer. Anal. 5 (1997) 1–6.
[16] K. Diethelm and N. J. Ford, Analysis of fractional differential equations, J. Math.

Anal. Appl. 265 (2002) 229-48.
[17] C. Li, L. Wu, H. Li, Y. Tong, A novel chaotic system and its topological horseshoe,
Nonlinear Analysis: Modelling and Control, 18(1) (2013) 66–77.
[18] E. Ahmed, A.M.A. El-Sayed, H. A.A. El-Saka, On some RouthHurwitz conditions for
fractional order differential equations and their applications in Lorenz, Rossler, Chua
and Chen systems, Physics Letters A 358 (2006) 1–4.
[19] E.N. Lorenz, Deterministic nonperiodic flow, J. Atmos. Sci. 20 (1963) 130.
[20] S. Celikovsky, G. Chen, On the generalized Lorenz canonical form, Chaos Solitons
Fractals, 26 (2005), 1271
[21] L. M. Pecora and T. L. Carroll, Synchronization in chaotic systems, Phys. Rev. Lett.
64(8) (1990) 821.
[22] L. M. Pecora and T. L. Carroll, Driving systems with chaotic signals, Phys. Rev. A 44
(1991) 2374.
[23] R. Hilfer, Applications of Fractional Calculus in Physics, World Scientific, USA,

2001.
[24] R. He and P. G. Vaidya, Implementation of chaotic cryptography with chaotic syn-

chronization, Phys. Rev. E 57(2) (1998) 1532.
[25] W. H. Deng and C. P. Li, Chaos synchronization of the fractional Lü system, Physica
A 353 (2005) 61–72.
Editors: Roy Abi Zeid Daou and Xavier Moreau c 2015 Nova Science Publishers, Inc.
Chapter 3
P RESSURE C ONTROL OF CNG E NGINES

BY N ON - INTEGER O RDER C ONTROLLERS :
A N EW T REND IN A PPLICATION OF F RACTIONAL
C ALCULUS TO AUTOMOTIVE S YSTEMS
Paolo Lino∗and Guido Maione†
Dipartimento di Ingegneria Elettrica e dell’Informazione,
Politecnico di Bari, Bari, Italy
Abstract
The massive use of electronic control in automotive vehicles improved performance,
comfort, safety and reduced pollutant emissions and consumption. In particular, the
accurate control of the fuel injected into cylinders allowed the common rail fuel in-
jection system to increase engine performance while reducing emissions, noise and
fuel consumption. In this context, compressed natural gas (CNG) engine systems can
further reduce emissions to adhere to environmental policy regulations. However, the
injection process is strongly non-linear, time-variant and highly coupled, so suitable
control systems must be designed to guarantee the desired performance.
This chapter describes how to synthesize and realize non-integer order controllers
for pressure control in common rail injection systems of CNG engines. The realization
is relatively simple and cheap, as required by the industrial application. Namely, not
only low sensitivity to parameter variations and load disturbances must be achieved,
but also a limited cost with respect to implementation by consolidated PID controllers.
The CNG common rail injection system includes: a tank storing high pressure gas;
a main chamber and a control chamber; a solenoid valve; an electronic control unit; a
common rail and electro-injectors. The tank feeds the downstream circuit. The main
and control chambers are separated by a moving piston. Both chambers receive fuel
from the tank and send it to the common rail, which is a constant volume accumulator
connected to the electro-injectors. The inlet flow to the main chamber is regulated by
a shutter that is integral with the piston, whose position depends on the equilibrium of
the pressures acting on its surfaces. Adjusting the pressure in the control circuit by the
∗
E-mail address: paolo.lino@poliba.it
†
E-mail address: guido.maione@poliba.it
38 Paolo Lino and Guido Maione
solenoid valve regulates the main chamber inflow. Moreover, as the main chamber and
the common rail have almost equal pressures, accurate metering of the injected fuel is
allowed by setting the injection timings at the same time.
This work reports recent advancements in the design and simulation of switched
fractional order PI controllers, in which the integral action is of non-integer order.
Performance, robustness and disturbance rejection are tested by simulation of virtual
prototypes based on non-linear models. The basic idea is to perform a loop-shaping
of the open-loop transfer function to obtain frequency domain performance specifica-
tions and achieve an optimal feedback system. To this aim, the fractional integrator
is profitably used and robust stability of the closed-loop system is guaranteed by a
D-decomposition method. There are several benefits of the design approach. Closed
formulas determine the controller gains by frequency domain specifications and can be
used for an automatic synthesis of the controller. Moreover, the realization of the non-
integer operator is by an efficient approximation method that prevents numerical prob-
lems and leads to a rational transfer function characterized by interlaced minimum-
phase zeros and stable poles, so that a reduced approximation error is obtained and
easy implementation is possible. To conclude, the non-integer order controllers al-
low higher accuracy in metering the injected fuel and promptness in setting the rail
pressure to the desired reference values.
PACS: 02.30.Yy, 05.45-a., 07.05.Dz, 88.85.mf

Keywords: Compressed natural gas engines, injection systems, non-integer-order con-
trollers, fractional-order controllers, fractional-order proportional-integral controllers, loop-
shaping, approximation of irrational operators, continued fractions, zero-pole interlacing,
control systems, stability, frequency-response methods
AMS Subject Classification: 93C, 93D, 93C80
1. Introduction: A Short Literature Review

The automotive industry is undergoing a deep transformation due to the massive intro-
duction of electronic control systems within vehicles. Electronic control has helped practi-
tioners to improve performance, comfort and safety and to reduce pollutant emissions and
fuel consumption. Among the control-engineered systems, the fuel injection system re-
ceived great benefits from electronic control. Specifically, the advent of the common rail
injection system led to improved engine performance and reduced emissions, noise and fuel
consumption by accurate control of the amount of fuel injected into the cylinders. Recently,
to further reduce emissions and adhere to environmental policy regulations, innovative com-
pressed natural gas (CNG) engine systems have also been studied, designed and analyzed.
Suitable control systems must guarantee the desired performance of the injection process,
which is non-linear and time-variant, even in the presence of large variations and uncertain-
ties of parameters and working points.
This chapter concerns a new and interesting application of the paradigm of fractional
calculus (e.g., see [3, 47, 55] for mathematical background) in this context. The authors’
interest in this subject is motivated by recent research results [27, 28] that are focused on a
methodology for designing non-integer (fractional) order controllers of the pressure in the
common rail injection system of CNG engines [23]. For high engine performance, indeed,
Pressure Control of CNG Engines by Non-integer Order Controllers 39
it is necessary to regulate the injection pressure for properly metering the air-fuel mix-
ture, notwithstanding the large variations of working conditions due to gas compressibility.
Moreover, control also has to adapt the working point of injection to changes of the power
request, speed, load, etc. To this aim, a suitable model is necessary for analyzing and con-
trolling the process dynamics, and both the controllers’ performance and their robustness
must be evaluated and analyzed.
The motivation for employing non-integer order controllers and their origin can be
traced back to the famous and seminal works by Bode, Manabe, and Tustin [7, 40, 65].
Bode found that a non-integer (fractional) order integrator is the ideal open-loop transfer
function for designing feedback amplifiers. Manabe investigated the dynamics of flexible
spacecraft structures and used non-integer integration for modeling purposes [40]. Tustin
worked on robust control of massive objects, approximated a fractional integrator of order
0.5, and achieved a phase margin of 45◦ over a sufficiently wide frequency range around
the gain crossover frequency. However, a renewed interest in fractional calculus applica-
tions has only been observed in recent decades. This interest grew especially thanks to
successful results in controlling car suspensions and other systems by using the CRONE
(Commande Robuste d’Ordre Non Entièr) control approach developed by the French team
headed by Oustaloup [48, 49, 51–53]. Besides the several generations of the CRONE ap-
proach [53], there are many ways to design and tune non-integer order controllers for auto-
motive, mechatronics and industrial systems (e.g., see [5,9,10,13,21,30,42,43,54,63,64]).
These and many other works let researchers understand that non-integer (fractional)
order differential and integral operators could offer more flexibility to the familiar PID con-
trollers. Derivatives and integrals of non-integer order were proposed in P I λ D µ -controllers
[56,57], where λ and µ are the non-integer orders of integration and differentiation, respec-
tively. However, one can use any non-integer order (a real or even complex number) [50].
Another way is to consider fractional power poles [10,11] or transfer functions powered to a
non-integer order, from which the term fractional order proportional-integral-derivative (or
FOPID) controllers is derived [43,58]. To synthesize, the term Fractional Order Controllers
(FOCs) is commonly used because the integral and derivative actions are often of a frac-
tional order. The common factor among all solutions is the extension of the potentialities
of the common PID controllers by offering more design degrees of freedom and enhanced
robustness.
Namely, FOCs may overcome PID or more complex integer order controllers because
they may guarantee better closed-loop performance and higher robustness for parametric
variations, both in the case of plants described as integer order systems and especially in
the case of plants characterized by non-integer order models [13]. Therefore, the potential
impact of FOCs in industrial loops is high. Recent results justify the efforts to exploit these
potentialities [43], even if there is no established and widely accepted method for designing
or tuning FOCs. In fact, there are still many issues to consider, especially for applications
requiring a relatively simple and cheap realization of non-integer order controllers. Namely,
the developed solutions should not only be less sensitive to parameter variations and load
disturbances, but should also have economic benefits with respect to standard PID con-
trollers of consolidated practice and limited cost. In conclusion, for automotive industries
it is important to obtain both efficient and low-cost controllers.
This chapter describes an approach for rail pressure control in injection systems of
CNG engines. The approach combines a gain scheduling technique and a systematic de-
sign methodology of FOCs to be applied for different working conditions of the injection
system. After the linearization of a non-linear state-space model, the process is modeled
as a first order lag plus a time delay (FOLPD) system. In particular, the chapter shows the
benefits provided by fractional order proportional-integral (FOPI) controllers, in which the
integral action is of non-integer order and is added to a standard proportional action. The
basic idea is to perform a loop-shaping [26] of the open-loop transfer function to obtain fre-
quency domain performance specifications and achieve an optimal feedback system [19].
The loop-shaping by the fractional integrator is inspired by the classical symmetrical opti-
mum principle. Deadtime in the plant is also taken into account. Gain scheduling is used
to adapt the FOC parameters to the various working points, then to switch between differ-
ent controllers. Moreover, robust stability of the closed-loop controlled system during the
switching between working points is guaranteed by a D-decomposition method [46] and
effective simulation tools.
To show the effectiveness of the approach, performance and robustness are tested by
simulation of virtual prototypes based on non-linear models and developed by commercially
available software.
2. The Injection System

The CNG common rail injection system includes the following main components (see
Figure 1): a tank storing high pressure gas that goes up to 200 bars; a pressure regulator
that is electronically driven by a solenoid valve and an electronic control unit (ECU); a fuel
metering system, consisting of a common rail and electro-injectors. The tank is a constant
volume steel cylinder that feeds the downstream circuit, so that pressure in the tank slowly
decreases because of the injection flow. The pressure regulator includes a main chamber
and a control chamber, which are separated by a moving piston. Both chambers receive
fuel from the tank, then send it to the common rail, which is a constant volume accumulator
connected to the electro-injectors. The inlet flow to the main chamber is regulated by a
shutter that is integral with the moving piston, whose position depends on the equilibrium
of the pressures acting on its surfaces. Therefore, adjusting the pressure in the control circuit
by means of a solenoid valve allows the regulation of the main chamber inflow. Moreover,
the main chamber and the common rail have almost equal pressures (p2 in Figure 1), so that
control of the rail pressure is indirectly achieved by operating on the valve. Then, the ECU
accurately meters the amount of injected fuel by operating on the valve control circuit to
determine the rail pressure and by setting the injection timings.
2.1. Modeling the CNG Injection System

The state-space model of the CNG injection system is derived by the Eulerian approach.
Two control volumes are considered, i.e., the regulator control chamber and the rail circuit.
These volumes have a uniform, time-varying, pressure distribution: p1 is the pressure in the
control circuit, p2 is the pressure in the main chamber and common rail. The tank pressure
p3 = ptk is considered an input rather than a state variable, because its measure is always
available on board as it is related to the fuel supply. Pressure ptk is approximately constant
fuel flow
from the tank ECU
p1
SOLENOID
VALVE
p1
CONTROL
CIRCUIT COMMON PRESSURE
SENSOR
PISTON RAIL
MAIN
CHAMBER p2 p2
fuel flow
from the tank
SHUTTER ELECTRO-
INJECTORS
PRESSURE REGULATOR
DISCHARGING
MANIFOLD
Figure 1. Scheme of the CNG common rail injection system.
within a large time interval, even if the fuel reservoir is slowly decreasing. Moreover,
it is likely to assume that injection and rail pressures are equal, so that injectors are not
modeled apart and are included in the rail circuit as electronic control valves. Finally, the
temperature is assumed to be constant in the whole injection system, so that the system
dynamics is completely defined by the pressure variations in the control chamber and the
rail circuit.
State-space equations of control volumes derive from the perfect gas law and can be
written in terms of the continuity equation [23]:
RΘ
ṗ = (ṁin − ṁout ) (1)
V
where p is the pressure of the considered control volume, R is the gas constant, Θ is the
temperature, and ṁin and ṁout are the input and output mass flows, respectively, the sum
of which has to be equal to the overall mass variation in the control volume V . Possible
volume changes due to mechanical part motions are disregarded, since they do not consider-
ably affect the pressure dynamics. Mass flows through control chamber and regulator inlet
orifices are considered to be isoentropic transformations, where an output/input pressure
ratio pout /pin ≤ 0.5444 is assumed (critical flow condition resulting in a sonic speed flow).
Applying the momentum equation leads to the following formula that holds for sonic speed
flow [68]: s
k+1
2 k−1
ṁin = cd ρin A k R Θ (2)
k+1
where A is the area of the inflow section, ρin is the intake gas density, k is the gas elastic
constant and cd is a discharge coefficient accounting for non-uniformity of the mass flow
rate.
In the case of a fluid-dynamic stationary flow, the following formula is used [68]:
p
ṁin = cd cL An,min ρout (pin − pout ) (3)
where cL takes into account the effect of kinetic energy losses in the nozzle minimal section
An,min . Equation (3) also assumes that no reversal flow occurs. This is the case of the flow
between the control chamber and rail circuit.
The inflow section of the pressure regulator, i.e., A in (2), changes with the axial dis-
placement hs of the shutter and piston [23]. Then, to compute A, the shutter and piston
dynamics are determined by applying Newton’s second law of motion to the forces acting
upon each of them. If the viscous friction and the inertias of the piston and the shutter are
disregarded with respect to the large hydraulic forces, then the force balance is:
X
pi Si − ks hs + Fso − Fc = 0 (4)
i
where Si is area of the surfaces on which pressure pi acts. It is assumed that pressure
gradients are applied to the flow minimal sections. Moreover, ks is the spring constant,
Fso is the spring pre-load, i.e., the force applied when the shutter is closed, and Fc is the
assumed constant coulomb friction. Hence, equation (4) provides hs and then A.
The shutter displacement of the electro-hydraulic valve regulates the incoming flux in
the control chamber. Since the shutter inertia is negligible, it is assumed that the inflow
section can be completely opened (As = As,max ) or closed (As = 0), depending on the ac-
tual driving current of the valve (energized/not-energized circuit). Consequently, the value
of the inflow section can be calculated as As = ETsv · As,max , where ETsv is a square
signal defining the energizing phase of the solenoid valve. ETsv takes values in {0; 1}
representing the valve energizing conditions.
In this model, the injector opening and closing transients are disregarded. Hence only
two conditions can occur: injectors are closed or opened with a flow section Ainj . In this
way, system order and computational effort are reduced, without introducing a considerable
error. Thus, the injector flow section can be expressed as ETinj · Ainj , where ETinj is a
square signal of amplitude {1; 2; . . .; n}, where n is the number of injectors. The amplitude
depends on the number of simultaneously opened injectors and on the signal period that
varies with the engine speed. Since critical flow condition always holds true, the injection
mass flow can be calculated by applying (2).
Now define two state variables, i.e., the control chamber pressure (x1 = p1 ) and the
rail pressure (x2 = p2 ). Moreover, assume as inputs the signals driving the solenoid valve,
u1 = ETsv , and the injectors, u2 = ETinj , respectively. Then the CNG injection system
resulting from (1)-(4) is described by the following second order non-linear state-space
model [23]:
 p

 ẋ 1 (t) = c 11 p tk (t) u 1 (t) − c 12 x2 (t) [x1 (t) − x2 (t)]
ẋ2 (t) = c21 ptk (t) [c24 x1 (t) − c25 x2 (t) − c26 ptk (t) − c27 ] − c22 x2 (t) u2 (t)+ (5)

 p
+ c23 x2 (t) [x1(t) − x2 (t)]
where cii are constant coefficients that depend on the fluid-dynamics gas characteristics and
on the system geometrical parameters.
Now, the state-space equations (5) are linearized around different equilibrium points
(x̄, ū), so that each tuning of the controller parameters refers to a current working point.
This simplification is justified because control must keep the pressure close to a reference
value, depending on the working conditions set by the driver power request, engine speed
and load. Linearization with respect to x̄ and ū yields: δ ẋ(t) =
A δx(t)+ B δu(t), where
u1 (t) x1 (t)
δx(t) = x(t) − x̄ and δu(t) = u(t) − ū, with u(t) = , x(t) = .
u2 (t) x2 (t)
Choosing δx2 (t) as output gives δy(t) = C δx(t), with C = [0 1]. The following
matrices A = [aij ] and B = [bik ] are obtained:
−c12 √ x̄2 −c12 √x̄1 −2 x̄2

 
2 x̄2 (x̄1 −x̄2 ) 2 x̄2 (x̄1 −x̄2 )
A=  (6)
c21 c24 p̄tk + c23 √ x̄2 −c21 c25 p̄tk − c22 ū2 + c23 √x̄1 −2 x̄2
2 x̄2 (x̄1 −x̄2 ) 2 x̄2 (x̄1 −x̄2 )
" #
c11 p̄tk 0
B= (7)
0 −c22 x̄2
The Laplace transform of the linearized model yields:
a12 b11

 X1 = X2 + U1 = G11 X2 + G12 U1


s − a11 s − a11
(8)
a21 b22
 X2 = X1 + U2 = G21 X1 + G22 U2


s − a22 s − a22
where, with some notational abuse, symbol δ has been dropped and the dependence of
X1 , X2 , U1 , U2 and of Gij on the complex variable s has been omitted. Equations (8)
show that the equilibrium tank pressure p̄tk greatly affects the effectiveness of the control
action through G12 , while the injection process can be considered to be a disturbance on the
rail pressure. Clearly, the pressure dynamics in the control chamber and rail are strongly
coupled. From (8), it is straightforward to obtain the relationship between U1 and X2 :
G12 · G21 a21 b11
X2 = U1 = 2 U1 (9)
1 − G11 · G21 s − (a11 + a22 )s + a11 a22 − a12 a21
Then, by putting U = U1 and Y = X2 , the input-output transfer function is given by:
Y (s) a21 · b11 1
= · 1 1
(10)
U (s) s1 · s2 (1 − s1 s)(1 − s2 s)
in which the constant coefficients√depend on the working point and the transfer function
poles are s1/2 = 0.5[a11 + a22 ± ∆], with ∆ = (a11 − a22 )2 + 4a12 a21 . It always holds
that a11 , a22 < 0 and a12 , a21 > 0, then ∆ > 0. Furthermore,√{a21 , |a22|} {|a11 |, a12}
and (a11 − a22 )2 4a12 a21 . Then s1 = 0.5[a11 + a22 + ∆] dominates the transient
response, independently of the working point, whereas s2 ≈ a22 . To sum up, if the smaller
time constant is neglected and a time delay τ is introduced to take into account the pressure
propagation from the main chamber to the common rail, a family of models described by a
first-order lag plus a time delay (FOLPD) system is obtained:
Ke−τ s
Y (s) = Gp(s) · U (s) = · U (s) (11)
1 + Ts
where K = a21 · b11 /(s1 · s2 ) and T = −1/s1 . The delay τ of pressure propagation can be
experimentally determined.
To determine K and T , (5) is manipulated to evaluate equilibrium pressures. However,
the valve and injectors’ driving signals, ū1 and ū2 , cannot be directly plugged into the
model equations because they result from modulation of discrete values. For this reason,
their mean values within an injection cycle are considered instead of the instantaneous
values. In particular, ū1 is the driving current duty cycle, while ū2 = 4 · tj · υ/120, i.e., the
ratio between the injection time interval tj and the duration of the whole injection cycle,
multiplied by the number (four) of injectors and by the engine speed υ (in rpm). Under this
assumption, the linear model coefficients of A and B are determined as follows.
First, some considerations on operation modes are useful. As shown by (5) and by
the coefficients of the linearized system, operation strongly depends on the tank pressure
p̄tk . The ECU takes into account the driver power request, speed and load, then it specifies
the proper amount of injected fuel by setting the rail pressure reference and the injection
timings. Namely, the injection flow rate only depends on the rail pressure because critical
flow conditions hold true [23]. Then, the ECU stores static look-up tables that specify the
rail pressure and injection timings, in terms of opening time interval tj (as a function of
load and rail pressure) and frequency 4tj υ/120 (as a function of speed). This guarantees
that the optimal injection profile is set for each working point that refers to specific values
of tank pressure, rail pressure and injection timings.
Therefore, a set of linearized models can be determined for each value of p̄tk . As
a matter of fact, p̄tk depends on the fuel supply, so it is almost constant in a large time
interval compared to the dynamics of x1 and x2 . Moreover, p̄tk affects the efficacy of the
control action u1 because it is included in coefficient b11 appearing in G12 . Each linearized
model in the set related to p̄tk depends on the rail pressure x̄2 and the injection timings
ū2 , as specified by the ECU. Then, the control input ū1 , which is able to keep x̄2 , and the
resulting control pressure x̄1 can be computed from (5) by setting derivatives to 0, and by
fixing p̄tk , x̄2 , and ū2 . In particular, the first equation in (5) gives the control chamber
pressure:
(c11 p̄tk ū1 )2
x̄1 = x̄2 + . (12)
c212 x̄2
Then, solving the equilibrium condition resulting from the second equation of (5) gives ū1 :
c211 c21 c24 3 2 c11 c23

p̄ ū + p̄tk ū1 +c21 (c24 x̄2 −c25 x̄2 −c26 p̄tk −c27 )p̄tk −c22 ū2 x̄2 = 0 (13)
c212 x̄2 tk 1 c12
Finally, given ū1 , x̄2 , and x̄1 , one can determine elements of A and B that affect K and T .
2.2. Scheduling of Parameters

Since the system has several equilibrium points, the parameters K and T in (11) assume
different values to which the FOC design will refer. A gain scheduling approach is adopted
to adjust the controller settings to variations in the CNG system parameters. Each equilib-
rium has an associated triple (K, T, τ ). More specifically, for each value of p̄tk , two 2-D
maps can be obtained for K and T by varying the rail pressure and the injection timings
at the same time in the whole working range. To synthesize, a fixed value of p̄tk is chosen
to select the proper 2-D look-up tables, which are used to compute K and T based on the
equilibrium actual rail pressure x̄2 and injection timings ū2 . The choice of equilibrium con-
ditions, and thus of the scheduling strategy, is based on a sensitivity analysis of the linear
model coefficients. Clearly, the larger the number of operating conditions for the schedul-
ing strategy, the larger the amount of memory to store controller parameters. However, the
computational effort does not increase. Namely, the complete set of controller parameters
is computed off-line and stored in static look-up tables as a function of the chosen operating
conditions.
3. The Control Strategy: Switching of Fractional Order

Controllers by Gain Scheduling
To take into account different equilibrium points of the controlled system and to achieve
a good trade-off between performance and robustness, a possible solution is to implement
switching among fractional order controllers by a scheduling strategy. This is a natu-
ral extension of the common approach in automotive applications that frequently employ
scheduling of different integer order PI controllers [14, 45, 62]. However, a PI controller is
employed for all operating conditions and its tuning is obtained by conventional rules like
Ziegler-Nichols [2, 15, 23, 66]. In what follows, the controller parameters depend on the
CNG engine working conditions. In other words, for each of these equilibrium points, there
is a different fractional order controller that is properly designed to achieve performance
and robustness specifications (see Section 4).
Scheduling works as follows. Firstly, the working point is established by the rail pres-
sure and by the average duration, say I, of the injection process, for each value of the
approximately constant tank pressure. More specifically, I depends on the injector open-
ing time intervals and on the engine speed. Secondly, the values of the rail pressure and
I are considered for the new working point that is required to be reached, and the varia-
tion between the two working points is computed. Thirdly, the controller is designed with
reference to the set (K, T, τ ) of parameters associated with the new equilibrium point.
This method is used when the variations between the two operating points are bounded
by 2 bar for the rail pressure and by 6 seconds for I. However, if the variations are higher
than these limit values, then several intermediate rail pressure and I reference values are
considered. Then, several sets (K, T, τ ) are determined and several associated controllers
are designed. In this way, maps of controller gains are built for the different equilibrium
points.
As regards stability of gain scheduling, the pressure reference and I are the scheduling
variables subject to step changes between the equilibrium points. Then, the amplitudes of
these steps must be limited, because each local linear fractional order controller guarantees
stability only close to the equilibrium point. If this holds true, then the initial equilibrium
point is in the region of attraction of the changed equilibrium point, and a smooth transition
is achieved [20].
If variations of variables are fast, then assessing stability in switching for classical gain
scheduling is impossible. In this case, only simulation can be used [59], for which the reader
can refer to Section 5. There are recent design approaches, such as those based on linear
parameter-varying systems (LPV) [44], that explicitly take into account time dependency
of scheduling variables, so that global closed-loop stability is a priori achieved [45]. How-
ever, the number and variation rate of scheduling parameters, as well as complexity of the
non-linear model, make the LPV framework inappropriate for the considered application.
Namely, analysis and design are complex and the computational effort is high [45].
To better analyze closed-loop stability with respect to variations in working conditions
of the injection process, one may refer to the classical D-decomposition methodology [46]
that is well known for designing integer order controllers [1] but that was also recently
applied to FOPID controllers operating on fractional order systems with delays [16, 18, 29].
Namely, this approach determines all the stabilizing controllers, i.e., the set of controller
parameters that guarantee a stable closed-loop system. A stability domain, say D, is defined
in the so-called parameter space and the designed controller is associated with a point inside
this space. If the point is far from the boundaries of the stability domain, then closed-loop
stability is obtained for (small) bounded variations of the controller gains.
To extend the D-decomposition methodology to the case of fractional order controllers,
consider the procedure shown by [16, 29]. The starting point is the characteristic equation
of a closed loop containing a FOPI controller. If the open-loop transfer function is
K KI (1 + TI sν ) −τ s
G(s) = e (14)
(1 + T s) sν
then the closed-loop transfer function is
K KI (1 + TI sν ) e−τ s
F (s) = (15)
(1 + T s) sν + K KI (1 + TI sν ) e−τ s
where ν is the fractional order of integration, TI = KP /KI , and KP and KI are the
proportional and integral gain, respectively. To analyze closed-loop stability, consider the
solutions of the fractional order characteristic pseudo-polynomial equation
E(s) = (1 + T s) sν + K KI (1 + TI sν ) e−τ s = 0 (16)
Then, if all solutions lie in the open left-half of the s-plane (LHP roots), then the closed-loop
system is BIBO stable [8, 16, 18].
So, if (KP , KI , ν) ∈ D (the set of all stabilizing controllers in the parameter space),
then all closed-loop roots are LHP. The domain D is defined by boundaries: the real root
boundary (RRB), the infinite root boundary (IRB), and the complex root boundary (CRB)
[1, 16]. The RRB is associated with the solutions of E(s = 0) = 0:
K KI = 0 ⇒ KI = 0 (17)
The CRB comes from E(s = jω) = 0
(1 + T jω) ω ν (C + jS)+
+ K KI [1 + TI ω ν (C + jS)] [cos(x) − j sin(x)] = 0 (18)
with C = cos(θ), S = sin(θ), θ = π2 ν, x = ωτ . Taking the real and imaginary parts of the
complex equation (18) leads to the following solutions:
ω ν (sin(x) + ωT cos(x)) (ωT S − C) sin(x) − (S + ωT C) cos(x)
KI = , KP = (19)
KS KS
which can be used to plot a curve in the 2-D-space (KP , KI ), when ν is fixed and ω is
varied from 0 to ∞, since the gains are functions of ω. If a different pair of controller gains
is chosen, analogous curves can be drawn in the 2-D-spaces (KP , TI ) or (TI , KI ).
The IRB for s → ∞ does not exist, because the condition αn ≤ βn provided in [16] is
not satisfied, where αn and βn are, respectively, the integer (in general fractional) orders of
the denominator and numerator polynomials of Gp(s) in (11).
To simplify the identification of the complete stability domain D, two steps can be
followed. Firstly, one of the parameters (e.g., the order ν) is fixed so that the RRB and
CRB boundaries are determined inside the plane defined by the remaining two parameters
(e.g., the (KP , KI )-plane). Secondly, the fixed parameter is varied and the complete 3-D
stability domain is obtained by sweeping over all the admissible values of the parameter
fixed initially.
If the analysis focuses on the plane (KP , KI ), then the boundaries divide it into unstable
and stable regions that can be checked by a test point. If a region is stable, then it does not
contain any closed-loop root in the closed right-half of the s-plane (RHP root); if a region is
unstable, then it includes some RHP root. For the stability check, algorithms were proposed
by [18] and [41].
For example, Figure 2a shows the stability regions obtained for ν = 1.3, 1.4, 1.5, 1.6,
when a FOPI controller is designed according to the procedure of Section 4, with reference
to an equilibrium point defined by K = 165 bar and T = 1.74 s. In particular, the equi-
librium points are determined by (K, T )-values corresponding to values of the rail pressure
p2 and injection timing tinj , for an engine speed of 2500 rpm. Note that the τ parameter
is approximately constant in the whole operating range, if an accurate model of the CNG
injection system is considered. Moreover, for a different equilibrium point, a different FOPI
controller and a different stability region are defined. Figure 2b shows the case of K = 127
bar and T = 1.97 s. The values of the gains in Figures are indicated but for a multiplying
factor of 10−2 .
Finally, it is remarked that, for each value of ν, the design point P (marked with ×
in the Figure 2) lies in the stability region and belongs to a peculiar curve; this curve is
called a “relative stability curve” and is obtained by the technique of the phase margin
tester [16, 29]. This curve is associated with the specified phase margin P Ms and, for each
point, provides a possible gain crossover frequency. Then, the point × determined by the
designed values of KP and KI corresponds to the specified gain crossover frequency, ωc .
Moving the design point along the relative stability curve changes the controller gains, by
keeping the same P Ms while changing ωc . Moreover, another curve (not shown) can be
defined for a specified gain margin GMs and passes through a design point that is associated
with the phase crossover frequency.
4. Fractional Order Control Design

In this section, the focus is on the P I ν type of controllers, because they recently ob-
tained promising results in mechatronics and automotive control systems. Namely, a sys-
tematic design methodology has been applied [24, 26] and is extendable to a wide class
of systems. Moreover, this kind of controller is the closest one to PI controllers that are
typically employed in industrial contexts [4]. Hence it is quite natural to compare the PI
4
10
4 10
ν = 1.6
P ν = 1.3 P ν = 1.4 ν = 1.6 P3 ν = 1.3 P ν = 1.4
4 4
3
P5 ν = 1.5 P6 ν = 1.6 CRB curves ν = 1.5

P ν = 1.5 P ν = 1.6 CRB curves
5 6
3 ν = 1.5 3 ν = 1.4
10 10
ωc = 3.2652 ν = 1.4 ωc = 2.8934 ν = 1.3
ν = 1.3
ν = 1.6
ν = 1.6
KI 2 KI 2 ν = 1.5
10 Stability Regions 10 Stability Regions
ν = 1.5
P ν = 1.4
P6 6 ω
ω
ν = 1.4 P5
P5
P4 P Relative Stability Lines
1
Relative Stability Lines 1 4 ν = 1.3
10 (phase margin = 60°) 10 (phase margin = 60°)
P3 P3
ν = 1.3
RRB line: KI = 0 RRB line: K = 0

I
0 0
10 10
−5 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50
KP KP
(a) (b)
Figure 2. Design of FOPI controllers (uc = 5.7, P Ms = 60◦ ) for two working points and
position of design points P in stability regions with respect to stability boundaries (CRB
curves) and relative stability curves: (a) K = 165 bar, T = 1.74 s, (b) K = 127 bar,
T = 1.97 s.
to the flexible and robust P I ν , with its high performance due to the additional freedom
of choosing the non-integer order ν of integration. The limited complexity of P I ν could
also help practitioners to make comparisons with various tuning rules of PI controllers and,
above all, to promote their acceptance of the fractional order control paradigm. Finally,
as shown below, the choice of the free controller parameters is determined by closed-form
formulas that combine a low computational complexity with analytic satisfaction of design
specifications (robustness and performance).
The design is guided by basically two ideas:
• the optimality of the closed-loop system, to obtain a system which is a good approx-
imation of an optimal system in a frequency band of interest [19];
• the loop-shaping of the open-loop frequency response, to obtain high robustness to

parametric variations and uncertainties, in a sufficiently wide frequency range.
The design obviously requires a realization phase. Namely, implementation of FOC

requires approximating the irrational operators that are basic units in Laplace representation
of FOC. Then, a rational transfer function must be obtained by a proper approximation
technique. Among the available techniques, a continued fractions approach is chosen that
shows several benefits:
a) good convergence properties and reduced approximation errors;
b) stability and minimum-phase characteristics, which are very important for control
applications;
c) interlacing between real poles and real zeros of the rational transfer function.
These properties are verified both in the analog and in the discrete domain. The implemen-
tation of the approximating transfer functions can be achieved by using interface equipment
allowing use of the Matlab/Simulink environment or micro-controllers or low cost compo-
nents.
4.1. Design Procedure

The design of robust FOPI controllers for each operating point is based on a loop-
shaping approach. By applying relatively easy closed-form formulas, the methodology
directly relates performance and robustness specifications to the values of the controller
parameters. More specifically, to achieve robustness to gain variations in a sufficiently
wide frequency range around the crossover [26], an integrator of non-integer order ν is
particularly suitable. Namely, it leads to a nearly flat phase diagram and a constant slope of
the magnitude diagram of the open-loop gain in that range. Moreover, to pursue optimality
of the feedback system [19], the open-loop frequency response is so shaped that the gain is
high at low frequencies and rolls off at high frequencies to avoid stability problems [61]. A
good input-output tracking is obtained in a specified bandwidth.
The proposed design procedure was conceived for controlling plants with transfer func-
K
tion Gp (s) = s (1+T s)
[26]. In particular, the background idea was inspired by the sym-
metrical optimum method that is often used to tune integer order PI controllers for electric
drives [39]. Subsequently, the effect of time delay τ has been added [24, 25]. To cover a
K −τ s
wide range of applications, processes characterized by Gp (s) = 1+T se have also been
considered [27, 28].
Here, the controlled plant of a unitary feedback control system is considered, in which
sensor dynamics are neglected for the sake of simplicity, and with a model defined by:
K
Gp(s) = e−τ s (20)
1+T s
where K is the static dc-gain, T is the time constant of the dominant dynamics and τ is the
dead time of the injection system. Moreover, the system includes a FOPI controller that is
characterized by a standard proportional element and an integral action of non-integer order
1 < ν < 2:
KI KI
Gc (s) = KP + ν = ν (1 + TI sν ) (21)
s s
with TI = K KI and where KP and KI are the proportional and integral gain, respectively.
P
In this way, the open-loop gain G(s) = Gc (s) Gp(s) is:
K KI (1 + TI sν ) −τ s
G(s) = e (22)
sν (1 + T s)
The non-integer order is chosen 1 < ν < 2 so that an integer unitary integration s−1 is
included to allow compensation of disturbances on the plant. The reason for this choice is
that, if the non-integer operator sν with 0 < ν < 1 is approximated by a rational transfer
function, then a non-zero dc-gain would be obtained by this realization and a steady-state
error would occur. Instead, by using 1 < ν < 2, an integer integrator (with order equal to
1) and a non-integer integrator (with order λ = ν − 1 and 0 < λ < 1) ensue. Therefore,
increased flexibility is gained due to three design parameters (KP , KI , ν), i.e., to one more
degree of freedom with respect to a PI controller. However, reliable and effective rules to
design and tune the parameters must also be established.
Putting s = jω, the open-loop frequency response (OLFR for brevity) associated with
G(s) = Gc (s) Gp(s) is given by:
K KI [1 + TI (jω)ν ] −jωτ
G(jω) = e
(jω)ν (1 + jω T )
K KI {1 + TI ω ν [cos( π2 ν) + j sin( π2 ν)]} −jωτ
= e . (23)
ω ν [cos( π2 ν) + j sin( π2 ν)] (1 + jωT )
Now, after simple algebraic computations, putting θ = π2 ν, S = sin(θ), and C = cos(θ)

and introducing a nondimensional frequency u = ω T lead to:
K KI T ν 1 + TI ( Tu )ν [C + jS] −j uτ

G(ju) = e T (24)
uν [C + jS] (1 + j u)
Then the magnitude and phase angle frequency behaviors are given by:
s
K KI T ν 1 + 2 TI ( Tu )ν C + TI2 ( Tu )2ν
|G(ju)| = (25)
uν 1 + u2
TI ( Tu )ν S

uτ
∠G(ju) = arctan − arctan(u) − θ − (26)
1 + TI ( Tu )ν C T
1
Now, the closed-loop frequency response F (ju) = 1+G−1 (ju)
is considered and the op-
timality requirement |F (ju)| ≡ 1 is approximated in a sufficiently large bandwidth uB .
In theory, however, an optimal feedback system that provides perfect input-output tracking
should be achieved only if this holds true for all frequencies. i.e., |G(ju)| 1, ∀u should
be verified. Namely, “a feedback system is optimal if and only if the absolute value of the
return difference is at least one at all frequencies” [19], but this condition cannot be satis-
fied by the return difference |1+G−1 (jω)|. Instead, |G(ju)| is shaped around the crossover
uc so that a high gain is obtained at low frequencies, whereas it rolls off at high frequencies
to avoid stability problems [61]. The appropriate shaping guarantees robust stability and de-
sired performance despite parameter variations in the controlled injection system. Namely,
the non-integer order integrator provides a magnitude plot with a fractional slope of −20
dB/decade and a nearly flat phase diagram in a range around uc so that a nearly constant
phase margin is achieved in this range.
To synthesize, start by fixing uB where optimal tracking is desired. In particular, a
trade-off is reached. Namely, the higher uB is, the lower the rise time in the response of
the closed-loop system. Moreover, the lower is uB , the more the crossover uc is shifted
toward a centered position in the flat region of the phase characteristic. For the sake of
convenience, uB is chosen higher than the plant bandwidth.
Then the normalized crossover frequency is estimated by a commonly used relation
[31]: uc ∈ [ u1.7
B uB
, 1.3 ], e.g., uc = u1.5
B
. The estimate may be changed inside the indicated
range to adjust TI . Then, to achieve a specified phase margin P Ms at uc , TI is properly

selected. Namely, the phase margin P M = π + ∠G(juc) is:
TI ( uTc )ν S

uc τ
P M = arctan − arctan(uc ) −
1 + TI ( uTc )ν C T
uc τ
+ π − θ = ϕ1 (uc ) − ϕ2 (uc ) − +π−θ (27)
T
where ϕ1 (uc ) and ϕ2 (uc ) are the first two arguments on the right side of (27). Then, TI is
determined to compensate the delay by putting ϕ1 (uc ) − ϕ2 (uc ) − uc Tτ = 0 and obtain a
closed-form formula:
T ν [uc + tan( uTc τ )]
TI = . (28)
uνc [S − uc C − (C + uc S) tan( uTc τ )]
This choice yields:
P Ms = π − θ = (2 − ν) π/2 ⇔ ν = 2 − 2P Ms /π (29)
which represents a direct and easy design relation between the fractional order ν and the
specification provided on P Ms . Note that ν > 1 is always obtained for a plant without an
integrator. Obviously, at the end of the design procedure, it is always verified that the FOPI
controller ensures a nearly flat Bode plot of ∠G(ju), so that the obtained phase margin is
nearly constant in a range around uc .
To complete the design procedure, |G−1 (juc )|2 = 1 is helpful to get another closed-
form formula for determining KI :
s
1 uc ν 1 + u2c
KI = (30)
K T 1 + 2 TI ( T ) C + TI2 ( uTc )2ν
uc ν
The (30) also leads to KP = KI TI .

It is to be noted that the value and sign of TI in (28) depend on the choice of ν (i.e.,
the specified P Ms ), which affects C and S, and on the choice of uc (i.e., the specified
uB ). Then, if uB is fixed, ν is restricted to allow TI > 0, i.e., a minimum-phase and
stable compensated system. This means that restrictions apply on the achievable phase
margin. Vice versa, if P Ms is fixed, the same kind of restriction applies on uc . In other
words, restrictions apply on the achievable bandwidth. For example, for P Ms = π/4 with
ν = 1.5, Table 1 provides the values of TI for several values of the pairs (r, uc), where r is
the delay ratio r = τ /T and uc ∈ {uB /1.7, uB /1.5, uB /1.3}, with uB = 2π.
Finally, another point must be noted. Given T from the plant
uc , and
model, ν, the
T S−uc C
maximum plant time delay allowing TI > 0 is: τmax = uc arctan C+uc S . Then,
τmax satisfying this requirement depends on plant parameters and specifications on uB and
P Ms . Moreover, values ν ≥ 1.6 are not convenient because they imply a too low phase
margin, which is usually taken as greater than 35◦ [61]. Table 2 reports the values of TI
corresponding to uc = uB /1.5 = 4.1888, τ = 0.18 seconds and T = 1.6857 seconds
(r = 0.1068), which are the parameters of the CNG injection system model for a particular
working condition. In all cases, τmax > τ so the controller design is possible.
Table 1. Values of TI (s) corresponding to pairs (r, uc) for P Ms = 54◦ , 45◦, 36◦ (with
ν = 1.4, 1.5, 1.6)
(r, uc) ν = 1.4 ν = 1.5 ν = 1.6

(0.8, 3.6960) 0.3408 0.2940 0.2607
(0.8, 4.1888) 0.4599 0.3523 0.2830
(0.8, 4.8332) 1.8525 0.6835 0.3930
(1.0, 3.6960) 0.9623 0.6210 0.4495
(1.0, 4.1888) −1.0476 −6.3933 1.2395
(1.0, 4.8332) −0.0251 −0.0265 −0.0297
(1.2, 3.6960) −0.4398 −0.6572 −1.6913
(1.2, 4.1888) 0.0261 0.0267 0.0286
(1.2, 4.8332) 0.1828 0.1600 0.1435
Table 2. Values of TI and τmax for a given working condition (T = 1.6857 s, τ = 0.18 s)
ν P Ms TI (s) τmax (s)

1.3 63◦ 1.1719 0.2839
1.4 54◦ 0.6773 0.3472
1.5 45◦ 0.4606 0.4104
1.6 36◦ 0.3418 0.4736
1.7 27◦ 0.2684 0.5368
4.2. Realization
The final step of the design procedure requires realization of the irrational non-integer
order operator sν . To this aim, a rational transfer function approximation is required. Many
approximation methods exist, starting with the well-known recursive Oustaloup’s technique
[48]. Other approaches are based on truncation of continued fraction expansions (CFEs)
and interpolation techniques [12, 67] or signal processing techniques [6]. Here, an efficient
method is used that derives approximations from the Lagrange’s CFE [33].
This CFE-based method a priori guarantees minimum-phase zeros and stable poles of
the approximating rational transfer function and is based on closed-form formulas express-
ing coefficients of the numerator and denominator polynomials in the transfer function.
Moreover, zeros and poles are interlaced on the negative real half-axis [35], and this prop-
erty also holds for digital realization [32, 34, 36]. This is very important for controller
synthesis, because right-half plane (RHP) zeros and RHP poles in the open-loop transfer
function imply inherent limitations to the benefits of feedback [17, 60]. More specifically,
RHP poles imply constrained large gain bandwidth, which leads to a highly amplified sen-
sor noise at the input to the plant. Finally, the method can obtain a reduced approximation
error by a relatively low number of zeros and poles, so an easy implementation is possible in
the analog or digital domain. The closed-form formulas are valid for sν when 0 < ν < 1.
Then, since ν > 1 in the non-integer order integrator of the FOPI controller, given that
sν = s sν−1 , only sν−1 is approximated. To synthesize, if λ = ν − 1, the approximation is
as follows:
αN,0 (λ) sN + αN,1 (λ) sN −1 + . . . + αN,N (λ)
sλ ≈ (31)
βN,0 (λ) sN + βN,1 (λ) sN −1 + . . . + βN,N (λ)
where N ≥ 1 is the number of zero-pole interlaced pairs and the coefficients αN,j (λ) =
βN,N −j (λ), for j = 0, . . ., N , depend on λ and are determined by the following closed-
form formula:
j N
αN,j (λ) = (−1) (λ + j + 1)(N −j)(λ − N )(j) (32)
j
in which the so-called Pochhammer functions are (λ + j + 1)(N −j) = (λ + j + 1)(λ +
j + 2) . . . (λ + N ) and (λ − N )(j) = (λ − N )(λ − N + 1) . . . (λ − N + j − 1), with
(λ + N + 1)(0) = (λ − N )(0) = 1. Simple algebraic manipulations allow us to express the
previous formula differently [37, 38]:
αN,j = C(N, j) (j + 1 + λ)(N −j) (N − λ)(j)∗ (33)
βN,j = C(N, j) (N − j + 1 + λ)(j) (N − λ)(N −j)∗ (34)

where (N − λ)(j)∗ := (N − λ)(N − λ − 1) . . .(N − λ − j + 1) and (N − λ)(N −j)∗ :=
(N − λ)(N − λ − 1) . . .(j − λ + 1) are falling factorials. It holds (N − λ)(0)∗ = 1.
5. Simulation Results
This section shows the effectiveness, performance and robustness of the described con-
trol strategy. To this aim, a detailed non-linear simulation model is created by the commer-
cial AMESim developing package [22]. AMESim is a simulation tool based on a virtual
prototyping environment. It enables modeling and integration of different physical compo-
nents, supports and combines different levels of abstraction. The polymorphism concept of
AMESim allows the analysis of a multi-disciplinary system as a whole and the representa-
tion of many different scenarios. All its peculiarities and properties make AMESim capable
of replacing real experiments for validation, so that the non-linear model of this study accu-
rately represents the complex fluid-dynamic phenomena characterizing the injection system
at different working points.
The AMESim model makes some assumptions. First, the distribution of pressure in the
control chamber, the common rail and the injectors is uniform; the elastic deformations of
solid parts by pressure changes are negligible; the pipes, which are considered incompress-
ible ducts with friction, are subject to non-uniform pressure distribution. Vice versa, the
model considers temperature variations affecting the pressure dynamics in each component
and only takes into account heat exchanges through pipes, by properly computing a ther-
mal exchange coefficient. The tank pressure plays the role of a maintenance input and it is
modeled by a constant pneumatic pressure source. Finally, to simplify the AMESim model
construction, some supercomponents have been created to group multiple elements.
The operating ranges of the main variables in the system are the following:
• pressure in the tank: ptk ∈ [30, 200] bar;
• pressure in the control chamber: x1 ∈ [4, 25] bar;
• pressure in the common rail: x2 ∈ [4, 25] bar;
• injector opening time: tinj ∈ [025] ms;
• duty-cycle of the command signal for the valve: [0, 100]%;
• engine speed: ωrpm ∈ [1000, 4000] rpm.
As a consequence, the model parameters vary in the following ranges: K ∈ [134, 1129]
bar, T ∈ [1.059, 18.546] seconds.
In the most representative simulation tests, the tank pressure, the engine speed and the
injector opening time are kept constant at ptk = 50 bar, ωrpm = 2500 rpm and tinj = 5ms,
respectively. However, Table 3 shows the system parameters in several conditions.
Table 3. Engine parameters in working points
tinj (ms) x2 (bar) ptk (bar) K (bar) T (s)

5 4 50 156 1.52
5 5 50 178 1.75
5 6 50 200 1.96
5 8 50 243 2.38
5 10 50 288 2.82
20 5 50 176 1.73
25 20 50 526 5.18
To test their efficiency and robustness, the gain-scheduled FOPI controllers, with ν =
1.3, 1.4, 1.5, are compared to a standard integer order PI that is typically used for injection
control and usually tuned by the open-loop Ziegler-Nichols rules. The PI is gain-scheduled
for fairness of comparison. Values of ν < 1.3 or ν > 1.6 are discarded because they
lead to too high or low (below 36◦ ) phase margins. With reference to the typical working
conditions in Table 3, two different cases are simulated.
In the first one, the reference pressure undergoes a small variation from 5 to 6 bar
and the system variables are perturbed slightly from their working point. Hence, the local
behavior can be analyzed by the linearized model. The parameters of the injection system
are K = 200 bar, T = 1.96 seconds (see Table 3). A single FOPI or PI controller is used
and designed according to these parameters associated with the final reference pressure that
must be reached.
Figure 3 shows the closed-loop step response.
With respect to a standard PI, the FOPI controllers improve the performance indices:
even if the rise time is slightly longer, the overshoot is highly reduced, so that a better
injection accuracy is achieved; the settling times are comparable for ν ≤ 1.5. With the
PI controller, the performance is influenced primarily by disturbances and nonlinearities
Figure 3. Response to a 5-to-6 bar step in the reference pressure.
related to operations of injectors, by PWM modulation of the solenoid valve command.

Large variations of the control signal cause high frequency pressure oscillations around the
equilibrium point. Moreover, the saturation in the control signal, which is associated with
a complete closure of the valve, determines large rail pressure overshoots and undershoots.
Conversely, performances of fractional controllers are less sensitive to system disturbances
and nonlinearities.
The second case simulates a large variation in the reference pressure from 6 to 10 bar,
with a step of 2 bar, and then back from 10 to 6 bar, with the same step. Then, the scheduling
switches between several controllers. In particular, since the variation is divided into two
steps, two FOPI/PI controllers are used. This choice is justified by simulation. Several tests,
indeed, verified that stability is achieved if variations of the rail reference pressure and of
injector opening times are within operating ranges.
This second experiment aims at testing control robustness. Namely, the emptying and
filling of the common rail follow different dynamics. This is due to several reasons. First,
the gradient between the tank pressure and the main chamber is higher than between the
common rail and the discharging manifold. Secondly, the rail fills more rapidly because
the feeding tank has a higher pressure than the rail. In contrast, the rail empties more
slowly because of the much lower rail pressure and the injection outflow. However, injection
timings do not vary because they are set by the ECU based on the amount of fuel to inject.
Therefore, the emptying process can be accelerated only by closing the solenoid valve.
Figure 4 shows that, for rising transients, FOPIs remarkably reduce the overshoots and
responses are similar to the first case. FOPIs are very prompt in resetting the rail pressure
Figure 4. Response to large variations in the reference pressure.
to the set-point.
Conversely, the PI controller reacts poorly to larger variations, and the previously de-
scribed non-linear phenomena considerably affect overall performance.
For descending transients starting from t = 13 s, owing to the request of a quick pres-
sure reduction, a duty cycle equal to zero closes the valve completely. Thanks to injections,
the rail pressure decreases to the final reference values (8 and 6 bar, respectively), with a
time constant depending on the system geometry. The error cannot be reduced with a higher
rate than that achieved, due to the saturation of the actuation variable. This is reflected by
the same slope of pressure plots by FOPIs or PI. Figure 4 clearly shows a better tracking of
the PI during the descending transient with respect to the rising transient. At the same time,
a large undershoot of the rail pressure still remains before reaching the steady state.
Therefore, FOPIs achieve an overall better quality of the response, so they may also
improve injection pressure regulation for large perturbations of working conditions.
All the achieved results confirm the higher robustness and performance of the FOPI
controllers in all simulated conditions. The tests show that FOPIs significantly im-
prove the usual performance indices for injection pressure control, by reducing under-
shoots/overshoots, steady-state errors and settling times [14].
This remarkable result can be explained by the proposed control approach. Namely,
on one side, it is already well known that fractional order control may help to improve
robustness and closed-loop performance. But the added value here is the robust stability
properties enforced by the D-decomposition. If one considers the variation of the working
points of the injection system, then the approach guarantees that, for each new working
point associated with a reference rail pressure and injection timing, the representative point
P of the designed controller (for each value of ν) belongs to the stability domain. In par-
ticular it lies on the relative stability curve associated with the designed phase margin (see
Figure 2 for two different points).
This occurs because, for each value of ν, the position and amplitude of both the CRB
boundary curves and the relative stability curves change, but their shape and mutual position
do not. In addition, consider parametric variations, uncertainties and model inaccuracies
that bring the design point P closer to a modified actual CRB curve than to an initially
used nominal CRB curve. In this case, the distance between the nominal CRB curve and
the relative stability curve on which P lies is sufficient to guarantee stability. If needed,
this distance can also be enlarged by shifting the point P along the relative stability curve
with the same phase margin specification (see possible different locations in Figure 2) with
increasing ω. In this way, P is relocated away from the CRB curve provided that ωc , then
ωB , can be increased.
Finally, the employed scheduling strategy guarantees stability when switching the frac-
tional order controllers. Namely, as far as variations in the reference pressure are bounded
by 2 bar and those in injection duration by 6 s (which is a good indication for real practice),
simulation results always verify closed-loop stable responses.
6. Conclusion
Injection pressure regulation is currently an important control problem in the automo-
tive industry because of the ever-growing focus on reducing pollutant emissions and con-
sumption and increasing engine efficiency. In this chapter, the case of CNG engines was
approached by an innovative fractional order control strategy. The typical advantages of-
fered by FOCs are amplified by combining the D-decomposition methodology and appro-
priate controller scheduling for the specific injection system working points. To synthesize,
injection pressure is regulated by tuning different FOPI controllers based on the reference
working points to achieve. Variations between working points and disturbances are com-
pensated by scheduling the FOPI controller gains. Moreover, several simulation studies
verify that switching between controllers does not lead to stability problems. To sum up,
the main benefits of the proposed approach for controlling rail pressure in injection systems
of CNG engines are the following.
Closed-form and relatively simple design formulas are applied to obtain the controller
gains in terms of frequency domain specifications; such formulas could be used for an
automatic synthesis of the controller. As far as relative stability is concerned, a specification
on the phase margin can be analytically achieved and is strictly related to the required non-
integer order ν and, vice versa, the selected order ν determines the phase margin that can
be obtained (see formulas in (29)).
The integral (or derivative) operator of the fractional order controller is realized by an
efficient approximation method that prevents numerical problems and leads to a rational
transfer function characterized by a low number of zeros and poles. Both reduced ap-
proximation errors and easy implementation are guaranteed by the proposed realization.
The approximating rational transfer function shows remarkable properties: the poles of the
transfer function are stable, the zeros are minimum-phase, and all singularities are interlaced
on the negative real half-axis of the s-plane [35]; interlacing, stability and minimum-phase
characteristics are also achieved by the discrete transfer function used for digital realiza-
tion [34, 36, 37].
Robust stability of the designed FOPI controllers is guaranteed by the D-
decomposition, which allows a greater robustness with respect to the usual solutions em-
ployed in the automotive industry, even if large variations of working points are considered.
Finally, if switching between sufficiently close working points is done such that variations
of injection timings are below 6 seconds and changes of rail pressure are below 2 bar, then
nonlinearity effects, oscillations and instability problems in the rail pressure are prevented.
Acknowledgment
This work was supported by the Italian Ministry of University and Research under
project “EURO6 - Advanced electronic control unit, injection system, combustion strate-
gies, sensors and production process technologies for low polluting diesel engines”.
References
[1] Ackermann, J., and D. Kaesbauer (2003). Stable polyhedra in parameter space. Auto-
matica, 39 (5), 937–943.
[2] Amorese, C., De Matthaeis, S., De Michele, O., and A. Satriano. The gaseous fuel
option: LPG and CNG. In Proc. Int. Conf. on Vehicles Alternative Fuel System &
Envir. Protection, Dublin, Ireland, 6–9 July 2004.
[3] Arena, P., Caponetto, R., Fortuna, L., and D. Porto (2000). Nonlinear Noninteger
Order Circuits and Systems: An Introduction. In L. O. Chua (Ed.) World Scientific
Series on Nonlinear Science, Series A (Vol. 38). Singapore: World Scientific.
[4] Åström, K. J., and T. Hägglund (1995). PID Controllers: Theory, Design, and Tuning.
2nd edition. Research Triangle Park, NC: Instrument Society of America.
[5] Barbosa, R. S., Tenreiro Machado, J. A., and J. M. Ferreira (2004). Tuning of PID
controller based on Bode’s ideal transfer function. Nonlinear Dynamics, 38, 305–321.
[6] Barbosa, R. S., Tenreiro Machado, J. A., and M. F. Silva (2006). Time domain de-
sign of fractional differintegrators using least squares. Signal Process., 86 (10), 2567–
2581.
[7] Bode, H.-W. (1945). Network Analysis and Feedback Amplifier Design. New York: D.
Van Nostrand Co., Inc..
[8] Bonnet, C., and J. R. Partington (2001). Stabilization of fractional exponential systems
including delays. Kybernetika, 37 (3), 345–353.
[9] Caponetto, R., Dongola, G., Fortuna, L., and I. Petráš (2010). Fractional Order Sys-
tems: Modeling and Control Applications. World Scientific Series on Nonlinear Sci-
ence Series A: Vol. 72. Singapore: World Scientific.
[10] Charef, A. C., Sun, H. H., Tsao, Y. Y., and B. Onaral (1992). Fractal systems as
represented by singularity function. IEEE Trans. Autom. Contr., 37 (9), 1465–1470.
[11] Charef, A. C., (2006). Analogue realisation of fractional-order integrator, differentia-

tor and fractional P I λ D µ controller. IET Contr. Theory and Appl., 153 (6), 714–720.
[12] Chen, Y.-Q., Vinagre, B. M., and I. Podlubny (2004). Continued fraction expansion
approaches to discretizing fractional order derivatives – An expository review. Non-
linear Dynamics, 38 (1-2), 155–170.
[13] Chen, Y.-Q. (2006). Ubiquitous Fractional Order Controls? In Proc. of the Second
IFAC Symposium on Fractional Derivatives and its Applications (IFAC FDA’06),
Porto, Portugal, 19-21 July 2006, vol. 2, part 1, 168–173.
[14] di Gaeta, A., Montanaro, U., Fiengo, G., Palladino, A., and V. Giglio (2012). A model-
based gain scheduling approach for controlling the common-rail system for GDI en-
gines. International Journal of Control, 85 (4), 419–436.
[15] Fiengo, G., di Gaeta, A., Palladino, A., and V. Giglio (2013). Common Rail System
for GDI Engines: Modelling, Identification, and Control. London: Springer Briefs in
Electrical and Computer Engin., Springer.
[16] Hamamci, S. E. (2007). An algorithm for stabilization of fractional-order time delay

systems using fractional-order PID controllers. IEEE Trans. Automatic Control, 52
(10), 1964–1969.
[17] Horowitz, I. (1963). Synthesis of Feedback Systems. New York, London: Academic
Press.
[18] Hwang, C., and Y. C. Cheng (2006). A numerical algorithm for stability testing of
fractional delay systems. Automatica, 42 (5), 825–831.
[19] Kalman, R.-E. (1964). When is a linear control system optimal? Transactions of the
ASME, Journal of Basic Engineering, 86, series D, 84–90.
[20] Khalil, H. K. (2002). Nonlinear Systems. 3rd edition. Upper Saddle River, NJ:
Prentice-Hall.
[21] Li, H. S., Luo, Y., and Y. Q. Chen (2010). A fractional order proportional and deriva-
tive (FOPD) motion controller: tuning rule and experiments. IEEE Trans. Control
Systems Technology, 18 (2), 516–520.
[22] Lino, P., and B. Maione (2007). Integrated design of a mechatronic system - The
pressure control in common rails. In Proc. 4th Int. Conf. on Informatics in Control,
Autom. and Robot. (ICINCO 2007), Angers, France, 9-12 May 2007, 11–18.
[23] Lino, P., Maione, B., and C. Amorese (2008). Modelling and predictive control of a
new injection system for compressed natural gas engines. Control Engineering Prac-
tice, 16 (10), 1216–1230.
[24] Lino, P., and G. Maione (2012). Fractional Order PI Tuning for Integrating Plants with
Time Delay. In R. Vilanova and A. Visioli (Eds.), 2nd IFAC Conf. on Advances in PID
Control (PID’12), Brescia, Italy, 28-30 Mar. 2012, IFAC Proceedings on line: Vol. 2,
Part I, 649–654.
[25] Lino, P., and G. Maione (2013). Laboratory experiments on fractional order PI control
of a servo system with delay. In E. Witrant, J. Sabatier, J.A. Tenreiro Machado, O.
Sename, and L. Dugard (Eds.), First IFAC Joint Conference on System Structure and
Control, Time-Delay Systems and Fractional Differentiation (2013 IFAC Joint Con-
ference SSSC-TDS-FDA) – Track FDA: Sixth Workshop on Fractional Differentiation
and Its Applications, Grenoble, France, 4-6 Feb 2013, IFAC Proceedings on line: Vol.
6, Part I, 899–904.
[26] Lino, P., and G. Maione (2013). Loop-Shaping and Easy Tuning of Fractional-Order
Proportional Integral Controllers for Position Servo Systems, Asian Journ. of Control,
5 (3), 1–10.
[27] Lino, P., and G. Maione (2013). Fractional order control of the injection system in
a CNG engine. In 2013 Europ. Contr. Conf., Zürich, Switzerland, 17-19 July 2013,
3997–4002.
[28] Lino, P., and G. Maione (2013). Design and simulation of fractional-order controllers
of injection in CNG engines. In T. Kawabe (Ed.), 7th IFAC Symp. Adv. Automot.
Contr., Tokyo, Japan, 4-7 Sep. 2013, IFAC Proceedings on line: Vol. 7, Part 1, 582–
587.
[29] Luo, Y., and Y. Q. Chen (2012). Stabilizing and robust fractional order PI controller
synthesis for first order plus time delay systems. Automatica, 48 (9), 2159–2167.
[30] Ma, C. B., and Y. Hori (2007). Fractional order control: Theory and applications in
motion control. IEEE Industrial Electronics Magazine, Winter 2007 (4), 6–10.
[31] Maciejowski, J. M. (1989). Multivariable Feedback Design. Wokingham, UK:

Addison-Wesley.
[32] Maione, G. (2006). Concerning continued fractions representations of noninteger or-

der digital differentiators. IEEE Signal Processing Letters, 13 (12), 725–728, 2006.
[33] Maione, G. (2008). Continued fractions approximation of the impulse response of

fractional order dynamic systems. IET Contr. Theory and Appl., 2 (7), 564–572, 2008.
[34] Maione, G. (2011). High-speed digital realizations of fractional operators in the delta
domain. IEEE Trans. Autom. Control, 56 (3), 697–702.
[35] Maione, G. (2011). Conditions for a class of rational approximants of fractional dif-
ferentiators/integrators to enjoy the interlacing property. In S. Bittanti, A. Cenedese,
and S. Zampieri (Eds.), Proc. 18th IFAC World Congr., Università Cattolica del Sacro
Cuore, Milan, Italy, 28 Aug. - 2 Sept. 2011, IFAC Proceedings on line: Vol. 18, Part
1, 13984–13989.
[36] Maione, G. (2013). On the Laguerre rational approximation to fractional discrete

derivative and integral operators. IEEE Trans. Autom. Control, 58 (6), 1579–1585.
[37] Maione, G. (2013). Closed-form rational approximations of fractional, analog and

digital differentiators/integrators. IEEE J. on Emerging and Selected Topics in Circuits
and Systems, 3 (3), 322–329.
[38] Maione, G. (2013). Correction to “Closed-form rational approximations of fractional,

analog and digital differentiators/integrators”. IEEE J. on Emerging and Selected Top-
ics in Circuits and Systems, 3 (4), 654.
[39] Maione, G., and P. Lino (2007). New tuning rules for fractional P I α controllers. Non-
lin. Dyn., 49 (1-2), 251–257.
[40] Manabe, S. (1961). The non-integer integral and its application to control systems.
Japanese Inst. Electrical Engineers Journal, 6 (3-4), 83–87.
[41] Matignon, D. (1996). Stability result on fractional differential equations with applica-
tions to control processing. In Proc. of IMACS-IEEE SMC Conf., Lille, France, 9-12
July 1996, vol. 2, 963–968.
[42] Monje, C. A., Chen, Y. Q., Vinagre, B. M., Xue, D., and V. Feliu-Batlle (2010).
Fractional-order Systems and Controls: Fundamentals and Applications. London,
U.K.: Springer-Verlag.
[43] Monje, C. A., Vinagre, B. M., Feliu, V., and Y. Q. Chen (2008). Tuning and auto-
tuning of fractional order controllers for industry applications. Control Engineering
Practice, 16, 798–812.
[44] Moze, M., Sabatier, J., and A. Oustaloup (2010). Air-fuel ratio control of an internal
combustion engine using CRONE control extended to LPV systems. In D. Baleanu,
Z. B. Güvenç, and J. A. Tenreiro Machado (Eds.) New Trends in Nanotechnology and
Fractional Calculus Applications (71–86). Dordrecht, The Netherlands: Springer.
[45] Naus, G. J. L. (2010). Model-based control for automotive applications. PhD Thesis,
Technische Universiteit Eindhoven, http://dx.doi.org/10.6100/IR690571, 2010.
[46] Neimark, Yu.I. (1949). Ustoichivost’ linearizovannykh sistem upravleniya (Stability

of Linearized Control Systems). LKVVIA, Leningrad.
[47] Oldham, K. B., and J. Spanier (1974). The fractional calculus: Integrations and Dif-
ferentiations of Arbitrary Order. New York, NY: Academic Press.
[48] Oustaloup, A. (1991). La Commande CRONE: Command Robuste d’Ordre Non

Entièr. Paris: Editions Hermès.
[49] Oustaloup, A. (1999). La Commande CRONE: du Scalaire au Multivariable. Paris:

Editions Hermès.
[50] Oustaloup, A., Levron, F., Nanot, F., and B. Mathieu (2000). Frequency band complex
non integer differentiator: Characterization and synthesis. IEEE Trans. Circuits Syst.
I, Fundam. Theory Appl., 47 (1), 25–40.
[51] Oustaloup, A., Moreau, X., and M. Nouillant (1996). The CRONE suspension. Con-
trol Engineering Practice, 4 (8), 1101–1108.
[52] Oustaloup, A., Sabatier, J., and P. Lanusse (1999). From fractal robustness to CRONE
control. Fractional Calculus and Applied Analysis, 2 (1), 1–30.
[53] Oustaloup, A., Sabatier, J., Lanusse, P., Malti, R., Melchior, P., Moreau, X., and M.
Moze (2008). An overview of the crone approach in system analysis, modeling and
identification, observation and control. In M. J. Chung and P. Misra (Eds.), Proc. of
the 17th IFAC World Congress, Seoul, Korea, 2008, Vol. 17, Part 1, 14254–14265.
[54] Petráš, I. (2009). Fractional-order feedback control of a dc motor. Journal of Electrical

Engineering, 60 (3), 117–128.
[55] Podlubny, I. (1999). Fractional Differential Equations. San Diego, CA: Academic
Press.
[56] Podlubny, I. (1999). Fractional-order systems and P I λ D µ controllers. IEEE Trans.

Aut. Contr., 44 (1), 208–214.
[57] Podlubny, I., Dorcak, L., and I. Kostial (1997). On fractional derivatives, fractional-
order dynamic systems and P I λ D µ -controllers. In Proc. 36th IEEE Conf. on Decision
and Control, San Diego, CA, USA, 10-12 Dec. 1997, vol. 5, 4985–4990.
[58] Raynaud, H.-F., and A. Zergaı̈noh (2000). State-space representation for fractional
order controllers. Automatica, 36, 1017–1021.
[59] Rugh, W. J., and J. S. Shamma (2000). Research on gain scheduling. Automatica, 36
(10), 1401–1425.
[60] Sidi, M. J. (2001). Design of Robust Control Systems: From Classical to Modern
Practical Approaches. Malbar (FLA), USA: Krieger Publishing Company.
[61] Skogestad, S., and I. Postlethwaite (2005). Multivariable Feedback Control: Analysis
and Design. 2nd edition. Chichester, England: Wiley & Sons, Ltd..
[62] Sun, Z., and S. S. Ge (2005). Switched Linear Systems: Control and Design. London:
Springer-Verlag.
[63] Tenreiro Machado, J. A. (1997). Analysis and design of fractional-order digital control
systems. Journal Systems Analysis Modelling Simulation, 27 (2-3), 107–122.
[64] Tenreiro Machado, J. A. (2001). Discrete-time fractional-order controllers. Journal of

Fractional Calculus & Applied Analysis, 4 (1), 47-66.
[65] Tustin, A. (1958). The design of systems for automatic control of the position of mas-
sive objects. Proc. Institution of Electrical Engineers, 105, Part C, Suppl. No. 1, 1–57.
[66] Ulsoy, A. G., Peng, H., and M. Çakmakci (2012). Automotive Control Systems. 1st
edition. Cambridge: Cambridge University Press.
[67] Vinagre, B. M., Podlubny, I., Hernandez, A., and V. Feliu (2000). Some approxima-
tions of fractional order operators used in control theory and applications. J. Frac.
Calc. Appl. Anal., 13 (3), 231–248.
[68] Zucrow, M., and J. Hoffman (1976). Gas Dynamics. New York: John Wiley & Sons.
Chapter 4
LINEAR INTEGER ORDER SYSTEM CONTROL

BY FRACTIONAL PI-STATE FEEDBACK
Rachid Mansouri1, Maamar Bettayeb2,3, Chahira Boussalem1

and Ubaid M. Al-Saggaf 3
1
L2CSP Laboratory, University Mouloud Mammeri of Tizi-Ouzou, Algeria
2
Electrical & Computer Engineering Department, University of Sharjah, UAE
3
Center of Excellence in Intelligent Engineering Systems (CEIES),
King Abdulaziz University, Jeddah, KSA
Abstract
This book chapter deals with a linear non-fractional order system control by a PI-state
feedback fractional-controller. The fractional aspect of the control law is introduced by a
dynamic state feedback as ( ) ( ) ( ( )).
Two methods, based on pole placement principle, are then proposed to design the
feedback vector gains and .
The first one uses the augmented system method. In this case, the closed-loop
characteristic polynomial is of order ( ) where is the order of the integer system and
is the fractional ratio of the non integer order ( /q).
The second proposed method allows to decompose the closed-loop characteristic
polynomial into a fractional polynomial of order , (0 < < 2) and an integer order
polynomial of order . In this case, a suitable choice of the poles provides to obtain, in
closed-loop, the Bode's ideal transfer function.
These two methods are then applied to stabilize an inverted pendulum-car system and the
effectiveness and robustness of the proposed controllers are examined by experiments on a
real system.
Keywords: Fractional calculus, Fractional systems, PI-state feedback, Pole placement,

Inverted pendulum-cart system stabilization
66 Rachid Mansouri, Maamar Bettayeb, Chahira Boussalem et al.
Introduction
The idea of fractional calculus has been known since the development of the regular
integer order calculus, with the first reference being associated with Leibnitz and L'Hopital in
1695 where half order derivatives were mentioned. A detailed chronology of this notion is
given in [23, 36]. The books [31, 41, 44] and more recently [18, 22, 33] constitute currently
the main references of this theory. Non integer differentiation is currently considered among
emerging thematics applied to many science and technology disciplines. All works
undertaken on systems modeling by linear differential equations using the usual integer
differentiation must be re-examined. From the automation point of view, the fundamental
notions analysis and simulation of fractional systems and the fractional controllers design
constitute highly promising fields of research.
In control theory, the non integer differentiation/integration has also proven its efficiency.
Indeed, Oustaloup, [40] with the CRONE-controller (Commande Robuste d'Ordre Non
Entier) and Podlubny, [42] with the -controller, involving the fractional-order
integrator and differentiator, have shown the advantage of the fractional-controllers over the
classical ones. Fractional-order controllers have received a considerable attention in the last
three decades (see [7, 12, 24, 33] for example). In fact, in principle, they provide more
flexibility in the controller design, with respect to the integer order controllers, because they
have more parameters to select.
However, almost all fractional controller design methods use the transfer function
representation because of the simplicity of the mathematical manipulations. There is still a
lack of strategies for introducing fractional dynamics in state space methods. The state space
representation of fractional order systems is introduced in [13, 19, 43, 49]. It has been
exploited in order to analyze performance of fractional order systems. Indeed, the stability of
the fractional order system has been investigated [28, 30]. Further, the controllability and the
observability properties have been defined and some algebraic criteria of these two properties
have been derived [8, 29, 51]. Recently, some new results on the minimal state space
realization for fractional order systems [47, 48] and approximation of fractional order state
space SISO and MIMO models in both commensurate and non-commensurate cases are
derived [25-27].
Control systems can include both the fractional order system to be controlled and the
fractional order controller. However, in practice, it is more common to consider only the
controller to be of fractional order. This is due to the fact that the plant model may have been
already obtained as an integer order model in the classical sense. In state space representation,
the introduction of the concept of fractional differentiation is not obvious because it does not
appear explicitly in the model. As in the transfer function representation, the fractional aspect
will be introduced by the control law. However, in state space, the control law is usually used
as a static state feedback that does not allow the introduction of the fractional aspect. In [34,
46], the authors present a fractional order state space control strategy for integer order
systems. It is based on rearranging integer order system equations into a set of fractional
(rational) order equations using system augmentation. By doing so, fractional order dynamics
can be introduced in the controlled system by using traditional state-space methods for
controller design. System augmentation is also used in [14, 15] to study the stabilization
Linear Integer Order System Control by Fractional PI-State Feedback 67
problem of commensurate fractional order systems by a pseudo-state feedback. However, the

non-integer order must be rational.
Pole placement via state feedback has been known for several decades [35]. Recently,
state-derivative feedback has been proposed [1-3, 18, 36] due to considerably low gain and
fast response. In this chapter, new approach is presented to design the fractional PI-state
feedback gain, two solutions are then proposed. In the first one, the integer PI-state feedback
solution proposed in [50] is generalized to the fractional PI-state feedback. We show that we
find the same solution but also the same problems as the solution proposed in [46] by using
system augmentation principle. Our second approach [10] provides a very simple method to
achieve the gain vectors of the fractional PI-state feedback. The desired closed-loop
characteristic polynomial is composed of two polynomials: The first one is integer and
permits to impose integer poles and the second one is a non-integer polynomial having
the form ( ), ( ) imposed to the closed-loop when CRONE principle [38-
40] (phase flatness of the open-loop) is used. These two methods are then applied to stabilize
an inverted pendulum-car system.
This chapter contains six sections including the introduction. In Section 2, some basic
definitions are recalled. These definitions include the fractional integral operation, the Bode's
ideal transfer function and the stability condition of fractional order systems. Pole placement
by integer PI-state feedback algorithm is also presented in this section. The main results are
presented in Sections 3 and 4. In Section 3, the pole placement method by integer PI-state
feedback is generalized to the fractional one. Limits of this method are also investigated and
presented through a first order system. Section 4 contains the second method which is more
general and simpler to implement. After presenting the inverted pendulum-car system in
Section 5, those two methods are used in Section 6 to stabilize it. Experimental results are
also presented in this section.
1. Basic Definitions and Preliminaries

1.1. Fractional Integral Operation
Let ( ) be a real function, of real variable , continuous and differentiable on , the

-th repeated integration, denoted ( ( )), is given by the Cauchy formula
( )
( ( )) ∫ ( )
( ) (1)
is an integer number. The generalization of the Cauchy formula (1) to a non-integer number
is given by [20]
( )
( ( )) ∫ ( )
( ) (2)
( ) is the Euler Gamma function
( ) ∫ (3)
( )
Note that, in Equation (2), the term ( )
is equal to when . We thus find the
usual definition of the integration. When is a real number, Equation (2) can be written as
( ( )) ( ) ( ) (4)
being the convolution product, and ( ) is given by
( )
( ) ( )
(5)
Oustaloup [38] calls the weight function ( ) the forgetting factor ("le facteur d'oubli").
1.2. Bode's Ideal Transfer Function
In his study on the design of feedback amplifiers, Bode [10] suggested an ideal shape of
the open-loop transfer function of the form
( ) (6)
where 1/ is the gain crossover frequency, that is, | ( )| . The parameter is

the slope of the magnitude curve, on log-log scale, and may assume integer as well as non-
integer values. In fact, the transfer function ( ) is a fractional-order differentiator when
and a fractional-order integrator for . In the Bode diagram, the amplitude curve
of ( ) is a straight line of constant slope dB/dec, and the phase curve is a horizontal
line at – rad.
Let us now consider the unity feedback system represented in Figure (1) with Bode's ideal
transfer function ( ) inserted in the forward path. This choice of ( ) gives a closed-loop
system with the desirable property of being insensitive to gain changes, also called the "iso-
damping" property [38]. If the gain changes, the crossover frequency will vary but the phase
margin of the system remains equal to ( – )rad, independently of the value of the gain.
Figure 1. Fractional-order control system with Bode's ideal transfer function.
The closed-loop system of Figure (1) is given by
( )
( ) (7)
( )
It exhibits important properties such as infinite gain margin and constant phase margin
(dependent only on ). Thus, this closed-loop system is robust to process gain variations and the
step response exhibits iso-damping property. The order and the time constant determine
the overshoot (dependent on )) and the settling time (dependent on ), respectively [38, 40].
1.3. Stability of the Fractional Order Systems
The definition of stability adopted here is the Bounded Input Bounded Output (BIBO)
stability, also called external stability. For commensurate non-integer order systems, the
stability condition is, as for the integer system, that the roots of the characteristic equation lie
in the left-hand s-plane.
In practice, checking the stability condition by calculating the roots of the characteristic
equation is very difficult due to the complexity of their calculation. Instead of using the roots
of the characteristic polynomial in , Matignon has established a stability condition by using
the integer order polynomial, of complex variable , obtained by the change of variable
. This condition can therefore be applied only to commensurate non-integer order
systems.
Theorem 1. [28-30] A non-integer system of commensurate order , ( ) is

BIBO stable, if and only if, the roots of the integer polynomial ( ),obtained from the
characteristic equation of the non-integer system ( ), by the change of variable ,
satisfy the condition
| ( )| (8)
is the order of ( ), , ( ) are the roots, and ( ) is the argument of the

root .
Remark: The commensurability condition of the fractional order is a necessary condition.
Indeed, when this condition is not verified, the study of fractional order system stability based
on the roots of its characteristic polynomial only is not feasible.
1.4. Pole Placement by Integer PI-State Feedback
State feedback is a well established tool for the stabilization and control of dynamic
systems, and it is widely used in the field of control engineering.
The idea of using state-PID feedback is relatively new, and so far only few applications can
be found in the literature. In [1-3, 18], the motivation for using state derivative feedback instead
of conventional state feedback is given by the easy implementation in mechanical applications
where accelerometers and velocity sensors are used for measuring the system motion.
Thus, the accelerations and velocities are the sensed variables, as opposed to the
displacement. Typical applications are in the vibration control of mechanical systems [2, 3,
18]. The use of acceleration feedback in vibration suppression problems has also been
discussed at length in [37].
The application of state derivative feedback to vibration problems has been discussed
from the perspective of robust control in [21]. In what follows, we recall the pole placement
method in the integer PI-state feedback case, see [50] for more details.
Let's consider a controllable LTI integer order system described by
̇( ) ( ) ( ) ( )
{ (9)
( ) ( )
where is the state vector, and is the control input, is the output. ,
, are the system matrix, the control and the output gain vectors
respectively.
The control law by PI-state feedback is
( ) ( ) ∫ ( ) (10)
where and are the designed gain vectors

to achieve a desired closed-loop characteristic polynomial. The block diagram of Figure (2)
represents a LTI system with PI-state feedback.
Figure 2. Block diagram representation of a linear system with PI-state feedback.
The corresponding characteristic polynomial is given by
( ) ( ) (11)
Since the system is controllable, there exists a linear transformation such that ,
System (9) becomes
̇( ) ( ) ( ) ( )
{ (12)
( ) ( )
and the control law by PI-state feedback (10) becomes
( ) ( ) ∫ ( ) (13)
where
[ ] [ ]
The closed-loop model is then given by
̇( ) ( ) ( ) ∫ ( ) ( )
{ (14)
( ) ( )
The closed-loop characteristic equation is
( ) ( ( )) * ( ) + (15)
( ) is given by
( )
[ ( ) ( ) ( ) ( ) ( )]
where
( )
( )
{ ( )
( ) having the controllable canonical form, the characteristic polynomial of the

closed-loop ( ) is given by
( ) ( ) ( )
( ) ( ) (16)
which can be written as
( ) ( )
( ) ( ) ( )
( ) ( )
( ) ( ) (17)
so
( ) ( ) ( )
( ) ( ) (18)
( ) is of order when the system is of order . Therefore, it is necessary to

impose poles to the closed-loop system. The desired polynomial is then
( ) ∏ ( ) (19)
term by term identifying between the polynomials (18) and (19) gives
(20)
Equation (20) shows that when the integral function is not used ( ), we find the
formulas of the usual static state feedback.
To calculate the coefficients of the feedback gain and , we can ensure that the
coefficients of compensate the coefficients ( ) of the open-loop
characteristic polynomial, except the coefficient which is equal to ( ).
Coefficients of are used to impose the coefficients ( ) of the closed-loop
characteristic polynomial.
Feedback gains and are then given by
( )
{ (21)
The PI-state feedback gains and of the original System (9) are then given by
and (22)
2. Pole Placement Based Design of the Fractional Pi-State

Feedback
The generalization of the PI-state feedback involves replacing the integer order
integration by a fractional order operator. Figure (3) shows the block diagram of an integer
system controlled by the fractional PI-state feedback. The generalization is given by the
following result.
Figure 3. Block diagram of a fractional PI-state feedback.
Theorem 2. Consider a controllable LTI integer order SISO system described by

Equation (9) and let the control law
( ) ( ) ( ( )) (23)
where, and , are gain vectors designed to impose the characteristic

polynomial of the closed-loop. ( ( )) is the non-integer integration operator of order
defined by Equation (2). Vectors and that can arbitrarily place the poles of the closed-
loop characteristic polynomial written in the form
( ) (24)
are given by
( )
{ (25)
being the linear transformation that gives the controllable canonical form.
Proof. The characteristic polynomial of the integer model of Equation (9) is given by (11).
Since the model (9) is controllable, it can be rewritten as (12). The control law (23) becomes
( ) ( ) ( ( )) (26)
where
and (27)
The state model in closed-loop is then
̇( ) ( ) [ ( ) ( ( )) ] ( )
{ (28)
( ) ( )
Laplace transform of Equation (28) yields

( ) ( ) ( ) ( ) (29)
( ) ( ) (30)
Matrix ( ) has the general form
( ) (31)
where
( )
[̃ ( ) ̃ ( ) ̃ ( ) ̃ ( ) ̃ ( )]
and
̃ ( )
̃ ( )
( )
{̃
The determinant of is ( )
( ( )) ( ) ( )
( ) ( ) (32)
( ) ( )
( ( )) ( ) ( )
( ) ( )
( ) ( )
The characteristic polynomial of the closed-loop system is then
( ) ( ) ( )
( ) ( ) (33)
Let ( ) the polynomial that we want to impose to the closed-loop

( ) (34)
Identification term by term between ( ) of Equation (33) and ( ) of Equation (34)

gives
and (35)
{ {
Therefore, and are calculated by
( ) ( ) ( )( )
{ (36)
Vector gains and of the original control law (23) are finally given by
and (37)
This completes the proof.
The fractional PI-state feedback design is based on the coefficients ( ) of

the desired polynomial (34). The problem is how one can find the values of these coefficients
from the poles imposed to the desired polynomial, as in the integer case. The polynomial
(34) being fractional, the number and the values of its roots depend on the value of .
Let ( and being two integer numbers). The fractional polynomial (34) can
then be put in the form
( ) (38)
or also by making the change of variable
( ) ( )
( ) (39)
The resulting polynomial is integer and of order ( ) but includes only ( )

parameters. So, there are powers of whose coefficients are zero. These constitute
constraints to be checked when one chooses the poles of the integer polynomial (39). In
addition, to ensure the stability of the closed-loop system, poles must verify the Matignon's
stability conditions given in Theorem 1. To illustrate the solution, consider the simple
example of a first order system and ( and ). In this case, the
integer polynomial (39) is written as
( ) (40)
This polynomial is thus of order where the coefficient of is zero. Therefore, we must
choose three poles whose sum is zero. But, when ( and ) the integer
polynomial (39) becomes
( ) (41)
In this case, we must choose four poles so that the coefficients of and must be zero.
The computation of the fractional PI-state feedback is summarized in the following

algorithm.
Consider the state space model of an integer order system for which the characteristic
polynomial is written as
( )
Step 1. Compute the linear transformation that gives the controllable canonical form.
Matrix can be computed as follows
( ) ( )
is a row vector
is the controllability matrix
is a unit vector
Step 2. Choose the fractional order variable , and must be integers.
Step 3. Write the integer order polynomial, representing the closed-loop characteristic
polynomial, in the form
( ) ( )
( )
Step 4. Select the powers of the variable for which the coefficient is zero.
Step 5. Choose poles of the integer polynomial of Step with respect the
constraints of Step , and with respect the Matignon's stability conditions.
Step 6. The feedback gain vectors and of the PI-state feedback ( ) ( )

( ( )) that place the poles of the system ( ̇ ) in closed-loop are
given by
( ) ( ) ( )( )
{
3. Bode's Ideal Transfer Function Based Design of Fractional

Pi-State Feedback
We presented, in Section 2, the generalization of the integer PI-state feedback. The pole
placement design method, although it can be used to determine the feedback gains and ,
is not practical because the non-integer order must be rational and its value is not chosen
according to a specified objective. In addition, the poles of the corresponding integer
polynomial are not arbitrary since they must verify constraints because some of the
polynomial coefficients are zero. In this section, another method based on the Bode's ideal
transfer function is proposed.
Theorem 3. Let's consider a controllable LTI integer order SISO system described by
Equation (9) and let the control law given by Equation (23). Vectors and that can
arbitrarily place the closed-loop characteristic polynomial written in the form
( ) ( )( ) (43)
are given by
( ) ( )( )
{ (43)
being the linear transformation that gives the controllable canonical form.
Proof. The closed-loop characteristic polynomial of the state space model (12) using the
fractional control law (26) being calculated by ( ) (see Equation (33), it is given by
( ) ( ) ( )
( ) ( ) (44)
It can be written as
( ) ( ) ( ) (45)
where
( ) ( ) ( ) ( ) (46)
( ) (47)
( ) and ( ) are two integer polynomials of order and , respectively.
and are then calculated as follows. First we require
(48)
( ) of Equation (46) can thus be written in the form
( ) ̃ ( ) (49)
where
̃ ( ) ( ) ( ) ( ) (50)
The closed-loop characteristic polynomial (Equation (45)) becomes
( ) ̃ ( ) ( ) ( )̃ ( ) (51)
̃ ( ) and ( ) have now the same order ( ). Their respective coefficients ,

( ) and ,( ) are finally calculated so that
̃ ( ) ( ) ̃ ( ) (52)
̃ ( ) is an integer polynomial of degree ( ) which allows to choose ( ) poles of

the closed-loop characteristic polynomial ( ).
Let, then, ( ) the desired closed-loop polynomial given by
( ) ( ) ∏( )
( )( ) (53)
( ) consists of two polynomials. A first one is integer and made from ( ) integer
poles and the second one is a non-integer polynomial ( ). Thus, when the pole ( )
is dominant, the open-loop system will be similar to the Bode's ideal function [9]. Pole ( )
is chosen to impose the dynamics of the closed-loop step response and the non-integer order
(with ) is chosen to impose its overshoot [7].
Identification term by term between ( ) and ̃ ( ), on one hand, and ( ), on
other hand, and taking into account Equation (48), gives
(54)
and
(55)
As (Equation (48)) and (Equation (54)), and are given by
( ) ( )( )
{ (56)
The pole placement of the closed-loop for the original system (9) is achieved by using the
gain vectors
and (57)
This completes the proof.
The following algorithm, for the computation of the fractional PI-state feedback, results
directly from the proof of the theorem.
Let the state space model of an integer order system for which the characteristic
polynomial is written as
( )
Step 1. Compute the linear transformation that gives the controllable canonical form.
Matrix can be computed as follows
( ) ( )
is a row vector
is the controllability matrix
is a unit vector
Step 2. Choose the fractional order polynomial
Step 3. Choose the ( ) poles , of the integer part of the desired

closed-loop polynomial.
Step 4. Put the desired closed-loop polynomial in the form
( ) ( )( )
Step 5. The feedback gain vectors and of the PI-state feedback ( ) ( )

( ( )) that place the poles of the system ( ̇ ) in closed-loop are
given by
( ) ( )( )
{
5. Inverted Pendulum-Cart System Modeling and Swing-Up

Control
The inverted pendulum problem is a challenging problem in control theory and has been
studied extensively in the control literature [11, 16, 17, 45]. It is a well established benchmark
problem that provides many challenging issues in control design. The system is nonlinear,
unstable, non-minimum phase and underactuated. The dynamics are quite similar to two-
wheeled mobile robots, flexible link robot, biped robot limbs, missiles and so on. Because of
their nonlinear nature, pendulums have maintained their usefulness and they are now used to
illustrate many of the ideas emerging in the field of non-linear control. The control of the
inverted pendulum can be divided into three aspects. The first aspect that is widely researched
is the swing-up control of the inverted pendulum. The second aspect is the stabilization of the
inverted pendulum. The third aspect is the tracking control of the inverted pendulum. In
practice, stabilization and tracking control are more useful for applications.
The pendulum-cart system is shown in Figure (4) and consists of
 A cart moving along a line on two rails of limited length m,

 A pendulum hinged in the cart by means of ball bearings rotating freely in the plane
containing this line,
 A cart driving device containing a DC motor, a DC amplifier and a pulley-belt
transmission system.
The control input is the supply voltage of the DC motor and the outputs are the cart
position and the pendulum angle .
The complete model describing the dynamic behaviour of the pendulum-cart system is
given by [4, 5]
̇ ( ) ̇ ( ) ̇
{ ̈ (58)
̈ ( ) ̇ ( )
where ( ) and ( ).
Figure 4. Inverted pendulum-cart system.
To avoid complicating the model of the pendulum-cart system, the DC motor dynamics
are neglected. The relationship between the force applied on the cart and the supply voltage
of the DC motor is a simple linear relationship . (where ). The pendulum
parameters are:
is its mass, is the cart mass, is the pendulum half
length, is the moment of inertia, is the friction coefficient
of the cart wheels, is the rotation friction coefficient of the pendulum,
and is the acceleration constant due to the gravity.
Swinging up an inverted pendulum will return it from the stable equilibrium point in
position to an unstable equilibrium point in upright position or . The
swing-up strategy used here is based on the Lyapunov stability theorem that allows the
convergence of the system energy to its energy value in an upright position. The Lyapunov
function used is the difference between the mechanical energy of the pendulum and the
desired mechanical energy . The swing up control is thus given by ̇ ( ) [4].
is a design parameter.
The swing-up control is implemented using a PC Pentium , , the sampling time is
and the design parameter . A saturation block between Volts is added
to protect the DC motor against the control effort. Figure (5) shows the evolution of the cart
position , the pendulum position and the control voltage .
The results obtained show that is sufficient to return the pendulum from its
stable equilibrium position to the unstable equilibrium point , respecting the
limits of the rail length m, and the control is less than Volts.
0.1
xp (m)
0
-0.1
0 2 4 6 8 10 12 14
Time
Theta (rad)
5
0
0 2 4 6 8 10 12 14
Time
Control (Volt)
0.5
-0.5
0 2 4 6 8 10 12 14
Time
Figure 5. Experimental results of the swing-up control.
5. Implementation of the Fractional PI-State Feedback

to Stabilize the Inverted Pendulum-Cart System
To test the theoretical results elaborated in Sections and , the linearized model of the
inverted pendulum-cart system around is used. It is given by
[ ( ) ( )] [ ]
* + (59)
The state vector is
̇ ̇ (60)
numerical values of and are
[ ], [ ] (61)
The system is unstable since its poles are , and .

To implement the fractional PI-state feedback control on the inverted pendulum-car
system, all the state variables, the cart position , the pendulum position , the cart linear
velocity ̇ and the angular velocity of the pendulum ̇ must be measured. These two last
measurements, being not available on the real system, we have used a differentiator followed
by a low-pass filter whose transfer functions are
(62)
Furthermore, The association between swing-up control and stabilization control requires
adding a suitable switching law [5, 45]. In our case, the switching is applied when the
absolute value of the angle is close to .
5.1. Implementation of the Fractional PI-State Feedback Obtained Using

the First Method
The linear model of the inverted pendulum-car system being unstable and as and
we choose , the method requires choosing poles of the integer
polynomial corresponding to the closed-loop characteristic polynomial. The integer
polynomial is given by
( ) (63)
This polynomial is of order , however there are only coefficients. The

zero coefficient is that corresponding to . So, the chosen poles must verify this constraint
∑ (64)
There are several ways to choose these poles. Equations (65) show the solution adopted
{ (65)
( )
The numerical values are
{ (66)
The numerical value of the parameter representing the DC motor model is ,

and by using the algorithm summarized in Section , gain vectors and of the fractional
PI-state feedback control are
{ (67)
Figure (6) shows the experimental results obtained. This figure shows the effectiveness of
the fractional PI-state feedback control on the experimental test bench. Indeed, it maintains
the stability of the pendulum in its unstable equilibrium position and controls the cart
position at . In addition, the control law values do not exceed Volts.
0.1
xp (m)
0
-0.1
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
0
0 5 10 15 20 25 30 35 40
Time
Control (Volt)
1
0
-1
0 5 10 15 20 25 30 35 40
Time
Figure 6. Experimental results swing-up + stabilization control using the first method.
0.2
xp (m)
0.1
0
-0.1
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
0
0 5 10 15 20 25 30 35 40
Time
Control (Volt)
2
1
0
-1
0 5 10 15 20 25 30 35 40
Time
Figure 7. Experimental results with variation of of the cart mass, using the first method.
Figure (7) shows the influence of the variation of of the cart mass (this perturbation
is obtained by adding a small mass of aluminum on the cart just after the end of the swinging
up step). Figure (8) shows the results obtained when the pendulum is subjected to disturbance
in the form of forces applied on it (by hitting the pendulum with a pen at different times).
Figures (7) and (8) show the robustness of the stabilization by the fractional PI-state
feedback. Indeed, despite the variation of of the cart mass and external disturbances on
the pendulum, it remains stable.
0.1
xp (m)
-0.1
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
0
0 5 10 15 20 25 30 35 40
Time
Control (Volt)
-2
0 5 10 15 20 25 30 35 40
Time
Figure 8. Experimental results with disturbance on the control at , and , using the first
method.
5.2. Implementation of the Fractional PI-State Feedback Obtained by Using

the Second Method
For the purpose of fractional order PI-state feedback control design, the linearized model
of the inverted pendulum-cart system (Equation 59) is considered again. The linear model
being of order , to achieve the closed-loop characteristic polynomial (Equation (43)), three
poles , and of the integer order polynomial are chosen. An
additional pole at of the fractional order polynomial is considered. The fractional
order considered is .
Gain vectors and of the fractional order PI-state feedback control obtained by using
the algorithm summarized in Section , are
{ (68)
Figure (9) shows the experimental results obtained. This figure shows the effectiveness of
the fractional PI-state feedback control on the experimental test bench. Indeed, it maintains
the stability of the pendulum in its unstable equilibrium position and controls the cart
position at .
0.1
xp (m)
0
-0.1
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
6
4
2
0
0 5 10 15 20 25 30 35 40
Time
Control (Volt)
2
1
0
-1
0 5 10 15 20 25 30 35 40
Time
Figure 9. Swing-up and fractional order PI-state feedback control experimental results, using the second
method.
0.4
xp (m)
0.2
0
-0.2
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
6
4
2
0
0 5 10 15 20 25 30 35 40
Time
Control (V)
-2
0 5 10 15 20 25 30 35 40
Time
Figure 10. Fractional order PI-state feedback control experimental results with mass variation,
using the second method.
Figures (10) and (11) show the experimental results obtained with disturbance. Figure
(10) shows the influence of the variation of of the cart mass. Figure (11) shows the
results obtained when the pendulum is subjected to disturbance applied at and
. Figures (10) and (11) show, in this case also, the robustness of the stabilization by
the fractional PI-state feedback. Indeed, despite the variation of of the cart mass and
despite external disturbances on the pendulum, it remains stable.
0.2
xp (m)
0.1
0
-0.1
0 5 10 15 20 25 30 35 40
Time
Theta (rad)
0
0 5 10 15 20 25 30 35 40
Time
Control (Volt)
2
0
-2
0 5 10 15 20 25 30 35 40
Time
Figure 11. Fractional order PI-state feedback control experimental results with disturbance on the
pendulum at and , using the second method.
Conclusion
Two new pole placement fractional PI-state feedback designs algorithms are presented in
this book chapter. The ﬁrst one consists in choosing poles of an integer polynomial
corresponding to the closed-loop fractional characteristic polynomial. Matignon’s stability
condition can thus be used to ensure the stability of the closed-loop. Nevertheless, the non-
integer order must be rational. In addition, the poles of the integer polynomial are not
arbitrary since they must verify constraints because some of the polynomial coefficients are
zero. The second method is simple to implement, it consists in choosing poles of an
integer polynomial and one pole of a fractional polynomial. A judicious choice of these poles
provides a closed-loop particular response obtained when the Bode’s ideal transfer function is
used, that is, an overshoot of the step response imposed by the fractional order and its
dynamics by the pole of the fractional polynomial. These methods are then used to stabilize
an inverted pendulum-cart system. The implementation on an experimental test-bed gave
good results, especially in terms of stability, accuracy and robustness with respect to the
variation of the cart mass as well as external disturbances applied on the pendulum.
Acknowledgments
This project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz
University, Jeddah, under grant no Gr/34/5. The authors, therefore, acknowledge with thanks
DSR technical and financial support.
References
[1] Abdelaziz T.H.S. and Valasek M., (2003). A direct algorithm for pole placement by
state derivative feedback for single-input linear systems. Acta Polytechnica, 43, 52-60.
[2] Abdelaziz T.H.S. and Valasek M., (2004). Pole-placement for SISO linear systems by
state-derivative feedback. IEE Proceedings Control Theory and Application, 151,
377-385.
[3] Abdelaziz T.H.S. and Valasek M., (2005). State derivative feedback by LQR for linear
time-invariant systems. Proceeding of 16th IFAC World Congress, Prague, Czech
Republic, July 3-8.
[4] Astrom K J, and Furuta K., (1996). Swinging up a pendulum by energy control.
Proceeding of IFAC 13th IFAC World Congress, San Francisco, CA, USA, June 30-
July 05.
[5] Astrom K.J., Aracil J., and Gordillo F., (2008). A family of smooth controllers for
swinging up a pendulum. Automatica, vol. 44, 1841-1848.
[6] Axtell, M., and Bise, E. M., (1990). Fractional calculus applications in control systems.
In Proceedings of the IEEE Nat. Aerospace and Electronics Conference, New York, 8
(3-4), 563-566.
[7] Barbosa R.S., Machado J.A.T., and Ferreira I.M., (2004). Tuning of PID controllers
based on Bode's ideal transfer function. Nonlinear dynamics, vol. 38, 305-321.
[8] Bettayeb M., and Djennoune S., (2008). New results on controllability and
observability of fractional dynamical systems. Journal of Systems Engineering and
Electronics, vol. 14 (9-10), 376-381.
[9] Bettayeb M., C. Boussalem, Mansouri R. and Al-Saggaf U. M., (2014), Stabilization of
an inverted pendulum-cart system by fractional PI-state feedback, ISA transactions, vol.
53 (2), 508-516.
[10] Bode H.W., (1945). Network Analysis and Feedback Amplifier Design. Van Nostrand,
New York.
[11] Chung C.C. and Hauser J., (1995). Non linear control of a swinging pendulum.
Automatica, vol. 31 (6), 851-862.
[12] Das S. Saha S. Das S. and Gupta A., (2011). On the selection of tuning methodology of
FOPID controllers for the control of higher order processes. ISA Transactions, vol. 50,
376-388.
[13] Dorcak, L., Petras, I., and Kostial, I., (2000). Modeling and analysis of fractional-order
regulated systems in the state space. In Proceedings of ICCC'2000, vol.45, High Tatras,
Slovak Republic, 185-188.
[14] Farges C. Moze M. and Sabatier J., (2009). Pseudo-state feedback stabilization of
commensurate fractional order systems. Proceedings of the 10th European control
conference, Budapest, Hungary, August 23-26.
[15] Farges C. Moze M. and Sabatier J., (2010). Pseudo-state feedback stabilization of
commensurate fractional order systems. Automatica, vol. 46, 1730-1734.
[16] Furuta K. and Yamakita, M. (1991). Swing up control of inverted pendulum. IEEE
Proceedings 1991 International Conference on Industrial Electronics, Control and
Instrumentation, Kobe, Japan, October 28-November 1.
[17] Furuta K., Yamakita M. and Kobayashi S. (1992). Swing-up control of inverted
pendulum using pseudo-state feedback. Journal of Systems and Control Engineering,
vol. 206, 263-269.
[18] Guo G. and Ma Z., (2006). State-PID feedback control with application to a robot
vibration absorber. International Journal of Modelling, Identification and Control, vol.
1 (1), 38-43.
[19] Hotzel R., and Fliess, M., (1998). On linear system with a fractional derivation:
introductory theory and examples. Mathematics and Computers in Simulation, vol. 45,
385-395.
[20] Kilbas A.A., Srivastava H.M. and Trujillo J.J., (2006). Theory and Applications of
Fractional Differential Equations. Elsevier, North-Holland.
[21] Leitmann G. and Reithmeier E. (2003) Robust vibration control of dynamical system
based on the derivative of the state. Archive of Applied mechanics, Vol. 72, 856-864.
[22] Magin R.L., (2006). Fractional Calculus in Bioengineering. Begell House Punlishers,
Inc., Connecticut.
[23] Machado, J.T., Kityakova V. and Mainardi F., (2011). Recent History of fractional
calculus. Communications in Nonlinear Science and Numerical Simulation, vol. 16 (3),
1140-1153.
[24] Manabe, S., (1963). The system design by the use of a model consisting of a saturation
and non-integer integrals. ETJ of Japan, vol. 8 (3-4), 147-150.
[25] Mansouri R., Bettayeb M., and Djennoune S., (2007). Multivariable fractional order
system approximation using derivative representation. International Journal of Applied
Mathematics, vol. 20 (7) 983-1003.
[26] Mansouri R., Bettayeb M., and Djennoune S., (2008). Non integer order system
approximation by an integer reduced model. in a special issue on fractional calculus in
the Journal Européen des Systèmes Automatisés, vol. 42, 689-700.
[27] Mansouri R., Bettayeb M., and Djennoune S., (2010). Multivariable fractional system
approximation with Initial conditions using integral state space representation. Int.
Journal on Computers and Mathematics with Applications, vol.59 (5), 1842-1851.
[28] Matignon, D., (1996). Stability results on fractional differential with application to
control processing. In IAMCS, IEEE SMC Proceedings Conference, Lille, France,
963-968.
[29] Matignon, D., and d'Andréa-Novel, B., (1996). Some results on controllability and
observability of finite dimensional fractional differential systems. In IAMCS, IEEE-
SMC Proceedings Conference, Lille, France, pp. 952-956.
[30] Matignon D., (1998). Stability properties for generalized fractional differential systems.
In Systèmes différentiels fractionnaires (Paris 1998), vol.5 of ESAIM: Proceedings,
145–158, Society for Industrial and Applied Mathematics, Paris, France.
[31] Miller K.S. and Ross B., (1974). An introduction to the Fractional Calculus and
Fractional Differential Equations. A Wiley Interscience Publication.
[32] Monje C.A., Vinagre B.M., Feliu V. and Chen Y.Q., (2008). Tuning and auto-tuning of
fractional order controllers for industry applications. Control Engineering Practice, vol.
16, 798–812.
[33] Monje, C.A., Chen, Y., Vinagre, B.M., Xue, D., and Feliu-Batlle, V., (2010).
Fractional-order Systems and Controls: Fundamentals and Applications. Springer
Verlag.
[34] Nia S.H.H., Sierociuk D., Calderon A.J. and Vinagre B.M., (2010). Augmented System
Approach for Fractional Order SMC of a DC-DC Buck Converter. Proceedings of the
4th IFAC Workshop on Fractional Differentiation and its Applications, Badajoz, Spain,
october 18-20.
[35] Ogata, K., (1997). Modern Control Engineering. Third Edition, Prentice Hall.
[36] Oldham K.B. and Spanier J., (1974). The Fractional Calculus. Academic Press, New
York and London.
[37] Olgac N., Elmali H., Hosek M. and Renzulli M., (1997). Active vibration control of
distributed systems using delayed resonator with acceleration feedback. Journal of
Dynamic Systems, Measurement and Control, vol.119 (3), 380-389.
[38] Oustaloup A., (1991). La Commande CRONE. Hermès Edition, Paris.
[39] Oustaloup, A., (1995). La Dérivation Non Entière: Théorie, Synthèse et Applications.
Hermès Edition, Paris, France.
[40] Oustaloup A. and Mathieu B., (1999). La Commande CRONE du Scalaire au
Multivariable. Hermès Editeur, Paris.
[41] Podlubny I., (1999). Fractional Differential Equations. Academic Press, San Diego.
[42] Poblubny I., (1999). Fractional-Order System and . JIEEE Transactions on
automatic control, vol. 44 (1), 208-214.
[43] Raynaud, H. F., and Zergainoh, A., (2000). State space representation for fractional
order controllers. Automatica, vol. 36, 1017-1021.
[44] Samko S.G., Kilbas A.A. and Marichev O.I., (1993). Fractional Integrals and
Derivatives. Gordon and Breach Science Publishers.
[45] Srinivasan B., Huguenin P., and Bonvin D., (2009). Global stabilization of an inverted
pendulum - Control stategy and experimental verification. Automatica, vol. 45, 265-
269.
[46] Sierociuk D. and Vinagre B.M., (2010). State and Output Feedback Fractional Control
by System Augmentation. Proceedings of the 4th IFAC Workshop on Fractional
Differentiation and its Applications, Badajoz, Spain, october 18-20.
[47] Tavakoli-Kakhki M., and Haeri M., (2010). The minimal state space realization for a
class of fractional order transfer functions. SIAM Journal on Control and Optimization,
vol. 48 (7), 4317-4326.
[48] Tavakoli-Kakhki M, Haeri M, and Tavazoei M.S, (2011). Notes on the state space
realizations of rational order transfer functions. IEEE Transactions on Circuits and
Systems I: Regular Papers, vol. 58 (5), 1099-1108.
[49] Vinagre, B. M., Monje, C. A., and Caldero, A. J., (2002). Fractional order systems and
fractional order actions. Tutorial Workshop # 2: Fractional Calculus Applications in
Automatic Control and Robotics, 41st IEEE CDC, Las Vegas.
[50] Wiboonjaroen W., and Sujitjorn S., (2011). Stabilization of an Inverted Pendulum
System via State-PI Feedback. International Journal of Mathematical models and
methods in applied sciences, vol. 5. (41), 763-772.
[51] Zeng Q., Cao G., and Zhu X., (2005). Research on controllability for a class of
fractional-order linear control systems. Journal of Vibration and Control, vol. 14 (9-
10), 1531-1541.
Chapter 5
FROM THE FORMAL CONCEPT ANALYSIS

TO THE NUMERICAL SIMULATION OF THE THERMAL
DIFFUSION PHENOMENA IN A FINITE MEDIUM
Riad Assaf1,2,, Roy Abi Zeid Daou2,†,

Xavier Moreau1,‡ and Fady Christophy1,2,#
1
University of Bordeaux, IMS Laboratory, Talence, Bordeaux, France
2
Lebanese German University, Faculty of Public Health,
Biomedical Technologies department, Sahel Alma, Jounieh, Lebanon
Abstract
This chapter shows the use of the fractional order approach in the study of the thermal
interface behavior in view to facilitate the identification of this phenomenon and the
optimization of its process control. After a short introduction, the main parameters are defined
then the semi-infinite plane case study is considered. A representative model is developed
based on the system equations. The operational, frequency and time responses are studied.
The case study of a finite plane follows using the same methodology, nevertheless exact
analytical solutions in the time domain don’t exist; their approximations will be treated in
other publications.
1. Context and Problematic

In general, the development of diffusion phenomena equations naturally leads to
fractional order systems (Podlubny, 2002). There are many examples in the fields of physics
such as electrochemistry (Kuhn, 2005), electromagnetism (Benchellal, 2005) (Canat, 2005),
or thermal (Agrawal, 2004) (Battaglia, 2001) (Kusiak, 2005).

E-mail addresses: riad.assaf@u-bordeaux.fr; r.assaf@lgu.edu.lb.
†
E-mail address: r.abizeiddaou@lgu.edu.lb.
‡
E-mail address: xavier.moreau@u-bordeaux.fr.
#
E-mail addresses: fady.chrystophy@u-bordeaux.fr; f.christophy@lgu.edu.lb.
94 Riad Assaf, Roy Abi Zeid Daou, Xavier Moreau et al.
Actually, the recognition of fractional order behavior when identifying physical systems
flourished starting the second half of the twentieth century (Oldham, 1974). One of these
fields where the fractional order integration can show up is the thermal diffusive interfaces. In
fact, for some special conditions, an integration of order 0.5 appears when identifying such
system (Abi Zeid Daou, 2012, Assaf, 2012). The objective of this chapter is to study the
presence of the fractional order and its effects on the system behavior, and to determine the
relation between the applied flux, the conduction and the point of measurement of the
temperature in a plane medium.
This objective is a milestone towards the definition of valid approximations for reliable
numerical simulations of the thermal diffusion phenomena in a finite medium where exact
analytical time response solutions don’t exist. For instance, one can define conditions for
which a finite medium can be considered as a semi-infinite medium, and can define other
conditions for which analytical approximations generate accurate time response simulations.
The understanding of the thermal diffusion phenomena, hence the establishment of a
validation model has two main goals:
- As part of the identification process: to facilitate the definition of the model structure,
and to pass from a black box approach to a gray box approach.
- As part of a process control model: to deduct from the validation model not only a
nominal model of synthesis, but also structural and parametric uncertainties.
This chapter contains six main paragraphs other than the Abstract and the Introduction at
the beginning; the Conclusion and the References at the end. A reminder of the main physical
characteristics needed to describe the thermal diffusion are found in Para 2, in Para 3 a model
of the diffusion phenomena in the semi-infinite medium is developed, then the relative
operational, frequency and time responses are studied in Para 4. In Para 5 a model of the
diffusion phenomena in the finite medium is developed, then only the corresponding
operational and frequency responses are studied in Para 6 since no exact analytical solutions
exist to represent the time response for the finite medium. In Para 7 one can find some step
responses using Simulink, it opens the door for the need for simple and accurate
approximations.
2. Parameters and Definitions

The main thermal parameters and definitions characterizing any thermal interface are
presented in this paragraph.
 Heat flow in the direction is represented by ( ).

 Heat flux in the direction is represented by ( ⁄ ).
 Thermal conductivity ( ) [Özişik, 1980] is the material property which
controls the rate of heat flow in the medium. For aluminum .
 Homogeneous and Isotropic medium [Özişik, 1980] is a material in which thermal
conductivity is independent of direction.
 Density of a material is represented by ( ⁄ ). For aluminum
⁄ .
From the Formal Concept Analysis to the Numerical Simulation … 95
 Specific heat capacity ( ) is [Schneider, 1957] the heat energy

required to raise the temperature of a unit weight of the material by one degree. For
aluminum .
 Thermal diffusivity ( ⁄ ) of a medium [Özişik, 1980] is associated with the
speed of propagation of heat into the solid during changes of temperature with time.
For aluminum ⁄ .
 Thermal effusivity ( ) is [de Wit, 2009] a measure of the material’s
ability to exchange thermal energy with its surroundings. For aluminum
√ .
 Volumetric heat capacity ( ). is a variant of the
volumetric heat capacity [US Army, 1988]; it describes the ability of a given volume
(crossed by a heat flux through its area along its length ) of finite medium to store
internal energy while undergoing a temperature change.
3. Semi-Infinite Plane
We consider a semi-infinite, homogeneous, isotropic, one-dimensional plane medium of
conductivity , diffusivity and of initial temperature zero all over the medium (Figure 1)
[Sabatier, 2008]. A heat flux ( ) is normally applied on its surface. Consequently, a
temperature gradient ( ) appears, which is a function of the independent time variable
as well as the abscissa of the temperature measurement point, inside the medium.
Figure 1. Diffusion in a semi-infinite plane medium.
3.1. System Definition
The one-dimensional heat conduction equation (Özişik, 1985) is given by a partial

differential equation:
( ) ( )
. (1)
Since the initial temperature is zero all over the medium:
( ) (2)
The Fourier law states the boundary condition (Özişik, 1985) where the flux is applied,
the second condition is trivial:
( )
( ) ;
(3)
( ) . (4)
3.2. System Resolution
The use of the fractional calculus to resolve the problems of the diffusive interfaces dates
back to the beginning of the 20th century [Battaglia, 2001]. The system of (Eq. 1-4) represents
the relations that model this interface knowing that the input signal is the flux ( ) whereas
the output signal is the temperature ( ) as shown in Figure 1.
The initial condition of the temperature being zero, Laplace transform * + of (Eq. 1) is:
̅( )
̅( ) where ̅ ( ) * ( )+. (5)
This is a differential equation of the 2nd order in x. Its general solution is:
̅( ) ( )
, (6)
leading to the characteristic equation:
( ) , (7)
which has the following 2 roots ( ) and ( ):
√ * + (8)
Hence the solution of (Eq. 5) is:
̅( ) ( ) √ ⁄ ( ) √ ⁄
(9)
Resolving (Eq. 9) for the boundary conditions (Eq. 3-4) gives the values of ( ) and
( ), let:
( )
{ ( ) ̅( ) (10)
√
Taking into consideration the definitions of the characteristic parameters and the outcome
of (Eq. 10) then resolving (Eq. 9), the solution of (Eq. 5) is found to be:
̅( ) √ ⁄
̅( ). (11)
√
Under the assumption of an initial condition of the temperature being zero, the transfer
function ( ) between the input flux ̅( ) and the output temperature ̅ ( ) is:
̅( ) √ ⁄
( ) , (12)
̅( ) √
or rewritten as:
( ) ( ) ( ), (13)
with the following considerations:
̅( )
̅( ) √
̅( )
( ) ̅( ) (14)
̅( ) . /
{ ( ) ̅( )
and defining a cutoff frequency as:
⁄ . (15)
Consequently, the transfer function ( ) of the system can be seen as three cascading
functions represented in Figure 2.
H  x, s 
 s  s0.5T 0, s  T 0, s  T  x, s 

H0 I 0.5
s  E  x, s 
Figure 2. Block diagram illustrating the transfer function ( ) of the semi-infinite medium.
The function depends on the medium physical characteristics, it links the input flux
)
̅( and the fractional derivative of order of the temperature at .
Then, ( ) is a fractional integrator of order that gives the temperature at .
The last function ( ) generates the output temperature ̅ ( ) at any point of the
semi-infinite medium.
This new approach allows understanding the link existing between the two kinds of study
encountered in the dynamics of thermal systems, i.e.:
- When the boundary condition at is the flux ̅( ); it is recognized as Neumann

(second-type) boundary condition.
- When the boundary condition at is the temperature ̅ ( ); it is recognized as
Dirichlet (first-type) boundary condition.
In both cases, for a semi-infinite medium the transfer ( ) is the same.
4. Responses in the Semi-Infinite Plane

4.1. Asymptotic Behavior Analysis at
At , ( ) ; thus the transfer function ( ) becomes:
( ) ( ) (16)
which asymptotic behaviors are given by:
( ) ( )
{ (17)
( ) ( )
The fractional integration behavior of order between the input flux ̅( ) and the
̅
temperature ( ) is at mid-way between an integration behavior of order which is
characteristic of a capacitive behavior, and a proportional behavior (integration of order )
characteristic of a resistive behavior.
The analysis of the influence of ( ) on the transfer function ( ) through its

asymptotic behaviors leads to:
( )
{ . /
(18)
( )
Hence,
( ) ( )
{ . / . (19)
( )
The same asymptotic behavior is noticed at any place of the semi-infinite medium.
4.3. Frequency Response Analysis at
The frequency response ( ) is given by:
( ) ( ) ( ). (20)
At , the block ( ) thus the transfer function ( ) becomes:
( ) ( ). (21)
The gain and the phase of the transfer function ̅ ( ) are:
| ( )|
{ . (22)
( ( ))
Bode diagrams and Black-Nichols diagram of ( ) are represented in (Figure 3) and

(Figure 4), respectively.
At , the 3rd block ( ) becomes:
. /
( ) (23)
√ √
Knowing that . / . / , the block ( ) is written as:
. / . /
( ) (24)
thus the gain and the phase of this block are:
0
-40
1
Gain (dB)
-80 5
10
-120 50
100
-160
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
0
-45
Phase (deg)
1
-90 5
10
-135 50
100
-180
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 3. Bode diagrams of ( ) for different positions of the temperature sensor in a semi-infinite
aluminum medium.
40
0
20 1
5
10
0
50
100
-20
Gain (dB)
-40
-60
-80
-100
-120
-270 -225 -180 -135 -90 -45 0 45
Phase (deg)
Figure 4. Black-Nichols plots of ( ) for different positions of the sensor in a semi-infinite

aluminum medium.
. /
| ( )|
{ . (25)
( ( )) . /
- At the low frequencies, the behavior of this block is similar to the one observed
when :
| ( )|
{ . (26)
( ( ))
- At the medium frequencies, especially when ⁄ , the cutoff

frequency:
| ( )| ( )
{ . (27)
( ( ))
- At the high frequencies, the behavior of this block is like:
| ( )|
{ . (28)
( ( ))
Now, we can look deeply into the transfer function:
. / . /
̅( ) . (29)
( )
The asymptotic behaviors of ( ) are found to be:
- At low frequencies:
| ( )|
{ . (30)
( ( ))
The behavior is similar to the one observed when .

- At high frequencies:
. /
| ( )|
{ . (31)
( ( )) ( . / )
Bode diagrams and Black-Nichols plots in a semi-infinite aluminum medium are merged
in Figure 3 and Figure 4 with various values of . The fractional integration behavior of
order is present all over the frequency range for , it disappears gradually at higher
frequencies when increases.
4.5. Time Response Analysis
When analyzing this system in time domain, a special attention should be taken. In fact,
because of the presence of the exponent in the expression of ( ), the inverse Laplace
transfer in simulation is not always possible when taking into consideration special inputs
̅( ). (Eq. 32) shows the inverse Laplace transform * + of the output temperature.
( ) * ̅( )+ * ( ) ̅( )+ (32)
Nevertheless, some time domain simulations could be made without using

approximations. In fact, three time domain simulations for three well known input types are
proposed: the impulse, the step and the sinusoidal inputs.
4.5.1. Impulse Response
The impulse response is obtained by substituting ( ) by its value shown in (Eq. 12)
and ̅( ) by * ( )+ . Accordingly, the impulse response is given by:
( ) √ ⁄ √ ⁄
( ) . / (33)
√ √
Knowing that (Battaglia, 2008):
√ ( )
. / (34)
√ √
where , the impulse response can be presented as follows:
( )
( ) (35)
√
Figure 5 shows the outputs for the impulse at different locations of and different values
of in a semi-infinite aluminum medium.
4.5.2. Step Response
The step ( ) response of magnitude is obtained by substituting ( ) by its value

shown in (Eq. 12) and ̅( ) by * ( )+ ⁄ . Accordingly, the step response is given by:
( ) √ ⁄ √ ⁄
( ) . ⁄ / (36)
√
-5 -5
x 10 x 10
1.4 1.4
0 1
1 10
1.2 1.2
5 20
10 40
1 50 1 80
Temperature Variation (°C)
100
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 10 20 30 40 50 60 0 20 40 60 80 100
(a): Time (s), T0=0°C (b): Sensor position x (mm), T0=0°C
Figure 5. Impulse responses of ( ) for (a) and ;

and (b) and .
Knowing that (Battaglia, 2008):
√ ( )
. ⁄ / √ . / (37)
√
where ( ) is the complementary error function and , the step response can be
presented as follows:
1 1
0 1
0.9 1 0.9 10
5 20
0.8 0.8
10 40
50 80

0.7 0.7
100
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 30 60 90 0 25 50 75 100 125 150 175 200
(a): Time (s), T0=0°C, Flux=2000W/m 2 (b): Sensor position x (mm), T0=0°C, Flux=2000W/m 2
Figure 6. Step responses of ( ) for (a) and ; and

(b) and .
( )
( ) ( √ . /) (38)
√ √
Figure 6 shows the outputs for the step function at different locations of and different
values of in a semi-infinite aluminum medium.
4.5.3. Stationary Harmonic Response
The entry at is supposed to be
( ) ( ). (39)
The system being supposed linear, the expression of the output temperature is given by:
( ) | ( )| . ( ( ))/ (40)
then replacing ( ) by its value in (Eq. 29)
. / . /
( ) ( ( )) (41)
Let
⁄ and ⁄ (42)
Then
( ) . / √ . /
√ (43)
25
1e-006
1e-005
20
0.0001
15
10
-5
-10
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Time (hour), T0=0°C, Flux=2000W/m 2
Figure 7. Harmonic responses of ( ) for ( ⁄ ) * + and ⁄ .
At :
( ) . / . /. (44)
Figure 7 illustrates the outputs for a harmonic function input for different values of
( ⁄ ) * + and for ⁄ in a semi-infinite aluminum medium.
At :
In order to make the analysis easier, let’s consider the temperature at as a

reference, thus we can write:
( ) ( ). (45)
Hence the temperature at will be written as:
( ) √ ( ).
√ (46)
For illustration purposes, let with . Figure 8 shows the outputs for
a harmonic function input for different values of * + in a semi-
infinite aluminum medium.
Figure 8. Harmonic responses of ( ) when , for ( ) ,( ) ,

( ) ,( ) ,( ) and ( ) .
 t 
0 x
Figure 9. Finite unidirectional plane of thickness L.

Note that, as long as the frequency of variation of the temperature is low relatively to
the diffusivity , the temperature variation is uniform and harmonic.
When the frequency of variation of the temperature increases, the diffusion depth of
the temperature decreases in the semi-infinite medium.
5. Finite Plane
We consider a finite, homogeneous, isotropic, one-dimensional plane medium of
thickness , conductivity , diffusivity , and of initial temperature zero all over the medium
(Figure 9). A heat flux ( ) is normally applied on its surface at ; it is assumed that
there are no losses at this point. Consequently, a temperature gradient ( ) appears, which
is a function of the independent time variable as well as the abscissa of the temperature
measurement point, with , - inside the medium.
5.1. System Definition
The one-dimensional heat conduction equation (Özişik, 1985) is given by a partial

differential equation:
( ) ( )
. (47)
Since the initial temperature is zero all over the medium:
( ) . (48)
The Fourier law states the boundary conditions (Özişik, 1985) where the flux is applied,
as well as at the second end:
( )
( )
(49)
( )
.
(50)
5.2. System Resolution
The system of (Eq. 47-50) represents the relations that model the diffusive interface
knowing that the input signal is the flux ( ) whereas the output signal is the temperature
( ) as shown in Figure 9.
The initial condition of the temperature being zero, Laplace transform * + of (Eq. 47) is:
̅( )
̅( ) where ̅ ( ) * ( )+ (51)
The solution of (Eq. 51) is similar to the previous case, hence:
̅( ) ( ) √⁄ ( ) √⁄
. (52)
Resolving (Eq. 52) for the boundary conditions (Eq. 49-50) gives the values of ( ) and
( ), let:
( ) ( ) ̅( )
√
{ (53)
√ √
( ) ( )
Taking into consideration the definitions of the characteristic parameters and the outcome
of (Eq. 53), thus:
√⁄
( ) √⁄ √⁄
̅( )
√
(54)
√⁄
( ) ̅( )
{ √ √⁄ √⁄
then the solution of (Eq. 51) is found to be:
( )√ ⁄ ( )√ ⁄
̅( ) ̅( ) . (55)
√ √⁄ √⁄
Under the assumption of an initial condition of the temperature being zero, the transfer
function ( ) between the input flux ̅( ) and the output temperature ̅ ( ) is:
( )√ ⁄ ( )√ ⁄
̅( )
( ) , (56)
̅( ) √ √ ⁄ √ ⁄
then, introducing the hyperbolic functions and the transfer function ( )

becomes:
̅( ) (( )√ ⁄ )
( ) , (57)
̅( ) √ . √ ⁄ / . √ ⁄ /
or rewritten as:
( ) ( ) ( ) ( ) , (58)
where:
̅( )
̅( ) √
̅( )
( ) ̅( )
̅( )
( ) ̅( ) (59)
(√ ⁄ )
̅( (√ ⁄ )
)
( ) ̅( ) (√ ⁄ )
{
with the following considerations:
. /
(60)
{ ( )
One can derive from (Eq. 60) that and are linked by the following relation:
. / . (61)
One can derive also the thermal diffusion time constant defined as:
. (62)
To clarify, the different temperatures found in the system (Eq. 59) are as follows:
- ̅( ) is the measured temperature at if the medium is considered to be

a semi-infinite medium.
- ̅( ) is the measured temperature at of the finite medium of length .
- ̅( ) is the measured temperature at , , -.
Consequently, the transfer function ( ) of the system can be seen as four cascading
functions represented in (Figure 10).
The function depends on the medium physical characteristics, it links the input flux
̅( ) and the fractional derivative of order of the temperature at .
Then, ( ) is a fractional integrator of order that gives the temperature at . Its
output is the temperature ̅ ( ).
The third block is the transfer function ( ) between the temperatures ̅ ( ) and
̅( ).
The fourth part is the transfer function ( ) having as input ̅ ( ) and as output
the temperature ̅ ( ).
H(x, s, L)
s0.5T(0, s, oo) T (0, s, oo) T (0, s, L) T (x, s, L)

H0 I
0.5
(s ) F(0, s L) G(x, s, L)
Figure 10. Block diagram illustrating the transfer function ( ) of the finite medium.
6. Responses in Finite Plane

At , ( ) ( ) , thus the temperature ̅ ( ) of the finite

medium of length is given by:
̅( ) ( ) ( ) ̅( ). (63)
The transfer function ( ) allows the analysis of the influence of the medium finite
character on the temperature at . As such, let the change of variable √⁄ .
Considering the definitions in (Para 2.) and (Eq. 60, 61), as a start the following is verified:
if then , as ( ) , thus ( ) , accordingly it is
found that:
( ) ( ) (64)
which is a result similar to what was found in a semi-infinite medium at .

Now consider when then , as ( ) , thus:
( ) , accordingly it is found that:
√ ⁄
( )
( ) (65)
√⁄
Once again, considering the definitions in (Para 2.) and (Eq. 60, 61), the (Eq. 65) is
rewritten as:
( ) . (66)
Thus for high instances, corresponding to low frequencies (i.e. ), the asymptotic
behavior of a finite medium is of capacitive type characterized by an integrator of order . It
is different than the behavior in a semi-infinite medium characterized by a fractional
integrator of order .
For low instances, corresponding to high frequencies, when then , as
( ) , thus:
( ) ( ) (67)
Accordingly an asymptotic fractional order behavior appears when between the

input flux and the temperature at the boundary where the flux is applied; in fact, it is the same
result that appears in the semi-infinite medium.
6.2. Asymptotic Behavior Analysis for
For high instances, corresponding to low frequencies, when then

( ) , thus:
( ) (68)
The same asymptotic behavior applies as at . It is of capacitive type characterized

by an integrator of order ; it is different than the behavior in a semi-infinite medium
characterized by a fractional integrator of order .
For low instances, corresponding to high frequencies, when then , as
√ ⁄
, thus ( ) ; it is resolved introducing (Eq. 61) a
√ ⁄
relation between and ; accordingly:
√ ⁄ √
( ) (69)
then,
√
( ) ( ) (70)
It is the same asymptotic behavior as it appears in a semi-infinite medium.
Now that the mathematical model is established, frequency domain analyses ( )

can be done. The resulting transfer function ( ) is the concatenation of four cascading
blocks as shown on the (Figure 10) and represented in the (Eq. 58-60). When the study is on
the boundary where the flux is applied, , the transfer function is expressed by:
( ) ( ) ( ) ( ) (71)
with ( ) , then ( ) reduces to::
( ) ( ) ( ) (72)
Bode diagrams of the classic fractional integrator of order 0.5 in a finite aluminum
medium are represented in Figure 11.
The frequency response study of the function ( ) allows understanding the
influence of the finite character of the medium on the temperature at . Actually Figure
12 represents Bode diagrams of this transfer for three different values of * +
in a finite aluminum medium.
40
0
Gain (dB)
-40
-80
0
-120
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
0
Phase (deg)
-45
0
-90
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 11. Bode diagrams of ( ) in a finite medium at .
Two asymptotic behaviors clearly appear:
At the low frequencies, a fractional integration behavior of order prevails. Actually,
, ( ) . /
| ( )| . /
{ (73)
( ( ))
t the high frequencies, a proportional behavior prevails. Actually, ,

( )
| ( )|
{ (74)
( ( ))
The transitional zone between the two asymptotic behaviors is four decades wide and it is
centered on the transitional frequency .
80
1
10
Gain (dB)
40 100
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
45
Phase (deg)
1
-45
10
100
-90
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 12. Bode diagrams of of ( ) for * + .
Now it is time to study the influence of each of the three cascading blocks on the global
transfer function ( ). For this purpose, for and , Figure 13 visualizes
the contribution of each block to the global response in a finite aluminum medium.
Two asymptotic behaviors clearly appear:
At the low frequencies, an integration behavior of order prevails. Actually,

, ( )
| ( )|
{ (75)
( ( ))
At the high frequencies, a fractional integration behavior of order prevails. Actually,
, ( ) . /
| ( )| . /
{ (76)
( ( ))
Here also, the transitional zone between the two asymptotic behaviors is four decades
wide and it is centered on the transitional frequency .
40
H(0,jw,1m)
0
H0I0.5(0,jw)
Gain (dB)
-40 F(0,jw,1m)
G(0,jw,1m)
-80
-120
-7 -6 -5 -4 -3 -2 -1 0 1
10 10 10 10 10 10 10 10 10
45
0
Phase (deg)
H(0,jw,1m)
-45
H0I0.5(0,jw)
-90 F(0,jw,1m)
G(0,jw,1m)
-135
-7 -6 -5 -4 -3 -2 -1 0 1
10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 13. Bode diagrams of ( ) ( ) ( ) and ( ).
For different values of * + , (Figure 14 represents the respective Bode

diagrams in a finite aluminum medium illustrating the translation of the transitional zone
towards the high frequencies when decreases.
40
0
Gain (dB)
-40
1
-80 10
100
-120
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
0
1
10
Phase (deg)
-45
100
-90
-135
-7 -6 -5 -4 -3 -2 -1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 14. Bode diagrams of ( ) for * + .

For different values of * + , Figure 15 represents respectively ( ) in

Black-Nichols plane in a finite aluminum medium.
6.4. Frequency Behavior Analysis for
For , ( ) . Thus it is interesting to begin with the study of its

frequency behavior in order to understand its influence on the global response ( ).
(Eq. 56-59) reformulate the expression of the third block:
( )√ ⁄ ( )√ ⁄
( ) (77)
√ ⁄ √ ⁄
Figure 16 represents Bode diagrams of the transfer of ( ) for a finite aluminum

medium of length and for four different values of the temperature sensor position,
let * + .
Two behaviors clearly appear:
At the low frequencies,
| ( )|
( ) { (78)
( ( ))
40
1
20 10
100
-20
Gain (dB)
-40
-60
-80
-100
-120
-270 -225 -180 -135 -90 -45 0 45
Phase (deg)
Figure 15. Black-Nichols plots of ( ) for * +

40
Gain (dB)
-40 50
10
-80 5
0
-120
-5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10
45
0
Phase (deg)
-45 50
10
-90 5
0
-135
-5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 16. Bode diagrams of ( ) for * + .
At the high frequencies,
. /
√ | ( )|
( ) { (79)
( ( )) . /
In order to make the analysis easier, a special frequency is introduced; it is defined as:
| ( )| ⁄√ (80)
The frequency decreases when increases. This is why, for a given frequency , the
gain and the phase are as smaller as is big as it is evident in Figure 16.
Now it is time to study the influence of each of the four cascading blocks on the global
transfer function ( ). For this purpose, for and , Figure 17
visualizes the contribution of each block to the global response in a finite aluminum medium.
Three behaviors clearly appear:
At the low frequencies, an integration behavior of order prevails similar to the one
found at . Actually, , ( )
| ( )|
{ (81)
( ( ))
40
Gain (dB)
H(5mm,jw,1m)
-40
H0I0.5(0,jw)
-80 F(0,jw,1m)
G(5mm,jw,1m)
-120
-7 -6 -5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
45
0
Phase (deg)
-45
-90
-135
-7 -6 -5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 17. Bode diagrams of ( ) ( ) ( ) and ( ).
At the medium frequencies, a fractional integration behavior of order prevails.

Actually, , ( ) ( )
| ( )| . /
{ (82)
( ( ))
At the high frequencies, a behavior identical to the one in the semi-infinite medium
√
prevails. Actually, , ( ) ( )
. /
| ( )| . /
{ (83)
( ( )) . /
Figure 18 represents Bode diagrams of the transfer of ( ) for a finite aluminum

medium of length and for four different values of the temperature sensor position,
let * + .
For a finite aluminum medium of length and for four different values of the
temperature sensor position, let * + , Figure 19 represents respectively
( ) in Black-Nichols plane.
40
50
0 10
Gain (dB)
5
-40
0
-80
-120
-7 -6 -5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
0
50
10
Phase (deg)
-45
5
0
-90
-135
-7 -6 -5 -4 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
Figure 18. Bode diagrams of ( ) for * +
40
50
10
20
5
0
0
-20
Gain (dB)
-40
-60
-80
-100
-270 -225 -180 -135 -90 -45 0 45
Phase (deg)
Figure 19. Black-Nichols plots of ( ) for * + .
7. Simulink Responses
In this last part of the chapter, the Simulink responses will be presented for both the semi-
infinite and the finite aluminum media. For both cases, the exact system is implemented on
Simulink.
So, Figure 20 shows the Simulink system whereas Figure 21 represents the temperature
variation with respect to time for different values of the sensor placement.
Time
Clock Scope4
To Workspace
Scope1
1
sqrt(u[1])
s
Transfer Fcn2 Fcn1
Dot Product
1 U1 H0(1)
Constant
Step1 Gain3 Gain1 Dot Product1
1
wx(1) f(u) exp(-(u[1])) Temp
s
Gain2 Transfer Fcn Fcn3 Fcn2 Scope To Workspace1
Figure 20. Simulink representation for the semi-infinite homogeneous plane.
0.8
0
0.7 1
5
10
0.6
50
100
0.5
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70 80 90
Time (s), T0=0°C, Flux=2000W/m2
Figure 21. Step responses of ( ) using Simulink for * + .
Note that, when comparing Figure 6 to Figure 21, a small time variation of about
appears due to the simulation time needed in Simulink that didn’t show up when
using Matlab.
As for the finite medium, Figure 22 and Figure 23 show the Simulink system and the
temperature variation with respect to time for different values of the sensor placement and for
two values of :
- for Figure 23a, is equal to

- for Figure 23b, is equal to .
Time
Clock
1 To Workspace
wLx power(u[1],-0.5) cosh(u[1]) Scope4
s
Gain3 Transfer Fcn2 Fcn4 Fcn6
1
wL power(u[1],-0.5) 1/cosh(u[1]) Dot Product2
s
Gain4 Transfer Fcn3 Fcn5 Fcn7
1 Dot Product
1
Constant wL power(u[1],-0.5) 1/tanh(u[1])
s
Gain1 Transfer Fcn1 Fcn1 Fcn2 Dot Product1
1
power(u[1],0.5) Temp
s
Transfer Fcn Fcn3 To Workspace1
Product
U1 H0(1)
Step1 Gain5 Gain6
Figure 22. Simulink representation for the finite homogeneous plane.
1 1
0 0
0.9 1 0.9 1
5 5
0.8 0.8
10 10
50 50
0.7 0.7
100 100
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a): Time (s), T0=0°C, Flux=2000W/m 2, L=100cm (b): Time (s), T0=0°C, Flux=2000W/m 2, L=15cm
Figure 23. Step responses of ( ) using Simulink for * + and for (a)
and (b) .
Conclusion
In this chapter we have studied the thermal interface behavior using the fractional order
approach. The identification of the diffusion phenomena as well as the optimization of their
process control were made easier using the fractional order approach, especially when both
operational and frequency domains are considered. In fact, based on the pseudo-state
(Sabatier, 2013) equations of the conduction phenomenon, representative models for both
semi-infinite and finite media were easily developed.
On the other hand, when it comes to time responses, it is more complicated as the
analytical solutions of Inverse Laplace transforms don’t exist for all the types of equations.
Nevertheless, approximations do exist and others are still a contemporary research subject of
our interest; they are the core of advanced studies related to fractional order applications.
References
Abi Zeid Daou R., Moreau X., Assaf R. and Christophy F. - Analysis of the Fractional Order
System in the thermal diffusive interface – Part 1: application to a semi-infinite plane
medium – 2nd International Conference on Advances in Computational Tools for
Engineering Applications (ACTEA), December 2012, Lebanon.
Agrawal O. M. P. – Application of Fractional Derivatives in Thermal Analysis of Disk
Brakes – Journal of Nonlinear Dynamics, Vol. 38, pp. 191-206, 2004, Kluwer Academic
Publishers.
Assaf R., Moreau X., Abi Zeid Daou R. and Christophy F. - Analysis of the Fractional Order
System in the thermal diffusive interface – Part 2: application to a finite plane medium –
2nd International Conference on Advances in Computational Tools for Engineering
Applications (ACTEA), December 2012, Lebanon.
Battaglia J. L., Cois O., Puissegur L. and Oustaloup A. – Solving an inverse heat conduction
problem using a non-integer identified model – International Journal of Heat and Mass
Transfer, Vol. 44, N 14, pp. 2671-2680, 2001.
Battaglia J. L. – Heat Transfer in Materials Forming Processes – ISTE Ed., London, 2008.
Benchellal A., Bachir S., Poinot T. and Trigeassou J. C. – Identification of non-integer model
of induction machines – Chapter in Fractional differentiation and its applications,
U-Books Edition, pp. 471-482, 2005.
Canat S. and Faucher J. – Modeling, identification and simulation of induction machine with
fractional derivative – Chapter in Fractional Differentiation and its Applications,
Cois O. - Systèmes linéaires non entiers et identification par modèle non entier: application en
thermique - Thèse de Doctorat de l’Université Bordeaux 1, 2002.
de Wit M. H. – Heat, air and moisture in building envelopes – Course book, Eindhoven
University of Technology, 2009.
Kuhn E., Forgez C. and Friedrich G. – Fractional and diffusive representation of a 42 V Ni-
mH battery – Chapter in Fractional differentiation and its applications, U-Books Edition,
pp. 423-434, 2005.
Kusiak A., Battaglia J. L. and Marchal R. – Heat flux estimation in CrN coated tool during
MDF machining using non integer system identification technique – Chapter in
Fractional differentiation and its applications, U-Books Edition, pp. 377-388, 2005.
Melchior P., Cugnet M., Sabatier J., Poty A. and Oustaloup A. – Flatness control of a
fractional thermal system – Chapter in Advances in Fractional Calculus: Theoretical
Developments and Applications in Physics and Engineering, Springer Ed., pp. 493-509,
2007.
Oldham K. B. and Spanier J. – The fractional calculus – Academic Press, Inc., New York,
1974.
Özişik M. N. – Heat Conduction – John Wiley & Sons, New York, 1980.
Özişik M. N. – Heat Transfer – A basic Approach – McGraw-Hill, New York, 1985.
Podlubny I. – Geometric and physical interpretation of fractional integration and fractional
differentiation – Chapter in Fractional differentiation and its applications, U-Books
Edition, pp. 3-18, 2005.
Sabatier J., Melchior P. and Oustaloup A. – A testing bench for fractional order systems
education – JESA, Vol. 42, N°6-7-8, pp. 839-861, 2008.
Sabatier J., Farges C. and Trigeassou J.-C. – Fractional systems state space description: some
wrong ideas and proposed solutions – Journal of Vibration and Control,
1077546313481839, 02-Jul-2013.
Schneider P. J. – Conduction Heat Transfer – Addison-Wesley Publishing Company Inc.,
Reading Massachusetts, 1957.
U.S. Army Corps of Engineers Technical Manual - Arctic and Subarctic Construction:
Calculation Methods for Determination of Depths of Freeze and Thaw in Soils - TM 5-
852-6/AFR 88-19, Volume 6, 1988.
Chapter 6
TEMPERATURE CONTROL OF A DIFFUSIVE MEDIUM

USING THE CRONE APPROACH
Fady Christophy1,2,, Xavier Moreau1,†, Roy Abi Zeid Daou2

and Riad Assaf1,2,‡
1
University of Bordeaux, Laboratory IMS, Talence, Bordeaux, France
2
Lebanese German University, Faculty of Public Health,
Biomedical Technologies department, Sahel Alma, Jounieh, Lebanon
Abstract
This chapter presents the design of the temperature control of a diffusive medium by using a
unique robust controller for three different materials: aluminum, copper and iron. For the
control-system design, the aluminum is selected as the material defining the nominal model.
Then, the second generation CRONE control is used because the parametric uncertainty (due
to the copper and the iron) leads to variations of open-loop gain. Finally, the responses in
frequency-domain and in time-domain illustrate the influence of the position of the sensor
versus the actuator on the stability degree robustness.
1. Introduction
The concept of fractional differentiation is an old one that was born at the end of the 17th
century when L’Hospital and Leibniz exchanged on the meaning of half order derivatives
[Oldham, 1974]. Since that date, many mathematicians worked on this concept from a
mathematical point of view. The most important contributions in this domain came in the 19th
century with Riemann and Liouville [Miller, 1993] when they gave a coherent definition of a
fractional derivation.
However, the applications on the fractional order systems started in the last quarter of the
20th century [Oustaloup, 1981]. In the last forty years, work on such systems propagated in

E-mail address: fady.chrystophy@u-bordeaux.fr.
†
E-mail address: xavier.moreau@u-bordeaux.fr.
‡
E-mail address: riad.assaf@u-bordeaux.fr.
124 Fady Christophy, Xavier Moreau, Roy Abi Zeid Daou et al.
several fields as the electrochemical processes [Kuhn, 2005] [Sabatier, 2006], dielectric
polarization [Bohannan, 2000], induction machines [Benchellal, 2005] [Canat, 2005],
viscoelastic materials [Moreau, 2002], suspension system [Abi Zeid Daou, 2010] [Abi Zeid
Daou, 2011.a], automatic control [Oustaloup, 1983] [Oustaloup, 1995] [Lanusse, 1994] [Abi
Zeid Daou, 2011.b], physiological movements as the muscle contraction [Sommacal, 2008],
the respiratory process [Ionescu, 2011], thermal diffusion [Battaglia, 2001] [Agrawal, 2004]
[Kusiak, 2005] [Melchior, 2007], and so on. In general, the development of diffusion
phenomena equations naturally leads to fractional order systems [Podlubny, 2005].
The aim of this study is to design the temperature control of a diffusive medium by using
an unique robust controller for three different materials (aluminum, copper and iron)
constituting the medium. In more details, this chapter is divided as follows. In section 2, the
material properties and the diffusive interface relations are recalled. Section 3 presents the
SISO Crone control-system design used in the frequency-domain for the temperature control
of the uncertain diffusive medium. Responses in frequency-domain and in time-domain are
given to illustrate the influence of the position of the sensor versus the actuator on the
stability degree robustness. At the end, a conclusion sums up all the results and some future
works are proposed.
2. Modelling
Consider a homogeneous one-dimensional semi-infinite medium, of conductivity  , of
diffusivity  and of initial temperature zero at every point (figure 1). It is subjected to a flux
density  t  (W/m2) on the outgoing normal surface n . This results in a temperature change,
denoted T(x,t), function of time t and of abscissa x  x  0 ;   of the measuring temperature
point inside the medium.
(t)
Semi-infinite medium
Figure 1. Illustration of the study area defining the process.
Initially, the goal is to achieve control of the temperature T(0,t) at x = 0 with a sensor
placed in this position. For this we have a heating resistor bonded to a surface S of 1 cm2 at
Temperature Control of a Diffusive Medium Using the CRONE Approach 125
the end x = 0 of a bar (high thermal conductivity glue) for generating a command u(t) of flux
f(t), or a flux density (t) = f(t)/S. The maximum flux Umax produced by the resistance is 12
W (1 A under 12 V over 1 cm2, thus 12 104 W/m2). The medium temperature is measured
with a platinum sensor of type PT100 via a voltage amplifier [Sabatier, 2008].
In a second step, the goal is always to regulate the temperature T(0,t) at x = 0 but with a
sensor placed at a position x > 0. An analysis of the influence of the position x of the sensor
on the performance of the command is given.
2.1. Fractional Model at x = 0
At x = 0, the transfer function G(0,s) of the process is given by [Abi Zeid Daou, 2012]:
T 0, s 
G 0, s    00.5 I 0.5 s  , (1)
U s 
with
 1
0  S 2   C
 p , (2)
 I 0.5 s   1
 s 0.5

0 is an uncertain parameter, 0  0 min ; 0 max  , not only because the parameters ,  et Cp
(relation (2)) are given with an uncertainty (data values at 20°C), but also because the
objective of the robust control for this study is to use the same control for different materials
constituting the semi-infinite medium (aluminum, copper, iron, …).
In this study, we chose three materials: aluminum, copper and iron. Table 1 summarizes
the materials’ characteristics necessary to calculate 0.
Table 1. Characteristics of the materials used
Materials  Cp 

 0 (rad/s)
at 20°C (kg/m3) (J.kg-1.K-1) (W.m-1.K-1) for S =10-4 m2
Aluminium 2700 889 237 1,76 10-1
Copper 8930 382 399 0,735 10-1
Iron 7860 452 81 -1
3,475 10
For these three materials, we get 0   0.735 ; 3.475  101 rad/s .

For the remainder of the control-system design, the aluminum is selected as a material
defining the nominal model, hence 0nom = 1.76 10-1 rad/s. Thus, the uncertainties of 0 result
in gain variations of G(0,j) of a factor 2.17 (between the iron and copper we have:
(3.475/0.735)0.5= 2.17). These gain variations cause gain variations  in open loop,  =
2.17, hence the use of the second generation CRONE control [Oustaloup, 1995].
Figure 2 shows the Bode plots of G(0,j) for the three materials.
20
Alu. x=0
Gain (dB)
0 Cop. x=0
Iron x=0
-20
-2 -1 0 1 2
10 10 10 10 10
Frequency (rad/s)
0
Phase (deg.)
-45
-90 -2 -1 0 1 2
10 10 10 10 10
Frequency (rad/s)
Figure 2. Bode diagram of G(0,j).
Moreover, the analytical expression of the step response T(0,t) for a flux f(t) of amplitude
U0 is given by [Schneider, 1957] [Özisik, 1980] [Özisik, 1985] [Battaglia, 2008]:
0
T 0, t  
2 U0
t  2 U0 t. (3)
S    Cp 
Figure 3 shows for the three materials the step responses at x = 0 for an input of
amplitude U0 = 1 W.
7
Alu. x=0
6 Cop. x=0
Iron x=0
5
Temperature (°C)
0
0 20 40 60 80 100
Time (s)
Figure 3. Step responses of the process at x = 0 for a flux of amplitude U0 = 1 W.

2.2. Fractional Model at x > 0
At x > 0, the transfer function G(x,s) of the process is expressed by [Assaf, 2012]:
T  x, s 
G  x, s    00.5 I 0.5 s  E x, s  , (4)
U s 
7
Alu. x=5mm
6 Cop. x=5mm
Iron x=5mm
5
Temperature (°C)
0
0 20 40 60 80 100
Time (s)
Figure 4. Step responses of the process at x = 5mm for a flux of amplitude U0 = 1 W.
with
  s 
0.5
 

E x, s   e  x 



 
, (5)
 x  2
 x
or, by introducing the nominal transfer function Gnom(0,s) established with the aluminum
at x = 0,
0.5
 s 
  
 
G x, s   Gnom0, s  e  x . (6)
Remark. The exponential function can be interpreted as a multiplicative structural

uncertainty m(x):
0.5
 s  s
0.5
    x 
 
 m x   e  x  e   . (7)
Moreover, the analytical expression of the step response T(x,t) for a flux f(t) of amplitude
U0 is given by [Battaglia, 2008]:
 1
  1  
T x, t   U 0
t 4 x t 1
0  2 e  erfc   , (8)
  x  2  t  
  x 
where erfc(.) is the complementary error function.

Figure 4 shows for the three materials the step responses at x = 5mm for an input of
amplitude U0 = 1 W.
2.3. Rational Model for Temporal Simulation  x  0
If the numerical simulation of the fractional integrator I0.5(s) and of function E x, s  cause
no problem in the frequency domain, it’s not the same for the time domain. In fact, apart from
some analytical signals such as the Dirac pulse or unit step for which analytical expressions
of time responses are well known, time simulation of the responses of the process model to
any of the other inputs is more problematic. This is the case in particular when the process
model is immersed in a control loop. It is then necessary to implement methods for
approximating the fractional model to have a rational model.
Thus, the fractional integrator I0.5(s) is first approximated by a fractional integrator
bounded in frequency between ωa and ωb that can be written as a I0.b5 s  where:
 0.5
 I s   s 0.5
1
0.5
 1 K0  1  s / a 
 0.5     , (9)
 1  s / a  s 0.5  1  s / b 
  I 0.5 s   K 0   s
  1  s / b 
a b
 s
with a < b and K0 a constant computed for the two integrators to have the same gain at
midband m  a b :
 1  m / b 2 
0.25
K0  m0.5   , (10)
 1   /  2 
 m a 
or by introducing the constant A = b/a > 1,
m0.5
K0  . (11)
A0.25
Then the fractional integrator bounded in frequency is approximated by a rational

integrator frequency bounded [Oustaloup, 1995]:
0.5
 a I b s   lim I N0.5 s  , (12)
N 
with
 s 
1 ' 
N
i
  ,
K0
I N0.5 ( s )  (13)
s  s 
i 1  1  
 i
 
where
1/ N
 
    b  ,      0.5 ,     0.5 ,
 a 
1 i
1'   a , N  b ,    1, , (14)
 i'
i' 1 i' 1 i 1
   1 et     1 .
i i' i
Finally, concerning the exponential function E x, s  , we use the limited Taylor expansion
z
of e when z tends to zero:

z2 z3 zk zk
lim e z  1  z 
z 0

2! 3!
 ... 
k!
 ... 
k 0
k!
.  (15)
By taking z = (s/x)0.5, we obtain:
1 1 1
0.5
 0.5 1.5 0.5 K
 0.5k
. (16)
 s 
 s  1 s  1  s 
K
1 s 

  s
 1         ...     
e x 2!x 3!  x  !  x 

 x  K!   x  k 0
k
Remark. z tends to zero is equivalent to tendering s to zero, and therefore  to zero and t
to infinity (time-frequency duality), or, considering the expression (5) of x, x tends to zero.
Thus, this approximation is more accurate than the truncation order K is high, when x is
small and it is with long time. In fact, this approximation is legitimate when  << x or
equivalently, when t >> x=1/x.
In a first step, K is limited to 1, hence:
0.5
 x 
 
T x, s   s 
E x, s  
1
  (14)
T 0, s   s 
0.5
 
0.5 ,
1    1  x 
 x   s 
the expression of order 0.5 could be interpreted as a closed loop system (figure 5), the open
loop transfer function is equal to the product of a constant  x0.5 and a fractional integrator
I0.5(s), the latter having already been subjected to the approximation described above.
E  x, s 
T 0, s  + T  x, s 
x0.5 I 0.5
s 
-
Figure 5. Block diagram associated with the approximation of the function E(x,s).
In summary, obtaining the rational model for temporal simulation,  x  0 , is due to the
rational integrator I N0.5 s  whose center frequency m is chosen equal to the crossover
frequency, unom, in nominal open loop.
Figure 6 shows the block diagram of the rational model used in Simulink for time
response simulation.
T 0,0 T x,0
0.5 
ut   f t  1  t  1 T 0, t  T 0, t   T  x, t 
  Cp
I0.5
0 t . + I0.5
0 t . +
S + - x
Figure 1. Block diagram of the rational model used in Simulink for time response simulation.
Remark. In a second step, an optimal approach to managing the dilemma between the
fidelity of the approximation which may be improved by increasing the order K of the
truncation of the series (16) (by increasing the complexity), and the simulation time which
unfortunately also increases with the degree of complexity is proposed. A fractional
representation of order 0.5 is studied in particular for the formulation of the optimal
approach, the value of the order K directly fixing the number of fractional integrators.
Similarly, the optimum choice of the number of decades and cells in the synthesis of rational
integrator will be a specific study.
Finally, the approximation errors of the rational model may be considered as
uncertainties and used for the control-system design.
3. Temperature Control
3.1. Scheme for the Control-System Design
Figure 7 shows the scheme used for the control-system design.

Input disturbance Sensor noise

Du(s) Dm(s)
Error signal
Yref(s) + (s) U(s) Y(s)
Reference C(s) + G(s) +
signal - Measured
Controller Plant model output
Figure 7. Scheme used for the control-system design.
The equations associated with the scheme of figure 7 are:
for the output signal:
Y s   S s  Dm s   S s  Gs  Du s   T s  Yref s  ; (18)
for the error signal:
 s    S s  Dm s   S s  Gs  Du s   S s  Yref s  ; (19)
for the control signal:
U s    S s  C s  Dm s   T s  Du s   S s  C s  Yref s  , (20)
with

 S s   1   s  : Sensitivity function
1

T s   1  S s  : Complementary sensitivity function . (21)
  s   C s  G s  : Open loop transfer function


3.2. Summary Data
The required data for the control-system design are divided into two groups:
 Signals models (often difficult to obtain, especially regarding the measurement

noise) and the uncertain especially linear model of the process which in the case of a
frequency approach is in the form of a transfer function G(s) (Figure 7), the areas of
uncertainty parameters i are specified, for example, in the form of bounded

intervals, i  i ; i ; 
 Translating specifications:
o Degree of stability, through the stability margins, M, MG, MM;

o Speed, through the unit gain frequency in open loop, u  u min ; u max  ;
o Accuracy in the steady state, through the relative error, (%);
o Saturation, through the maximum value of the command, Umax.
Remark. It should be noted that the translation of specifications requires preliminary

work, especially regarding the choice of u.
In the absence of additional information, particularly regarding the measurement noise, it
is difficult (or impossible) to determine precisely unom.
However, a simplified approach leading to a first iteration can be made by taking unom
= 3 0nom, initial choice based on experience…
Another approach is to set the contribution cm of measurement noise relative to the
maximum value Umax of control signal (for instance cm = 10%). In the absence of precise
model of the measurement noise, knowing that its frequency spectrum is more concentrated
towards the high frequencies, we place ourselves in stationary harmonic, we estimate the
amplitude Dm of measurement noise over the full scale range Em of the sensor (for instance
Dm = 1% Em) and from equation (3) we impose:
lim U  j   lim S  j  C  j  Dm  cm U max , (22)

   
knowing that lim S  j   1 ,

 
lim U  j   lim C  j  Dm  cm U max . (23)

   
Thus, a strain is obtained regarding the controller gain at high frequencies:

C  lim C  j  
cm U max
, (24)
  Dm
constraint leads to limit u.
3.3. Loop Shaping
There are three generations of the CRONE controller [Oustaloup, 1995]. When
parametric uncertainties result in gain variations  in open loop, the second generation is
used. Hence, the expression of the open loop transfer function (s) is given by:
nl n
 1  s / l   1  s / h 
 ( s )  0     1  s / h  nh , (25)
 s / l   1  s / l 
where ωl and ωh represent the transitional low and high frequencies, n the non-integer order
between 1 and 2 near the frequency ωu, nl and nh are the orders of the asymptotic behavior at
low and high frequencies (figure 8) and β0 a constant which ensures unity gain at frequency
ωu. β0 is given by:

0  u /l nl 1  u /l 2 
n  nl  / 2
1   /  
u h
2 nh  n  / 2
. (26)
The fractional asymptotic behavior (figure 8) is set to a frequency interval [ωA, ωB]
around the nominal frequency ωunom. To meet the robustness of the degree of stability, it is
necessary to impose:
  u min
 u  u min ; u max  ,  A  u  B   A . (27)
B  u max
Open-loop gain
Log10(h)
0 dB Log10(B)
Log10(unom) Log10()
Log10(A)
Log10(l)
Open-loop phase Fractional asymptotic behavior
10 10
0°
Log10()
-nl 90 °
-n 90 °
-nh 90 °
Figure 8. Asymptotic Bode diagrams of (j).
In addition, to obtain a fractional asymptotic behavior on the interval [ωA, ωB] (Figure 8),
a practical rule leads to choose [Oustaloup, 1995]
l   A / 10
 . (28)
h  10 B
Taking l and h geometrically distributed around unom and introducing a = B/A:
 l h  unom

 B , (29)
a  
 A
l and h are given by:

 l h  unom

 h  

l  unom / 10 a
.
 (30)
   100 a h  unom 10 a
 l
Log10
20 Log10 0 dB
Log10A Log10unom Log10B
Log10B - Log10A = Log10(B/A)

= Log10a
Figure 9. Illustration of computing a.
As for the calculation of the ratio a, it is derived from the slope of –n20dB/dec near unom
and the gain variation  in open loop due to parametric uncertainties (figure 9). Thus, we
find out:
a   1 / n . (31)
Finally, given that
 ( s)  Ccrones  Gs  , (32)
the fractional form CF(s) of the transfer function of the CRONE regulator Ccrone(s) is deducted
from the transfer function Gnom(s) of the nominal process:
CF s    ( s) / Gnoms  . (33)
Finally, one last step is necessary to determine the rational form CR(s) of the fractional
transfer function CF(s) which is obtained either analytically using a recursive distribution of
poles and zeros [Oustaloup, 1995], either numerically from the frequency response of CF(j)
and a specific module of the CRONE Toolbox [Oustaloup, 2005].
Remark. Be careful, if the process is non minimum phase (positive real part zeros) and if
it has a delay and / or very little damped modes, then there are precautions to take
[Oustaloup, 2005].
3.4. Specifications and Results
Finally, the translation of specifications led to choose:
o for the degree of stability, phase margin M  45;

o for the speed, crossover frequency, unom, in nominal open loop the greatest possible;
o for accuracy in the steady state, a static error equal to zero;
o to saturation, a maximum value of the control signal, Umax = 12 W.
Given the summary data, we choose:
 nl = 1.5, to ensure zero steady-state error;

 n = (180° - M)/90° = 1.5, to ensure a phase margin M = 45;
 nh = 1.5, to limit the input sensitivity;
 unom = 30nom = 0.53 rad/s (simplified approach).
From these data and taking into consideration the relation (31), we deduce:
 A  u min  0.41 rad/s   0.041 rad/s

a  2.171 / 1.5  1.68     l . (34)
B  u max  0.684 rad/s h  6.84 rad/s
The expression of the transfer function is given by:
1.5 1.5
 1  s / l   1  s / h 
 ( s )  0     1  s / h 1.5 , (35)
 s / l   1  s / l 
with
0  46.6 . (36)
From the relations (32) and (33), we deduce the expression of the fractional form CF(s) of
the CRONE regulator:
1.5 1.5 0.5

 1  s / l   1  s / h   
CF ( s )  0     1  s / h 1.5  s
 , (37)
 s / l   1  s / l  
 1nom 
after simplification,
C0
CF ( s )  , (38)
s
with
 1.5 
C0  0  0l.5   C0  9.145 101 W / K . (39)
 1nom 
Remark. The nature of the plant (integrator of order 0.5) and the choice of orders
asymptotic behavior at low (nl = 1.5), at average (n = 1.5) and at high (nh = 1.5) frequencies
lead to a particular case where the expression CF(s) is already in a very simple rational form
(an integrator of order 1). In fact, for this particular case, the transfer function (s) in open
loop is an integrator of order 1.5:
1.5
 
 ( s)   u  . (40)
 s 
Sensitivity function S(s) and complementary sensitivity function T(s) have the following
expressions:
S ( s) 
1

s / u 1.5 (41)
1   ( s) 1  s / u 1.5
and
 ( s) 1
T ( s)   . (42)
1   ( s ) 1  s / u 1.5
Open-loop Gain (dB)
40 Alu.
20 Cop.
0 Iron
-20
-40 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Open-loop Phase (deg.)
0
-45
-90
-135
-180 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Figure 10. Bode diagrams of (j).

3.4.1. Frequency Responses at x = 0
Figures 10 and 11 show Bode diagrams and Black-Nichols diagrams of the transfer
function in open loop (s) for the nominal parametric state, and for the extreme values of
gain. The robustness of the degree of stability is ensured, due to the insensitivity of the phase
margin Mto parametric uncertainties.
40
0 dB
30
20 1 dB
Open-loop Gain (dB)
10 3 dB
6 dB
0
-10
-20
-30
-40
-270 -225 -180 -135 -90 -45 0
Figure 11. Black-Nichols diagram of (j).
-5
-10
Gain (dB)
-15
-20
-25 Alu.
-30 Cop.
Iron
-35
-40 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Figure 12. Gain diagram of the complementary sensitivity function T(s).

Figures 12 and 13 show diagrams of the complementary sensitivity function T(s) (figure
12) and the sensibility function S(s) (figure 13) for the nominal parametric state, and for the
extreme values of the uncertainties. The robustness of the degree of stability appears with
insensitivity of the resonance factors T(j) et S(j) to parametric uncertainties.
Figure 14 shows diagrams of the sensitivity function input R(s) = S(s)C(s). We can notice
that the gains R(j) are maximum at medium frequencies.
-5
-10
Gain (dB)
-15
-20
-25
Alu.
-30 Cop.
-35 Iron
-40 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Figure 13. Gain diagram of the sensibility function S(s).
10
Alu.
Cop.
5
Iron
0
Gain (dB)
-5
-10
-15
-20 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Figure 14. Gain diagram of the sensitivity function of the input R(s)= S(s)C(s).
3.4.2. Time Responses at x = 0
The rational model of the process used for the time simulation is derived from the
rational integrator I N0.5 s  (relation (13)) obtained by taking a median frequency m = unom =
0.53 rad/s, a constant A = b/a = 108 (an approximation of 8 decades) and a number of cells
N = 20.
Figure 15 shows, at x = 0 and for aluminum, the step responses of the temperature to a
flux of amplitude 1 W obtained from the exact response (relation (3)) and the response of the
approximated model.
3.5
2.5
Temperature (°C)
1.5
0.5 Alu. x=0 exact.

Alu. x=0 approxim.
0
0 10 20 30 40 50
Time (s)
Figure 15. Step responses of the temperature at a flux of amplitude 1 W, at x = 0, for aluminum,
obtained from the exact response (relation (3)) and the response of the approximated model.
-3
x 10
0.5
0
Temperature difference (°C)
-0.5
-1
-1.5
-2
-2.5
-3
0 10 20 30 40 50
Time (s)
Figure 16. Temperature difference, at x = 0 and for aluminum, between the exact step response (relation
(3)) and the approximated model.
Figure 16 shows, always at x = 0 and for aluminum, the temperature difference between
the exact step response and the approximated model, to assess the quality of the
approximation since the maximum error in 50 seconds is less than 3/1000 of °C.
Figure 17 shows the step responses of the output at a step input of amplitude 1°C for the
nominal parametric state, and for the extreme values of the uncertainties. The robustness of
the degree of stability clearly appears with the insensitivity of the first overshoot and the
damping factor.
1.4
Alu.
1.2 Cop.
Iron
Temperature T(0,t) (°C)
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50
Time (s)
Figure 17. Step responses of the output to an input step of 1°C: for aluminum (________), copper (________)
and iron (________).
1
Alu. x=0
0.8 Cop. x=0
Iron x=0
0.6
Error (°C)
0.4
0.2
-0.2
-0.4
0 10 20 30 40 50
Time (s)
Figure 18. Step responses of the error to a step input of 1°C: for aluminum (________), copper (________) and
iron (________).
Figure 18 shows the step responses of the error for the same simulation conditions.
According to specifications, the static error is zero.
Finally, figure 19 shows the corresponding step responses of the control signal where the
maximum value Umax = 12 W is never reached.
2
Alu. x=0
1.8
Cop. x=0
1.6 Iron x=0
1.4
Control signal (W)
1.2
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50
Time (s)
Figure 19. Step responses of the control signal to a step input of 1°C: for aluminum (________),
copper (________) and iron (________).
3.4.3. Frequency Responses at x > 0
The objective is always to control the temperature T(0,t) at x = 0, but by analyzing the
influence of the position x of the temperature sensor when it can’t be positioned at x = 0.
Thus, the transfer function of open loop (x,s) is then given by:
0.5
 s 
  
 x, s   C s  G 0, s 

e  x 
, (43)
by introducing the nominal transfer function of the open loop nom(0,s),
0.5
 s 
  
 
 x, s    nom0, s  e  x , (44)
whose frequency response (x,j) is expressed by:
0.5
  
  j 
 x, j    nom0, j 

e  x . (45)
Knowing that
   
0.5
  
0.5
    x
 
0.5
   
m  x ,    e
  j     
e  
2 2
 
x
e  x  mx,   e  j  x ,  avec  0.5 0.5 , (46)

 x,          x  
  2 
  x  2 
the gain and the argument of the open loop are given by:
  x, j    nom0, j  mx,  


 . (47)
arg  x, j   arg  0, j    x,  
 nom
Illustratively, the controller synthesized for x = 0 is applied to the process at x = 0.5 cm.
This value is selected to 0.5 cm to compare the results with those shown in [Sabatier, 2008].
Figures 20 and 21 show Bode diagrams of Black-Nichols diagram for the transfer
function of open loop (x,s) for the nominal case (aluminum at x = 0) and at x = 0.5 cm with
aluminum, copper and iron.
Open-loop Gain (dB)
50 Alu. x
Cop. x
25
Iron x
0 Alu. 0
-25
-50 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
0
-45
-90
-135
-180
-225 -2 -1 0 1
10 10 10 10
Frequency (rad/s)
Figure 20. Bode diagrams of (x,j) at x = 0.5 cm for aluminum (_____), copper (_____) and iron (_____),
and for the nominal case (_____).
3.4.4. Time Response at x > 0
Figure 22 shows, at x = 0.5 cm and for aluminum, the step responses of the temperature
to a flux of amplitude 1 W obtained from the exact response (relation (8)) and the response of
the approximated model.
Figure 23 shows, still at x = 0.5 cm and for aluminum, the temperature difference
between the exact step response and the approximated model, to assess the quality of the
approximation since the maximum error in 50 seconds is less than 33/1000 of °C.
40
0 dB
30
20 1 dB
Open-loop Gain (dB)
10 3 dB
6 dB
0
-10
-20
-30
-40
-270 -225 -180 -135 -90 -45 0
Figure 21. Black-Nichols diagrams of (x,j) at x = 0.5 cm for aluminum (_____), copper (_____) and iron
(_____), and for the nominal case (_____).
3.5
2.5
Temperature (°C)
1.5
Alu. x=5mm exact.

0.5
Alu. x=5mm approxim.
0
0 10 20 30 40 50
Time (s)
Figure 22. Step responses of the temperature at a flux with an amplitude of 1 W, at x = 0.5 cm and for
aluminum, obtained from the exact response (relation (8)) and the response of the approximated model.
-0.005
Temperature difference (°C)

-0.01
-0.015
-0.02
-0.025
-0.03
-0.035
0 10 20 30 40 50
Time (s)
Figure 23. Temperature difference, at x = 0.5 cm and for aluminum, between the exact step response
(relation (8)) and the approximated model.
Alu.
2 Cop.
Iron
1.5
0.5
0
0 10 20 30 40 50
Time (s)
(a)
Alu.
2 Cop.
Iron
Temperature T(5mm,t) (°C)
1.5
0.5
0
0 10 20 30 40 50
Time (s)
(b)
Figure 24. Simulated step responses of the temperature T(0,t) at x = 0 (a) and T(5mm,t) at x = 5mm (b)
for a step input of 1°C for aluminum, copper and iron, in the case of a position sensor x capt = 0 mm.
To facilitate comparison, figure 24 recalls for a position sensor xcapt = 0 mm the simulated
step responses of the temperature to a step input of 1°C for aluminum, copper and iron, not
only at x = 0 mm (figure 24.a), but also at x = 5 mm (figure 24.b).
Figure 25 shows for a position sensor xcapt = 5 mm the simulated step responses of the
temperature at a step input of 1 ° C for aluminum, copper and iron, at x = 0 mm (figure 25.a)
and at x = 5 mm (figure 25.b).
Thus, with a position sensor xcapt = 5 mm, not only the specifications regarding degree of
stability and precision for steady-state temperature T(0,t) are not complied with whatever
material, but also the robustness of the degree of stability achieved with sensor position xcapt =
0 mm disappears. This result was predictable by viewing figure 21.
Alu.
2 Cop.
Iron
1.5
0.5
0
0 10 20 30 40 50
Time (s)
(a)
Alu.
2 Cop.
Iron
Temperature T(5mm,t) (°C)
1.5
0.5
0
0 10 20 30 40 50
Time (s)
(b)
Figure 25. Simulated step responses of the temperature T(0,t) at x = 0 (a) and T(5mm,t) at x = 5mm (b)
for a step input of 1°C for aluminum, copper and iron, in the case of a position sensor x capt = 5 mm.
4. Conclusion
In this chapter we have studied the design of the temperature control of a diffusive
medium by using an unique robust controller for three different materials, namely: aluminum,
copper and iron. For the control-system design, the aluminum has been selected as the
material defining the nominal model. Then, the second generation CRONE control has been
chosen because the parametric uncertainty (due to the copper and the iron) leads to variations
of open-loop gain. Finally, the responses in frequency-domain and in time-domain have
illustrated the influence of the position of the sensor versus the actuator on the stability degree
robustness.
As a future work, three tracks will be studied during the design of robust control:
 the first will be to take into account the uncertainties associated with the sensor
position xcapt;
 the second will be to estimate the temperature T(0,t) from the measurement of the
temperature T(xcapt,t);
 the third will be to combine the first two.
References
[Abi Zeid Daou, 2010] Abi Zeid Daou R., Francis C. and Moreau X. – Study of the inertial
effect and the nonlinearities of the CRONE suspension based on the hydropneumatic
technology – Nonlinear Dynamics, an International Journal of Nonlinear Dynamics and
Chaos in Engineering Systems, Vol.63, N°1-2, pp.1-17, 2010.
[Abi Zeid Daou, 2011.a] Abi Zeid Daou R., Moreau X. and Francis C. – Effect of
hydropneumatic components nonlinearities on the CRONE suspension – IEEE
Transactions on Vehicular Technology, Vol. 61, N°2, pp. 466-474, 2011.
[Abi Zeid Daou, 2011.b] Abi Zeid Daou R., Moreau X. and Francis C. – Control of hydro-
electromechanical system using the generalized PID and the CRONE controllers - 18th
World Congress of the International Federation of Automatic Control (IFAC), Milano,
Italy, August 18 – September 2, 2011.
[Abi Zeid Daou, 2012] Abi Zeid Daou R., Moreau X., Assaf R. and Christophy F. - Analysis
of the Fractional Order System in the thermal diffusive interface – Part 1: application to a
semi-infinite plane medium – 2nd International Conference on Advances in
Computational Tools for Engineering Applications (ACTEA), December 2012, Lebanon.
[Agrawal, 2004] Agrawal O.M.P. – Application of Fractional Derivatives in Thermal
Analysis of Disk Brakes – Journal of Nonlinear Dynamics, Vol. 38, pp. 191-206, 2004,
Kluwer Academic Publishers.
[Assaf, 2012] Assaf R., Moreau X., Abi Zeid Daou R. and Christophy F. - Analysis of the
Fractional Order System in the thermal diffusive interface – Part 2: application to a finite
plane medium – 2nd International Conference on Advances in Computational Tools for
Engineering Applications (ACTEA), December 2012, Lebanon.
[Battaglia, 2001] Battaglia J.L., Cois O., Puissegur L. and Oustaloup A. – Solving an inverse
heat conduction problem using a non-integer identified model – International Journal of
Heat and Mass Transfer, Vol. 44, N°14, pp. 2671-2680, 2001.
[Battaglia, 2008] Battaglia J.L. – Heat Transfer in Materials Forming Processes – ISTE Ed.,
London, 2008.
[Benchellal, 2005] Benchellal A., Bachir S., Poinot T. and Trigeassou J.C. – Identification of
non-integer model of induction machines – Chapter in Fractional differentiation and its
applications, U-Books Edition, pp. 471-482, 2005.
[Bohannan, 2000] Bohannan G. W. - Application of Fractional Calculus to Polarization
Dynamics in Solid Dielectric Materials - Ph. D. Thesis, Montana State University, 2000.
[Canat, 2005] Canat S. and Faucher J. – Modeling, identification and simulation of induction
machine with fractional derivative – Chapter in Fractional Differentiation and its
Applications, U-Books Edition, pp. 459-470, 2005.
[Ionescu, 2011] Ionescu C., Desager K., De Keyser R. - Fractional order model parameters
for the respiratory input impedance in healthy and in asthmatic children - Computer
Methods and Programs in Biomedicine, Vol. 101, Issue 3, pp. 315-323, March 2011.
[Kuhn, 2005] Kuhn E., Forgez C. and Friedrich G. – Fractional and diffusive representation
of a 42 V Ni-mH battery – Chapter in Fractional differentiation and its applications,
[Kusiak, 2005] Kusiak A., Battaglia J.L. and Marchal R. – Heat flux estimation in CrN coated
tool during MDF machining using non integer system identification technique – Chapter
in Fractional differentiation and its applications, U-Books Edition, pp. 377-388, 2005.
[Lanusse, 1994] P. Lanusse, De la commande CRONE de première génération à la commande
CRONE de troisième génération, PhD Thesis, Bordeaux I University, France, 1994.
[Melchior, 2007] Melchior P., Cugnet M., Sabatier J., Poty A. and Oustaloup A. – Flatness
control of a fractional thermal system – Chapter in Advances in Fractional Calculus:
Theoretical Developments and Applications in Physics and Engineering, Springer Ed.,
pp. 493-509, 2007.
[Miller, 1993] Miller K.S. and Ross B. – An Introduction to the Fractional Calculus and
Fractional Differential Equations – Wiley, New York, 1993.
[Moreau, 2002] Moreau X., Ramus-Serment C. and Oustaloup A. – Fractional Differentiation
in Passive Vibration Control – Nonlinear Dynamics, An International Journal of
Nonlinear Dynamics and Chaos in Engineering Systems, ISSN 0924-090X, Vol. 29, pp.
343-362, 2002.
[Oldham, 1974] Oldham K.B. and Spanier J. – The Fractional Calculus – Academic Press,
New York, 1974.
[Oustaloup, 1981] Oustaloup A. - Fractional order sinusoïdal oscillators: optimization and
their use in highly linear FM modulation - IEEE Transactions on Circuits and Systems,
vol. cas 28, n°10, pp. 1007-1009, 1981.
[Oustaloup, 1983] Oustaloup A., Systèmes asservis linéaires d’ordre fractionnaire: théorie
et pratique. Masson, Paris, 1983.
[Oustaloup, 1995] Oustaloup A., La dérivation non entière: théorie, synthèse et
applications. Hermès, Paris, 1995.
[Oustaloup, 2005] Oustaloup A., Melchior P., Lanusse P., User’s Guide, CRONE Control
Design Module, CRONE Toolbox for MatLab, Version 3.5, May 27, 2005.
[Özişik, 1980] Özişik M.N. – Heat Conduction – John Wiley & Sons, New York, 1980.
[Özişik, 1985] Özişik M.N. – Heat Transfer – A basic Approach – McGraw-Hill, New York,
1985.
[Podlubny, 2005] Podlubny I. – Geometric and physical interpretation of fractional
integration and fractional differentiation – Chapter in Fractional differentiation and its
applications, U-Books Edition, pp. 3-18, 2005.
[Sabatier, 2006] Sabatier J., Aoun M., Oustaloup A., Grégoire G., Ragotand F., Roy P. -
Fractional system identification for lead acid battery sate charge estimation - Signal
Processing, 86 (10), 2645–2657, 2006.
[Sabatier, 2008] Sabatier J., Melchior P. and Oustaloup A. – A testing bench for fractional
order systems education – JESA, Vol. 42, N°6-7-8, pp. 839-861, 2008.
[Schneider, 1957] Schneider P.J. – Conduction Heat Transfer – Addison-Wesley Publishing
Company Inc., Reading Massachusetts, 1957.
[Sommacal, 2008] Sommacal A., Melchior P., Cabelguen J-M., Oustaloup A., Ijspeert A-J. -
Fractional Multi-Models of the Gastrocnemius Frog Muscle - Journal of Vibration and
Control. Sage Publishing, Vol. 14, No. 9-10, pp. 1415-1430, 2008.
Chapter 7
ADAPTIVE SECOND-ORDER FRACTIONAL

SLIDING MODE CONTROL WITH APPLICATION
TO WATER TANKS LEVEL CONTROL
Danial Senejohnny1,, Mohammadreza Faieghi2,†

and Hadi Delavari 3,‡
1
School of Electrical Engineering,
Sharif University of Technology, Tehran, Iran
2
School of Electrical Engineering, Iran University of Science
and Technology, Tehran, Iran
3
Faculty of Electrical Engineering, Hamedan University
of Technology, Hamedan, Iran
Abstract
Combining the fractional calculus with second-order sliding mode control, a novel type of
control strategy called second order fractional sliding mode control is introduced for a class of
nonlinear dynamical systems subject to uncertainty. A fractional-order switching manifold is
proposed and the corresponding control law is formulated based on the Lyapunov stability
theory to guarantee the sliding condition. A novel adaptation algorithm is derived to ensure
perfect tracking, diminish chattering effect and steady state error by estimating switching
controller parameters. Finally, numerical simulation results utilizing the dynamic model of
interconnected twin tank system are presented to illustrate the effectiveness of the proposed
method.

E-mail address: d.senejohnny@gmail.com (Corresponding author).
†
E-mail address: mfaieghi@gmail.com.
‡
E-mail address: hdelavary@gmail.com.
150 Danial Senejohnny, Mohammadreza Faieghi and Hadi Delavari
1. Introduction
Among existing control methodologies, the sliding mode control method, first proposed
in the early fifties, is one of the control design methods to dominate the uncertainties and
disturbances (i.e., the so-called 'matched' disturbances) acting on the systems. The method has
gained significant research attention since early sixties in the former USSR and in the modern
world since the late seventies and has been widely applied in a variety of applications [1, 2].
In first order sliding mode controller design, the sliding surface is selected such that it has
relative degree one with respect to the control input. That means the control input acts on the
first derivative of the sliding surface. Higher-order sliding mode is the generalization of the
first-order sliding mode and the control input is performed so that it acts on higher derivatives
of the sliding surface. The chattering is due to the inclusion of the sign function in the
switching term and it can cause the control input to start oscillating around the zero sliding
surface, resulting in unwanted wear and tear of the actuators. In general two approaches have
been proposed in the literature to solve the problem. The first is to smoothen the switching
term as the sliding surface gets closer to zero (soft switching) by using the continuous
approximations of the discontinuous sign function, and the second is to generate "higher-
order sliding modes'', first introduced by Levant in 1987 [3, 4].
The second-order and higher-order sliding mode approaches have been actively
developed over the last two decades for chattering attenuation and robust control of uncertain
systems in variety of applications with relative degree two and higher respectively [3-12]. The
second-order sliding mode control compared to first-order SMC has the advantage of
providing a smooth control and better performance in the control implementation yielding
less chattering and better convergence accuracy while preserving the robustness properties.
In recent years, numerous studies and applications of fractional-order systems in many
areas of science and engineering have presented [13-16]. Fractional calculus, as old as the
ordinary differential calculus, goes back to times when Leibniz and Newton invented
differential calculus. Emergence of effective methods in differentiation and integration of
non-integer order equations make fractional-order systems and controllers more and more
attractive to the control engineers.
In most cases, researchers consider the fractional order controllers applied to the integer
or fractional order plants to enhance the system control performance [17-25]. In [17] a new
tuning method for fractional order proportional and derivative controller (FO-PD) is proposed for
a class of typical second-order plants, existence of fractional adaptive controller based on high
gain output feedback for linear, time-invariant, minimum phase, and single input single output
systems of relative degree one is investigated in [18], several alternative methods for the control
of power electronic buck converters applying fractional order control (FOC) [19], a fractional
order disturbance observer for robust vibration suppression in [20], fractional order reference
models in model-reference adaptive control investigated in [21], and various kind of
fractional sliding mode control strategies with different fractional sliding surfaces in [22- 25].
In this paper, the objective is to construct a second-order sliding mode control based on a
fractional PIDDα sliding surface with independent gain coefficients. Although the theme of 2-
SMC is inherently proposed for high frequency chattering suppression, but an adaptive
algorithm for of parameter estimation in switching surface is proposed to further suppress this
phenomenon. As a way of illustrating the control scheme, simulations are carried out on
interconnected twin tank system.
Adaptive Second-Order Fractional Sliding Mode Control … 151
The rest of this paper is organized as follows, basic mathematical definitions of fractional
calculus and problem statement are presented in Section 2. Second-order fractional sliding
mode control strategy with fractional PIDDα sliding surface, widely investigated in section 3.
Adaptive algorithm for estimating switching surface parameters is given in Section 4. In
Section 5, simulation results are given to support the theoretical analysis of the proposed
second order fractional sliding mode control and validate its usefulness. A summary of the
present research is given in Section 6.
2. Preliminaries
2.1. Basic Definition of Fractional-Order Calculus
Fractional-order integration and differentiation are the generalization of the integer-order

ones. Efforts to extend the specific definitions of the traditional integer-order to the more
general arbitrary order context led to different definitions for fractional derivatives [13-16].
One of the most commonly used definitions is Caputo definition. In this paper, authors have
used Caputo fractional operators
c
t0 Dt as a main tool. The Caputo fractional derivative of
order  of a continuous function f (t ) is defined as follows
f ( )
t
1
 (  1)  (t   )
c
t0 Dt f (t )  d , 0   1 (1)
t0
where t0 and t are the limits of operation, and  is the Gamma function.
The initial conditions for the fractional order differential equations with the Caputo’s
derivatives are in the same form as for the integer-order differential equations. The formula for
the Laplace transform of the Caputo’s fractional derivative for t0  0 has the form [13-16]:
e st  0c Dt f (t )dt  s F (s)  s 1 f (0) 0    1


0
(2)
where s denotes the Laplace operator and F ( s) is the Laplace transform of f (t ) . Laplace
transform of the Caputo derivative allows utilization of initial values of classical integer order
derivatives. For zero initial conditions, Laplace transform of any fractional derivatives
reduces to:
L  0 Dt f (t )  s F (s) (3)
2.2. Problem Statement
Suppose a generalized nonlinear dynamic model can be described by a coupled second-

order nonlinear system of the form
x2 k 1 (t )  x2 k (t )
(4)
x2 k (t )  f k ( x)  bk uk (t )  f k ( x), k  1,..., n
where f k ( x) is the lumped model uncertainty, and x(t ) is state vector
x(t )   x1 (t ), x2 (t ), x3 (t ),..., x2n (t )  R 2n

T
(5)
also uk (t )  R is a control action, bk ( x) and f k ( x, t ) , are the control gains and the nonlinear
dynamics of the system, respectively.
The desired state variables are defined as:
T
xd (t )   x1d (t ), x2d (t ), x3d (t ),..., x2dn (t )   R 2n (6)
The tracking error can be defined as:
e2 k 1 (t )  x2dk 1 (t )  x2 k 1 (t )
(7)
e2 k (t )  x2dk (t )  x2 k (t )
Replacing (4) in (7) yields:
e2 k 1 (t )  e2 k (t )  x2dk 1 (t )  x2dk (t )
(8)
e2 k (t )  x2dk (t )  f k ( x)  bk uk (t )   k (t )  f k ( x),
The control objective is to steer the system’s states x(t ) to the desired trajectory x d (t ) . In
next section the theory of second order sliding mode control with fractional PIDDα sliding
surface is studied.
3. Second Order Sliding Mode Control Strategy

In first order sliding mode controller design a desirable sliding surface, namely S , is
selected and the feedback control law, acting discontinuously on S , aims to fulfill the
constraint S  0 in finite time.
Provided that S , S , , S ( r 1) are continuous functions and the rth order sliding set
S S   S ( r 1)  0 , the rth order sliding mode control objective is to steer to zero not only
the sliding surface S in finite time, but also its (r-1) first successive time derivatives by defining
a suitable discontinuous control function acting on the rth time derivative of S [9, 12].
In the particular case of the second-order sliding mode control, the control acting on the
second derivative of the sliding variable, namely S , aims to steer to zero not only the sliding
surface S , but also its first-order time derivative as S  S  0 [11].
The idea of the proposed second-order sliding mode, designed for the special class of
nonlinear systems (4), is based on the controller proposed in [10]. The selected sliding surface
in this research is a fractional PIDDα function of error (section 3.1) and the switching
controller parameters are estimated according to an adaptation law (section 4). The block
diagram of the proposed control scheme is depicted in Figure 1
Equivalent
Controller
ueq
Fractional +
xd + e s Switching usw + Nonlinear x
Sliding
_ Controller System
Surface
Adaptation
Law
Figure 1. Block diagram of second-order sliding mode controller with fractional sliding surface.
3.1. Second Order Fractional Slidng Surface Synthesis
The problem of interest in the present case is to generate a second-order sliding mode
control on a chosen sliding surface S (t ) . In the literature, some different sliding functions
were used in the derivation of sliding mode controllers such as, Terminal and fractional
sliding surface [24, 26], fractional integral sliding surface [27], PD and fractional PDα sliding
surface [25, 28]. In order to utilize the robustness properties of fractional order controller and
c 
reduce the sliding surface settling time, we used the term D e (t ) beside the term PID where
unanimously construct a robust fractional sliding surface PIDDα [22, 29].
The PIDDα fractional sliding surface with constant coefficients can be introduced as:
t
Sk (t )  k Sk (t )  kkP e2k 1 (t )  kkD e2k 1 (t )  kkFD Dk e2k 1 (t )  kkI  e2k 1 ( )d (9)
0
where kkP , kkD , kkI and kkFD are the independent positive constants denoting proportional
derivative and fractional derivative gains, respectively,  k is the order of fractional
derivative action,  k  0,1 ,  k is also a positive constant, that determines Sk (t ) rate of
decay. Substituting (8) in (9) one obtains
Sk (t )  k Sk (t )  kkP e2 k 1 (t )  kkD  e2 k (t )  x2dk 1 (t )  x2dk (t )   kkFD Dk e2 k 1 (t )  kkI  e2 k 1 ( )d (10)

t
0
Taking time derivative from sliding surface given in (10) one gets
Sk (t )  k Sk (t )  kkP e2k 1 (t )  kkD  e2k (t )  x2dk 1 (t )  x2dk (t )   kkFD Dk e2k 1 (t )  kkI e2k 1 (t ) (11)
The control input can be given as uk  ukeq  uksw , where ukeq and uksw are the equivalent
control and the switching control, respectively. The equivalent control, ukeq , proposed by
Utkin [30] is based on the nominal (estimated) plant parameters (i.e. model uncertainty are
not considered) and provides the main control action, while the switching control, uksw ,
ensures the discontinuity of the control law across sliding surface, supplying additional
control to account for the presence of model uncertainty.
By substituting (8) into (11) and then equating Sk (t ) to zero, the overall control signal
become as:
uk (t )  bk1  kkD  k 
D e2 k 1 (t )  kkI e2 k 1 (t )  kkP e2k 1 (t )  k Sk (t )  f k ( x)  x2dk 1 (t )   uksw (12)
1 FD
 k

where uksw is the switching function and is introduced to be as
uksw  ks Sk (t )  kks sgn(Sk (t ))  (13)
where ks , kks are switching feedback control gains to be chosen in sequel, and
1 if Sk (t )  0

sgn( Sk (t ))  0 if Sk (t )  0 (14)

1 if Sk (t )  0
Substitution of (12) into (11), results in:
Sk (t )  kkD f k ( x)  kkDbk ks Sk (t )  kkDbk kks sgn(Sk (t )) (15)
The switching controllers (13) involve a discontinuous control action owing to the term
sgn(Sk (t )) . There arises the practical issue of chattering due to imperfections in switching
devices and delays. In addition, the chattering causes oscillations of the control input around
the zero sliding surface, resulting in unwanted wear and tear of actuators. Recently,
researchers have utilized the adaptive techniques together with the sliding mode control for
many engineering systems to smooth the output from a sliding mode controller and alleviate
the chattering in the pure sliding mode control [31]. Therefore, an adaptation law is derived in
section 4 to estimate switching control gains ks , kks . Stability and Lyapunov analysis are
carried out in the next subsection.
3.2. Stability Analysis
If the system is trapped on the sliding surface, namely Sk  Sk  Sk  0 , then the tracking
error e2 k 1 converges to zero exponentially if the coefficients, kkP , kkD , kkI and kkFD , are
selected properly, in turn, the fractional-order differential equation (FODE)
kkP e2k 1  kkD e2k 1  kkFD D e2k 1  kkI e2k 1  0 is stable. This implies that the closed-loop system
is globally asymptotically stable.
The following Lemma addresses that how sliding surface parameters kkP , kkD , kkI and
kkFD could be chosen such that FODE kkP e2k 1  kkD e2k 1  kkFD D e2k 1  kkI e2k 1  0 be
asymptotically stable.
Lemma 1. FODE kkP e2k 1  kkD e2k 1  kkFD D e2k 1  kkI e2k 1  0 is asymptotically stable if
  
   arg( )   [32].
 2qk 
Proof. Regardless of initial conditions, the Laplace transform of


P
k e
k 2 k 1 k e D
k 2 k 1  k D e2k 1  k e
FD
k
I
k 2 k 1  0 is Fk (s) E2k 1 (s) , where E2k 1 (s) is the Laplace
transform of e2 k 1 (t ) and Fk ( s) is
Fk (s)  kkP s  kkD s 2  kkFD s1  kkI (16)
Assuming  k  pk qk where, pk ,qk are positive constants so that (pk ,qk )  1 , and
1
( )
s qk
, then (16) can be rewritten as
Fk ( )  kkP  qk  kkD  2qk  kkFD  pk  qk  kkI (17)
Then it is obvious that the FODE kkP e2k 1  kkD e2k 1  kkFD D e2k 1  kkI e2k 1  0 is
asymptotically stable if parameters kkP , kkD , kkI and kkFD could be chosen so that
  
   arg( )  .
 2qk 
According to the proposed Lemma 1 the sliding surface is asymptotically stable;
e2k 1 (t )  0 as t   .
In order to analyze the stability analysis of the proposed second order sliding mode
control, under control action (12) take the following Lyapunov function
Vk 
2

1 2
Sk  Sk2  (18)
The stability is guaranteed if the derivative of the Lyapunov function is negative definite,
also known as the reaching condition. Taking the first time derivative from (18) one gets:
Vk  Sk Sk (1  ks bk kkD )  kkD Sk f k ( x)  kkDbk kks Sk (19)
Assuming that f k ( x) are bounded, namely k  sup x f k ( x) , and ks  1 kkD bk , then
one gets 1  ks kkD bk  0 . As a result, the expression (19) can be rewritten as
Vk   Sk Sk 1  ks bk kkD   kkDk Sk  kkD bk kks Sk

(20)
  Sk  S 1   b k   k
k
s D
k k k
D
b k  kkDk
s
k k k 
1  Sk  k kk bk  1 
s D
Therefore, provided k   s
k   k
 to dominate the model matched
bk  kkD 

uncertainties [33], then Vk will be negative which implies asymptotic stability of the system
and Sk (t ) and Sk (t ) tend to zero as t   .
4. Adaptation Law Synthesis

In this section an adaptive scheme based on Lyapunov function for switching feedback
control gains is derived. The adaptation law guarantees the system stability regardless of
model uncertainty bound. Take the following Lyapunov function
Vk 
1 2
2

Sk  Sk2  2k11 (ks )2  2k1 (kks )2  (21)
where k  ˆk  k , kk  kˆk  kk denote the estimation error. Taking time derivative from
s s s s s s
(21) with respect to time and with the sense of (20), one has:
Vk   Sk Sk 1  ks bk kkD    kkDbk kks  kkDk  Sk  2k11ks ks  2k1kks kks (22)
adding and subtracting some terms, Eq. (22) can be rewritten as follows
Vk   Sk  S 1  ˆ k b   kˆ b k
k
s D
k k k
s
k k
D
k 
 kkDk  kks bk kkD Sk  ks bk kkD Sk Sk
(23)
 1
   k k
s
2 k 1 k
s
k
1 s s
2k k k
It is clear that kˆk and ˆk can be chosen such that

s s
 Sk  S 1  ˆ b k   kˆ b k
k
s
k k
D
k
s
k k
D
k 
 kkDk  0 ,
so the adaptation law can be derived directly from (23) as follow
ks bk kkD Sk Sk   2k11ks ks  0  ks   2k 1kkD bk Sk Sk

(24)
kks bk kkD Sk   2k1kks kks  0  kks  2k kkD bk Sk
The switching controller parameters ks and k ks in switching control (13) are evaluated
through the adaptive algorithm derived in (24).
As the bound of uncertainty is not available, choosing switching controller gains
manually can make chattering effect worst. Therefore, by virtue of adaptation law (24)
switching controller parameters assume their gains by making a compromise between
chattering suppression and reaching condition.
5. Numerical Studies
The twin-tanks system comprised of two small tanks mounted above a reservoir which
provides storage for the water, Figure 2. Water is pumped into the bottom of each tank by two
independent pumps. The pump only increases the liquid level and is not responsible for
pumping the water out of the tank [8].
Figure 2. The plant of a liquid-level control system.
The twin-connected tanks system is a nonlinear dynamical system and the governing
dynamical equations can be written as [8, 34]
k1 1
h1   h1  h2 sign(h1  h2 )  q1
A1 A1
(25)
k k 1
h2  1 h1  h2 sign(h1  h2 )  2 h2  q2
A2 A2 A2
where h1 and h2 are the total water heads in Tank 1 and Tank 2 respectively, which are the
two outputs of interest, and q1 , q2 are the two inflows into the tanks. It is assumed that A1
and A2 are the capacities of Tank-1 and Tank-2 respectively. Both tanks assumed to have the
same cross sectional area i.e. A1  A2  A .
It has been seen in [35] that in chemical plants, selecting the flow rate as an input is more
effective than using flow as the input. Thus, if the flow rates are considered as the inputs, i.e.,
q1  u1 and q2  u2 , the system dynamics can be written as
k1 h1  h2 1
h1    u1
2 A h1  h2 A
(26)
k h1  h2 k2 1
h2  1  h2  u2
2 A h1  h2 2 A h2 A
It seems from (25) and (26) that there is a discontinuity at h1  h2 , so desired water level
is selected as hd1  hd 2 .
A mathematical model of Twin Tank can be expressed as follows:
x1 (t )  x2 (t )
k1 x2  x4 1
x2 (t )    u1
2 A x1  x3 A
(27)
x3 (t )  x4 (t )
k1 x2  x4 k2 1
x4 (t )   x4  u2
2 A x1  x3 2 A x3 A
System constraints: The flowed fluid into the tanks ( q1 and q2 ) cannot be negative
because the pumps can only pump water into the tanks. Therefore constraints on the inflow
are given by q1 , q2  0 . In the steady state, for constant water level set points, the respective
derivatives must be zero separately i.e. h1  0, h2  0 . Therefore, the inequality
2
k h
1
 2  1 must hold [8].
k  k2 h1
1
2 2
Both tanks have the same area of 70 cm , k1  14 and k2  10 , the set point water level is
2
selected as hd ,1  xd ,1  8 cm and hd ,3  xd ,3  6 cm , with initial water level h1 (0)  4 ,

h2 (0)  2 .
In Eq. (9) sliding surface parameters are selected to be as k1P  3 , k1D  7 , k1FD  1 ,
k1I  0.5 , 1  1 , 1  0.8 , k2P  3 , k2D  7 , k2FD  1 , 2  1 , k2I  0.5 ,  2  0.8 . The
initial conditions are chosen as S1 (0)  0.5 and S2 (0)  0.5 as well. In order to smooth the
control action, instead of sgn(.) function tanh(.) is used.
The simulation results with 20% variations in system’s nominal parameters (i.e
A  84 cm2 , k1  16.8 and k2  12 ) shows good stabilization of the water levels in both tanks
Figs. 3, dashed-line and solid-line represents system’s response and desired trajectory,
respectively. The value of inflow in both tanks settles to the steady state values shown in
Figure 3, the pumps only increases the liquid level, so their value is always positive.
9 7
8 6
7 5
h2(m)
h1(m)
6 4
5 3
4 2
3 1
0 10 20 30 0 10 20 30
time (sec) time (sec)
300
200
inlet flow, q1
100
-100
0 5 10 15 20 25 30 35
time (sec)
300
200
inlet flow, q2
100
-100
0 2 10 15 20 25 30 35
time (sec)
Figure 3. 2-SMC with PIDDα surface and +20% variation in parameters of Twin Tank system.
λ81
k81
λ82
k82
Figure 4. Adaptive parameters versus time.

Ṡ1
Ṡ2
S1 S2
Figure 5. Phase plot.
Adaptive parameters 1s , k 1s , 2s , k 2s in (24) are estimated as follow with zero initial
conditions shown in Figure 4.
1s  1k1D b1 S1 S1 , k1s  2 k1D b1 S1 , 2s  3 k2D b2 S2 S2 k2s  4 k2D b2 S2
where 1  0.8, 2  2.8, 3  1, 4  3.2 . In order to depict the transient behavior of

estimated parameters figures are zoomed with respect to time access and are depicted for 1
second.
The trajectories of the proposed control algorithm are demonstrated on the phase portrait
of the sliding variable in Fig 5, where S1 - S1 and S 2 - S 2 converge to zero.
Conclusion
In this study chattering-free robust second order sliding mode controller with PIDDα
sliding surface for nonlinear systems is investigated. The main virtue of this study relies on
fractional PIDDα sliding surface which provides more flexibility in the controller design. The
other objective is to find an effective chattering-free robust method for nonlinear dynamic
systems, by applying appropriate control action (12). Adaptation algorithm is derived based
on second order sliding mode control, to estimate switching controller parameters without any
knowledge about model uncertainty bound.
References
[1] U. Itkis, Control Systems of Variable Structure. NewYork: Wiley, 1976.
[2] W. Perruquetti, J. P. Barbot, Sliding mode control in engineering, New York, NY,
Marcel Dekker Inc., 2002.
[3] Levant A. Sliding order and sliding accuracy in sliding mode control. Internat. J.
Control. 1993; 58:1247-63.
[4] Ferrara A, Rubagotti MA. Sub-optimal second order sliding mode controller for
systems with saturating actuators. IEEE Trans Automat. Control. 2009; 54: 1082_7.
[5] Boiko I, Fridman L, Iriarte R, Pisano A, Usai E. Parameter tuning of second-order
sliding mode controllers for linear plants with dynamic actuators. Automatica 2006;
42:833-9.
[6] Davila J, Fridman L, Levant A. Second-order sliding mode observer for mechanical
systems. IEEE Trans Automat. Control. 2005; 50:1785-9.
[7] Bartolini G, Pisano A, Usai E, Second-order sliding-mode control of container cranes,
Automatica 38 (2002) 1783 – 1790.
[8] Khan MK, Spurgeon SK. Robust MIMO water level control in interconnected twin-
tanks using second order sliding mode control. Control Eng. Pract. 2006; 14:375-86.
[9] Levant A. Principles of 2-sliding mode design. Automatica 2007, 43(4), 576–86.
[10] Eker l, Second-order sliding mode control with experimental application, ISA
Transactions 49 (2010) 394-405.
[11] Mihoub M, Nouri AS, Abdennour RR. Real-time application of discrete second order
sliding mode control to a chemical reactor, Control. Eng. Pract. 2009; 17: 1089_95.
[12] Levant A. Homogeneity approach to high-order sliding mode design. Automatica, 2005,
41,823–30.
[13] I. Podlubny, Fractional differential equations. New York: Academic Press; 1999.
[14] R. Hilfer, Applications of fractional calculus in physics. Singapore: World Scientific;

2001.
[15] K. Miller, B. Ross, An Introduction to the Fractional Calculus and Fractional
Differential Equations, Wiley, New York, 1993.
[16] K. Oldham, J. Spanier, The Fractional Calculus, Academic Press, New York.
[17] H. Li, Y. Luo, Y. Chen, A Fractional Order Proportional and Derivative (FOPD)
Motion Controller: Tuning Rule and Experiments, IEEE Transaction on Control System
Technology, VOL. 18, NO. 2, MARCH 2010, 516 – 520.
[18] S. Ladacia, J. J. Loiseaua, A. Charefb, Fractional order adaptive high-gain controllers
for a class of linear systems, Communication in nonlinear science and numerical
simulation, Volume 13, Issue 4, July 2008, Pages 707–714.
[19] Y. Luo, Y.Q. Chen, Fractional order [proportional derivative] controller for a class of
fractional order systems, Volume 45,Issue 10, October 2009, Pages 2446–2450.
[20] Y. Q. Chen, B. M. Vinagre, I. Podlubny, Fractional Order Disturbance Observer for
Robust Vibration Suppression, Nonlinear Dynamics 38:355-367, 2004.
[21] B. M. Vinagre, I. Petras, I. Podlubny, Y. Q. Chen, Using Fractional Order Adjustment
Rules and Fractional Order Reference Models in Model-Reference Adaptive Control,
Nonlinear Dynamics 29: 269–279, 2002.
[22] H. Delavari, D.M. Senejohnny, Fractional-order controllers for robot manipulator, In:
Legnani G, Fassi I, editors. Robotics: state of the art and future trends, New York:
Nova Science Publishers; 2012, pp. 187–209. ISBN: 978-1-62100-403-5.
[23] B. Zhang, Y. Pi, Y. Luo, Fractional order sliding-mode control based on parameters
auto-tuning for velocity control of permanent magnet synchronous motor, ISA
Transactions, Volume 51, Issue 5, September 2012, Pages 649–656.
[24] Dadras S, Momeni HR. Fractional terminal sliding mode control design for a class of
dynamical systems with uncertainty. Commun Nonlinear Sci Numer Simulat, Volume
17, Issue 1, January 2012, Pages 367–377.
[25] M.Ö. Efe. Fractional fuzzy adaptive sliding-mode control of a 2-DOF direct-drive robot
arm. IEEE Trans Syst Man Cybern Part B: Cybern. 388 2008; 38:1561–70.
[26] Z. Man, A.P. Paplinski, H.R. Wu. A robust MIMO terminal sliding mode control for
rigid robotic manipulators. IEEE Trans Autom Control 1994; 39:2464–9.
[27] S. Dadras, H. Momeni. "Passivity-based fractional-order integral sliding-mode control
design for uncertain fractional-order nonlinear systems." Mechatronics, 23(7) (2013):
880-887.
[28] H. Delavari, R. Ghaderi, A. Ranjbar, S. Momani, Fuzzy fractional order sliding mode
controller for nonlinear systems, Communication in Nonlinear Science and Numerical
Simulation 15 (2010) 963–978.
[29] L. Bruzzone, G. Bozzini, Non dimensional Analysis of Fractional-Order PDD1/2 Control
of Purely Inertial Systems, Journal of Mechatronics and Applications, 2010.
doi:10.1155/2010/903420.
[30] VI. Utkin. Sliding modes in optimization and control problems. New York: Springer-
Verlag; 1992.
[31] W. Perruquetti, J. P. Barbot, Sliding mode control in engineering, New York, NY,
Marcel Dekker Inc., 2002.
[32] W.Deng, C.Li, J.Lu, Stability analysis of linear fractional differential system with
multiple time delays, Nonlinear Dynamic (2007) 48:409–416, Springer.
[33] M. Naderi, MR. Faieghi, Comments on second-order sliding mode control with
experimental application, ISA Transations, 2012;51(6):861–862.
[34] L.Chang, A MIMO sliding control with a second order sliding condition. In ASME
winter annual meeting, Dallas, Texas, (1990), paper No. 90-WA/DSC-5.
[35] J. Kantor, Non-linear sliding mode controller and objective function for surge tanks.
International Journal of Control, (1989) 50, 2025–2047.
Chapter 8
F EATURES OF F RACTIONAL O PERATORS

I NVOLVING F RACTIONAL D ERIVATIVES
AND T HEIR A PPLICATIONS TO THE P ROBLEMS
OF M ECHANICS OF S OLIDS
Yury A. Rossikhin∗and Marina V. Shitikova†

Research Center on Dynamics of Solids,
Voronezh State University of Architecture and Civil Engineering,
Voronezh, Russia
Abstract
The given Chapter consists of Introduction, 8 paragraphs and Conclusion. The state-
of-the-art review of papers devoted to the fractional derivatives, fractional operators
and their applications is presented in paragraph 1. In first four original paragraphs,
from 2 to 5, the theoretical foundations of Rabotnovs fractional operators are pre-
sented, namely: the definition of the fractional operators, their connection with frac-
tional derivatives, the multiplication theorem of fractional operators, the class of re-
solvent operators originated by Rabotnov operator. Besides, the definitions of the gen-
eralized Rabotnov operators are formulated, which are represented in terms of finite
sums of the fractional Rabotnov operators with one and the same fractional parameter.
The class of the resolvent operators, which are originated by the generalized Rabot-
nov operator, has been constructed. The main viscoelastic operators of frequent use in
applications, and particularly, the operator of cylindrical rigidity, are calculated under
two assumptions: (1) when the operator of volume expansion-compression is constant,
and (2) when it is an operator involving the Rabotnov fractional operator. The con-
nection between the Rabotnov fractional operator and Koellers operators is shown in
paragraph 5. The applications of the presented theories to the problems of vibrations of
oscillators, propagation of stationary shock waves in hereditarily elastic media, and of
impact interaction of an elastic sphere with viscoelastic beams are given, respectively,
in paragraphs from 6 to 8.
∗
E-mail address: yar@vgasu.vrn.ru
†
E-mail address: mvs@vgasu.vrn.ru
166 Yury A. Rossikhin and Marina V. Shitikova
Keywords: hereditarily elastic media, operators of fractional order, Volterra integral equa-
tions, Green functions, Rabotnov fractional exponential function
AMS Subject Classification: 26A33, 34E13, 34C15, 34C26, 37N15, 45D05, 70J30,
70K25
1. Introduction
The number of researchers who utilize fractional derivatives in their studies is growing
steadily year after year. It hardly probable that many of them even realize the reasons to
insert the fractional derivatives in all conceivable and inconceivable processes, even to the
detriment of the processes in themselves [1, 2, 82]. It seems likely that the exotic character
of the equations describing such processes charm researchers. As this takes place, some
slighting treatment of research advances of other scientists in the field, sometimes bordering
through one’s ignorance. Thus, for example, Usuki [85] when investigating Rayleigh waves
in a fractional derivative viscoelastic medium cited the second review paper by Rossikhin
and Shitikova [64] completely ignoring their first state-of-the-art article [55], wherein both
volume harmonic waves in a three-dimensional fractional derivative viscoelastic medium
(fractional derivative standard linear solid model) and Rayleigh waves in the same medium
have been considered in detail.
Recently our attention was attracted by a paper by Dal [16], wherein the author has
argued that Caputo derivative along with Riemann-Liouville derivative had been utilized by
Samko et. al. [73] and Rossikhin and Shitikova [64], what is a pure invention. It seems
likely that the author of [16] has read neither [73] nor [64].
In 1998, Rossikhin and Shitikova in their paper [58] devoted to the application of frac-
tional calculus in nonlinear vibrations of suspension bridges under the conditions of differ-
ent internal resonances pioneered in utilizing the approximate formula
D γ eiωt ≈ (iω)γ eiωt , (1)
which differs from the exact formula by one term

Z
γ iωt γ iωt sin πγ ∞ uγ
D e = (iω) e + e−ut du, (2)
π 0 u + iω
√
where ω is the frequency, t is the time, i = −1, γ (0 < γ ≤ 1) is the fractional parameter,
Z
1 d t x(s)ds
D γ x(t) = , (3)
Γ(1 − γ) dt 0 (t − s)γ
Γ(1 − γ) is the gamma-function, and x(t) is an arbitrary function.
And this approximate formula (1), together with a nontraditional definition of the frac-
tional derivative (see formula (5.82) in Samko et al [73] referring to Krasnosel’skii et al
[35] and Iosida [30]) γ
γ d
D = , (4)
dt
which is a good combination with the method of multiple time scales, allowed the authors
of [54, 58] to solve completely the stated problem.
Fractional Calculus in Mechanics of Solids 167
The same authors in 2009 [69] and 2012 [66] succeeded in showing that if one uses the
method of multiple time scales for the analysis of nonlinear vibration of mechanical systems
and restricts oneself by the zero- and first-order approximations, then the utilization of the
approximate formula is quite reasonable, since the second term of the exact formula (2)
does not enter into this approximation. This result is rather important, and it could not be
ignored. However, this practice recently was applied repeatedly by various authors without
any reference to its originators [16, 17].
This list could be continued, but we restrict ourselves by the enumerated examples,
because there is more important problem. This problem is connected with the fact that
the willingness of some researchers to accustom by all means to the usage of fractional
integro-differential calculus results in inept findings published in scientific literature. Thus,
equations with fractional forces of inertia made its appearance [3, 32, 37, 79, 82, 87], as
well as rheological models inconsistent with the laws of thermodynamics [34], intrinsic
damping [82] instead of the internal friction due to Zener [38, 66, 91], contradictions in
the evaluation of the applicability of different definitions of a fractional derivative, and in
particular, Riemann-Liouville definition and so-called Caputo definition [16], etc.
Many of enumerated and unstated absurdities could be prevented if instead of fractional
derivatives one uses fractional operators which connected at once with fractional integro-
differentiation, in so doing they possess pictorial physical meaning.
Investigation of properties of the fractional operators and their applications to the prob-
lems of Mechanics of Solids were carried out with a great success by the Russian school
of thought leaded by the prominent scientist of the XXth century, Academician Yury N.
Rabotnov [47, 48] (the authors of this Chapter also belong to this school of mechanicians).
The chronological list of the main contributions made prior to 1980 by Western and Rus-
sian researchers in the field of fractional calculus applications in linear viscoelasticity was
presented by Rossikhin in his retrospective article [50], wherein they have been summarized
in Table 1 showing to the present generation of researchers applying fractional calculus in
many branches of science that the work of these pioneers was primarily responsible for the
development of the theory of fractional calculus viscoelasticity. The simplest fractional cal-
culus equations modeling the viscoelastic features of materials using the two approaches,
i.e., via the Boltzmann-Volterra relationships with weakly singular kernels and via frac-
tional derivatives or fractional integrals, as well as other types of fractional operators, are
presented in the first column, while the Russian and Western authors who proposed them
for the first time (to our knowledge) in problems of viscoelasticity are cited in the second
and third columns, respectively. The last column shows the papers wherein these models
were implemented for the first time for solving different dynamic problems of mechanics
of solids and geophysics.
The enumeration of the models is started from the fractional Newtonian model sug-
gested by Scott Blair [8] and Gerasimov [20] using the second and first approaches, respec-
tively. This simplest fractional calculus element was further named by Koeller [33] as the
spring-pot with the memory parameter γ, because it has the property that its constitutive
equation has continuity from the ideal solid state at γ = 0, to the ideal fluid state at γ = 1.
This model was used by Caputo [13] to study vibrations of an infinite viscoelastic layer.
The fractional calculus standard linear solid model proposed by Gross [21] and Rabot-
nov [46] was employed with great success in the 1970s by Russian scholars for solving a
rich variety of dynamic engineering problems [22, 23, 24, 40, 42, 70, 88]. Thus, in 1966,
Rozovskii and Sinaiskii [70] were the first to study the impulse response of an oscillator
representing the order of the fractional exponent as the proper fraction, γ = q/n, where
q and n are integers, that results in the rationalization of the corresponding characteristic
equation. This procedure is now widely used by the authors dealing with numerical algo-
rithms in treating equations with fractional operators. The Green functions for the fractional
standard linear solid and Maxwell oscillators were obtained in 1970 by Zelenev et al. in
Refs. [88] and [89], respectively.
As for the wave propagation in hereditarily elastic media, then harmonic bulk waves
were investigated in 1968 by Meshkov and Rossikhin [42], and the impulse load propaga-
tion in a rod made of the fractional standard linear solid material was studied by Gonsovskii
and Rossikhin [23] (since this paper was originally published in a hard-to-get Russian vol-
ume, then the main results of Ref. [23] were presented in Sec. 4.3 of Ref. [55]). The
impact of a hereditarily elastic rod with the Rabotnov kernel of heredity against a rigid
barrier was studied by Gonsovskii et al. [22] in 1972. The nearby-front asymptotic of the
stress waves in a semi-infinite rod originated by the constant load application to its free end
at the initial time was obtained in 1973 by Gonsovskii and Rossikhin [24]. Two years later,
the asymptotic solution in the neighborhood of the step pulse onset in a similar problem for
a viscoelastic rod made of the fractional Maxwell material was presented by Buchen and
Mainardi [9], using the second approach for γ = 1/2.
Rabotnov [47] was a ground breaker in the generalization of the fractional calculus stan-
dard solid model suggesting in 1966 the rheological models which describe the dissipative
processes with several relaxation (retardation) times based on the fractional exponential
function proposed by him in Ref. [47]. He further used such models in Ref. [48] for an
approximation of the relaxation and creep functions determined experimentally for sev-
eral polymeric materials. The presence of a large variety of rheological parameters allows
one to describe adequately the behavior of the advanced polymeric materials possessing
the complex relaxation and creep functions involving several plateau and several transition
zones.
Rzhanitsyn [49] proposed in 1949 another weakly singular kernel of heredity as a creep
kernel, which was written in the form of a fractional operator different from the fractional
derivative or fractional integral. Its resolvent kernel of relaxation was found by Vulfson
[86] in 1960. In 1967, Meshkov [38] suggested the equivalent form for the Rzhanitsyn
model, while Slonimsky [77] employed it for describing relaxation processes in polymers
with a chain structure of macromolecules. Four years later, Meshkov and Rossikhin [43]
suggested to use the Rzhanitsyn kernel as the relaxation kernel in the Boltzmann-Volterra
equation and utilized this model to study the impulse response of a viscoelastic oscillator.
Belov and Bogdanovich [6] used the model proposed in Ref. [43] for investigating the
propagation of a load impulse in a semi-infinite rod and constructed its asymptotic solution.
Table 1. A historical overview of Russian and Western contributors to the use of Fractional Calculus in Mechanics of Solids, from
the 1940’s to the 1970’s
The simplest fractional calculus models Russian researchers Western researchers Applications carried out prior to 1980
fractional Newtonian model
σ(t) = E∞ τεγ D γ ε(t) Scott Blair (1944) [8]
σ(t) = E∞ τεγ I 1−γ dε(t) Gerasimov (1948) [20] Bland (1960) [93] Caputo (1976) [13]
standard linear solid model
(a) via Boltzmann
h − Volterra relationshipsi
Rt
σ = E∞ ε − νε 0 3γ (−t0 /τε )ε(t − t0 )dt0 Rabotnov (1948) [46] Gross (1947) [21] Rozovskii and Sinaiskii (1966) [70]
h i
ε = J∞ σ + νσ 0t 3γ (−t0 /τσ )σ(t − t0 )dt0
R
Meshkov and Rossikhin (1968) [42]
Zelenev et al. (1970) [89]
Meshkov et al. (1971) [40]
Gonsovskii et al. (1972) [22]
Gonsovskii and Rossikhin (1972) [23],
(1973) [24]
(b) via fractional derivatives
σ + τεγ D γ σ = E0 (ε + τσγ D γ ε) Meshkov (1967) [38] Caputo and Mainardi (1971) [14] Caputo and Mainardi (1971) [14, 15]
(b) via fractional
» operators –
τε−γ I γ
σ = E∞ 1 − νε −γ ε Rabotnov (1966) [47]
1+τε Iγ
Kelvin − Voigt model
(a) via Boltzmann − Volterra relationships
Rt
ε = J0 0 3γ (−t0 /τσ )σ(t − t0 )dt0 Shermergor (1966) [76]
γ
σ = E0 (ε + τσ D γ ε) Shermergor (1966) [76] Caputo (1967) [11] Caputo (1967) [11], (1974) [12]
Smit and de Vries (1970) [78] Bagley and Torvik (1979) [4]
Table 1. (Continued)
The simplest fractional calculus models Russian researchers Western researchers Applications carried out prior to 1980
Maxwell model
(a) via Boltzmann
h − Volterra relationships
i
Rt
σ = E∞ ε − 0 3γ (−t0 /τε )ε(t − t0 )dt0 Shermergor (1966) [76] Zelenev et al. (1970) [88]
1/2
σ + τε D 1/2 σ = η ε̇ Gemant (1936) [19]
σ + τεγ D γ σ = E∞ τεγ D γ ε Meshkov (1967) [38] Caputo and Mainardi (1971) [15] Buchen and Mainardi (1975) [9]
(c) via fractional integral
ε = J∞ (σ + τε−γ I γ σ) Shermergor (1966) [76]
generalized
h standard linear solid model i
RtP
σ = E∞ ε − 0 n 0 0
j=1 ej 3γ (−t /τεj )ε(t − t )dt
0 Rabotnov (1966) [47]
Fractional operator with Rzhanitsyn kernel
(a) as a »
creep kernel
“ ”−γ –
d
ε = J∞ 1 + νσ 1 + τσ dt σ Meshkov (1967) [38]
» – Slonimsky (1967) [77]

R γ−1 −t0 /τσ
ε = J∞ σ + νσ 0t t τ γeΓ(γ) σ(t − t0 )dt0 Rzhanitsyn (1949) [49]
σ
h i
σ = E∞ ε − 0t e−t /τσ 3γ (−νσ t0 /τσ )ε(t − t0 )dt0
R 0
Vulfson (1960) [86]
(b) as a relaxation
» kernel –
R t tγ−1 e−t0 /τε 0 )dt0
σ = E∞ ε − νε 0 γ
τ Γ(γ)
ε(t − t Meshkov and Rossikhin Meshkov and Rossikhin (1971) [43]
ε
(1971) [43] Belov and Bogdanovich (1976) [6]
The fractional derivative Kelvin-Voigt and standard linear solid models were first pro-
posed by Shermergor [76] and Meshkov [38] in 1966 and 1967, respectively, and then
independently but at a somewhat later time by Caputo [11] and Caputo and Mainardi [14],
respectively. The early applications of these models were made by Caputo [11] in 1967 and
Caputo and Mainardi [14, 15] in 1971 for solving the problems dealing with geophysics.
In 1974, the fractional derivative Kelvin-Voigt model was utilized by Caputo [12] to study
vibrations of an infinite viscoelastic layer.
In the late 1970s, an investigation on fractional derivative models and their application
to the problems of structural dynamics was initiated in the USA by Torvik and Bagley. Their
first paper in the field [4] dated to 1979 was devoted to modeling an elastomer damper
as a fractional derivative Kelvin-Voigt oscillator. During the next decade, these authors
contributed significantly to incorporate the fractional derivative standard linear solid model
into the numerical procedures for investigating viscoelastically damped structures [5, 83,
84].
Thus, from the papers cited in the forth column of Table 1 it is clearly evident what a
tremendous work was carried out in the late 60’s and 70’s by Russian and Italian researchers
in the application of fractional calculus viscoelastic models (formulated earlier or in the
same period using the two approaches as shown in the first three columns of Table 1) for
solving dynamic problems in the mechanics of solids and geophysics.
2. Rabotnov’s Fractional Operators and Main Formulas

of Algebra of Fractional Operators
We shall proceed our presentation in this Section from the analysis of the rheological
model, which is called as the fractional derivative standard linear solid model
σ + τεγ D γ σ = E0 (ε + τσγ D γ ε), (5)
in so doing
τεγ τσ−γ = E0 E∞
−1
, (6)
where σ is the stress, is the strain, τ and τσ are the relaxation and retardation times,
respectively, E0 and E∞ are the relaxed (prolonged modulus of elasticity, or the rubbery
modulus) and nonrelaxed (instantaneous modulus of elasticity, or the glassy modulus) mag-
nitudes of the elastic modulus, respectively.
Meshkov [38] was the first to derive formula (5) in 1967.
Expressing σ in terms of ε or ε in terms of σ from (5) with due account for (6) yields

1
σ = E∞ ε − νε γ ε , (7)
1 + τε D γ

1
ε = J∞ σ + νσ σ , (8)
1 + τσγ D γ
where E0 = J0−1 , E∞ = J∞
−1
,
E∞ − E0 J0 − J∞ E∞ − E0 J0 − J∞
νε = = , νσ = = . (9)
E∞ J0 E0 J∞
Relationships (7) and (8) involve one and the same operator which will be denoted
further as 3∗γ (τiγ ), i.e.,
1
3∗γ (τiγ ) = (i = σ, ε). (10)
1 + τiγ D γ
Operator (10) was introduced into consideration by Yu. N. Rabotnov in a little bit
another form [46].
In order to obtain operator (10) in another form, we multiply the numerator and denom-
inator of the fraction in (10) by I γ τi−γ , where
Z t
γ (t − s)γ−1
I x(t) = x(s)ds (11)
0 Γ(γ)
is the fractional integral.

Considering that D γ I γ = 1, we find
I γ τ −γ
3∗γ (τiγ ) = i . (12)
1 − −I γ τi−γ
If we suppose that the right part of formula (12) is the sum of an infinite decreasing
geometrical progression, the denominator of which is equal to d = −I γ τi−γ , then 3∗γ (τiγ )
could be represented as
∞
X −γ(n+1) γ(n+1)
3∗γ (τiγ ) = (−1)n τi I , (13)
n=0
or Z t
3∗γ (τiγ ) x(t) = 3γ (−s/τi ) x(t − s)ds, (14)
0
where
∞
tγ−1 X (−1)n (t/τi)γn
3γ (−t/τi ) = γ . (15)
τi n=0 Γ[γ(n + 1)]
Considering (14), formulas (7) and (8) take the form
Z t
σ = E∞ ε − νε 3γ (−s/τε )ε(t − s)ds , (16)
0
Z t
ε = J∞ σ + νσ 3γ (−s/τσ )σ(t − s)ds . (17)
0
Formulas (16) and (17) are Boltzmann-Volterra relationships with weakly singular ker-
nels of heredity 3γ (−t/τi ), which attenuate at t → ∞, in so doing resolvent kernels occur
to be the same. Only exponential kernels possess this feature, and the kernels (15) go over
into exponential kernels at γ = 1, i.e.,
3γ (−t/τi ) = τi−1 exp (−t/τi ) . (18)

These properties of 3γ (−t/τi )-function allowed Rabotnov to call it as a fractional

exponential function.
Relationships (16) and (17) sometimes are written in the form

e
σ = Eε, Ee = E∞ 1 − νε 3∗γ (τεγ ) , (19)

e
ε = Jσ, Je = J∞ 1 + νσ 3∗γ (τσγ ) . (20)
In conclusion of this Section we present the important formula for the multiplication of
the operators
∗ γ ∗ γ
τεγ 3∗γ (τεγ ) − τσγ 3∗γ (τσγ )
3γ (τε ) 3γ (τσ ) = . (21)
τεγ − τσγ
In fact
1 A1 A2
3∗γ (τεγ ) 3∗γ (τσγ ) = = γ γ + , (22)
(1 + τεγ D γ ) (1 + τσγ D γ ) 1 + τε D 1 + τσγ D γ
where
τεγ τσγ τσγ
A1 = γ, A2 = −A1 γ = − γ .
τεγ
− τσ τε τε − τσγ
It should be noted that formula (21) does not coincide with that presented in Rabotnov
[48] (see formula (5.10) there), since we use the dimensionless operator 3∗γ (τiγ ), while the
dimensional operator was utilized in [48].
3. Generalized Rabotnov Operators

Generalized Rabotnov operators are written in the form
 
X n
E∗ = E∞ 1 − mj 3∗γ tγ  , j (23)
j=1
" n
#
X
J∗ = J∞ 1 + ni 3∗γ (τiγ ) , (24)
i=1
where mj , tj (j = 1, 2, ..., n) and ni , τiγ (i = 1, 2, ..., n) are constants.
Since operators (19) and (20) are reciprocal, i.e., E∗ J∗ = J∗ E∗ = 1, then the following
relationship should be valid:
" # 
Xn Xn
1= 1+ ni 3∗γ (τ γ ) 1 − i mj 3∗γ tγ  . (25)
j
i=1 j=1
Assume that the constants ni and τiγ (i = 1, ..., n) are known, and it is necessary to
determine the constants mj and tγj (j = 1, ..., n). Using formula (21), let us rewrite (25) in
the form

Xn n
X X n X n τiγ 3∗γ (τiγ ) − tγj 3∗γ tγj
ni 3∗γ (τiγ ) − mj 3∗γ tγj − ni mj = 0,
τiγ − tγj
i=1 j=1 i=1 j=1
whence it follows that

  " #
n n γ n n γ
X γ
X m j τ X γ
X ni tj
i 
ni 3∗γ (τi ) 1 − γ γ − mj 3∗γ tj 1− γ γ = 0. (26)
τi − tj τi − tj
i=1 j=1 j=1 i=1
From (26) we find

n
X mj τiγ
1− = 0, (27)
τiγ − tγj
j=1
n
X ni tγj
1− γ γ = 0, (28)
τi − tj
i=1
or in another form
n
X mj t−γ
j
1+ = 0, (29)
j=1
τi−γ − t−γ
j
n
X ni τi−γ
1+ = 0. (30)
i=1
τi−γ − t−γj
From the n-th order Eqs. (28) or (30) we could define n magnitudes of t±γ j (j =
±γ
1, ..., n), while knowing tj from the set of n Eqs. (27) or (29) we find the values mj
(j = 1, ..., n).
γ
Suppose now that constants mj and tj (j = 1, ..., n) are known, and it is a need to
γ
determine constants ni and τi (i = 1, ..., n). In this case, from the n-th order Eqs. (27) or
(29) we could define n magnitudes of τi±γ , while knowing τi±γ , we could find the values of
ni from the set of n Eqs. (28) or (30).
The set of equalities (30) implies that the values t−γ
j are the roots of the equation
n
X ni τi−γ
F1 (x) = 1 + = 0. (31)
i=1
τi−γ − x
Equation (31) possesses n real positive roots if τi−γ > 0 and ni > 0.
−γ
If we suppose that τk+1 > τk−γ , then zeros of the function F1 (x) are located between
the poles, thus
τk−γ < t−γ −γ
k < τk+1 , τn−γ < t−γ
n . (32)
The plot of the function F1 (x) for n = 2 is presented in Fig. 1a, where x = t−γ .
The set of equalities (29) represents a system of n linear equations for n unknown
coefficients mj . Assign that
n
X mj t−γ
j
F2 (y) = 1 + , (33)
j=1
y − t−γ
j
where y = τ −γ .
Function F2 (y) (33) possesses n zeros at y = τi−γ , while at y → t−γ
j it tends to +∞ or
−∞ depending on the sign of mj and the direction of the transition to the limit. Since the
Figure 1. Functions F1 (x) and F2 (y).
points τi−γ are located between the points t−γ −γ

j , except the point τ1 (see formula (32)), then
the plot for the function F2 (y) for n = 2 should be as that in Fig. 1b, but this is possible
only when all mj are positive.
At n = 1, from (27) and (28), (19) and (20) we obtain
τσγ
1 + νε = 0, (34)
τεγ − τσγ
γ
τε
1 + νσ = 0, (35)
τεγ − τσγ
whence it follows γ γ
τε τσ
1 − νε = , 1 + νσ = . (36)
τσ τε
Considering (9), formulas (36) go over in relationships (6).
At n = 2, operators J∗ and E∗ take the form

J∗ = J∞ 1 + n1 3∗γ (τ1γ ) + n2 3∗γ (τ2γ ) , (37)
and
E∗ = E∞ 1 − m1 3∗γ (tγ1 ) − m2 3∗γ (tγ2 ) . (38)
Multiplying reciprocal operators (37) and (38) yields
n1 τ1−γ n2 τ2−γ
1+ + = 0, x = t−γ
i , (39)
τ1−γ − x τ2−γ − x
(
a1 m1 + a2 m2 = 1,
(40)
b1 m1 + b2 m2 = 1,
where it is evident from Fig. 1 that
t−γ t−γ
a1 = −γ 1 −γ > 0, a2 = −γ 2 −γ > 0,
t1 − τ1 t2 − τ1
t−γ
1 t−γ
2
b1 = < 0, b2 = > 0.
t−γ
1 − τ2−γ t−γ
2 − τ2−γ
Note that the signs of the values a1 , a2 , b1 , and b2 could be established with help of
Fig. 1.
Equation (39) could be written in the form
h i
x2 − τ1−γ (1 + n1 ) + τ2−γ (1 + n2 ) x + τ1−γ τ2−γ (1 + n1 + n2 ) = 0, (41)
and its roots are defined by formula

1 h −γ i
x1,2 = t−γ 1,2 = τ1 (1 + n1 ) + τ2−γ (1 + n2 )
2
rh i2
1
± τ1−γ (1 + n1 ) − τ2−γ (1 + n2 ) + 4τ1−γ τ2−γ n1 n2 , (42)
2
and both roots are positive.
Substituting the found roots (42) in (40), we have
b2 − a2 a1 − b1
m1 = > 0, m2 = > 0, (43)
a1 b2 − a2 b1 a1 b2 − a2 b1
which are positive as well.
If, vice versa, it is necessary to determine the values of τi−γ and ni , then using Eqs. (29)
and (30), we have
m1 t−γ m2 t−γ
1 − −γ 1
− −γ 2 = 0, x = τi−γ , (44)
t1 − x t2 − x
 −γ −γ
 −γτ1 −γ n1 + −γτ2 −γ n2 = 1,

t1 −τ1 t1 −τ2
−γ −γ (45)

 −γ 1 −γ n1 + −γτ2 −γ n2 = 1.
τ
t2 −τ1 t2 −τ2
It is convenient to rewrite (44) in the form

h i
−γ −γ −γ −γ
x2 − t1 (1 − m1 ) + t2 (1 − m2 ) x + t1 t2 (1 − m1 − m2 ) = 0. (46)
The solution of (46) gives us the values of τ1−γ and τ2−γ , and then substituting τ1−γ and
τ2−γ in (45), we find n1 and n2 .
Let us present here two more useful formulas, which are the immediate generalization
of formulas (36).
Using the Viéte theorem, from (41) and (46) we have
t−γ −γ −γ −γ
1 t2 = τ1 τ2 (1 + n1 + n2 ),
τ1−γ τ2−γ = t−γ −γ

1 t2 (1 − m1 − m2 ),
or
t−γ
1 t2
−γ
= 1 + n1 + n2 , (47)
τ1−γ τ2−γ
−γ −γ
τ1 τ2
−γ = 1 − m1 − m2 . (48)
t−γ
1 t2
Formulas (47) and (48) are generalized easily over the case of the operators defined by
relationships (23) and (24). In this case, they look like as follows:
 γ
Qn
 τi  Xn
 i=1 
Q  = 1 + ni , (49)
 n 
tj i=1
j=1
 n
γ
Q
 tj n
 j=1  X
Q  = 1 − mj . (50)
 n 
τi j=1
i=1
4. Calculation of the Main Viscoelastic Operators
4.1. e
The Case of Constant Operator of the Bulk Extension-Compression K
For the majority of viscoelastic materials, the bulk modulus remains a constant value
during the process of mechanical deformation, i.e.,
Ee E∞
= , (51)
1 − 2e
ν 1 − 2ν∞
where ν∞ is the nonrelaxed Poisson’s ratio, and νe is Poisson’s operator.

From (51) with due account for (19) we have
1
νe = ν∞ + (1 − 2ν∞ ) νε 3∗γ (τεγ ) . (52)
2
Further we will require the operators (1 + νe)−1 and (1 − νe)−1 , which are determined
considering (52) according to the above developed procedure, i.e.,
1 1
1 γ = 1 − B 3∗γ (tγ1 ) , (53)
∗
1 + ν∞ + (1 − 2ν∞ ) νε 3γ (τε )
2
1 + ν∞
1 1 ∗ γ
1 γ = 1 + D 3γ (t2) , (54)
1 − ν∞ − (1 − 2ν∞ ) νε 3∗γ (τε )
2
1 − ν∞
where B, tγ1 , and D, tγ2 are yet unknown constants.

Multiplying the operators in the right-hand sides of Eqs. (53) and (54), respectively, by
the denominator of the fraction in their left-hand side parts and considering (21), we are led
to the following equations:

1 (1 − 2ν∞ ) νε τεγ 1 (1 − 2ν∞) νε tγ1
1−B γ ∗ γ
3γ (τε ) − B 1 − 3∗γ (tγ1 ) = 0,
2 1 + ν∞ τε − tγ1 2 1 + ν∞ τεγ − tγ1
(55)
τγ tγ2
» – » –
1 (1 − 2ν∞ ) νε 1 (1 − 2ν∞ ) νε
− 1 + D γ ε γ 3∗γ (τεγ ) + D 1 + γ γ 3∗γ (tγ2 ) = 0. (56)
2 1 − ν∞ τε − t2 2 1 − ν∞ τε − t2
Vanishing to zero the terms in square brackets in (55) and (56), we could determine the
unknown constants
(1 − 2ν∞ )νε τεγ − tγ1
B= = , (57)
2(1 + ν∞ ) + νε (1 − 2ν∞ ) τεγ

(1 − 2ν∞ ) νε τεγ
t−γ
1
−γ
= τε 1+ , tγ1 = , (58)
2(1 + ν∞ ) A
(1 − 2ν∞ )νε τεγ − tγ2
D= =− , (59)
2(1 − ν∞ ) − νε (1 − 2ν∞ ) τεγ

(1 − 2ν∞ ) νε τεγ
t−γ
2 = τε−γ 1 − , tγ2 = , (60)
2(1 − ν∞ ) C
where
2(1 + ν∞ ) + νε (1 − 2ν∞ ) 2(1 − ν∞ ) − νε (1 − 2ν∞ )
A= > 1, C= < 1.
2(1 + ν∞ ) 2(1 − ν∞ )
Now we could calculate operators µ e Really, considering (19) and (53), (57),
e and λ.
(58), we find
Ee E∞ 1
µ
e= = 1 − νε 3∗γ (τεγ ) 1 − B 3∗γ (tγ1 ) . (61)
2(1 + νe) 2 1 + ν∞
Multiplying operators entering in (61) and considering (21), we have

γ γ
τε ∗ γ t1 ∗ γ
µ
e = µ∞ 1 − νε 1 − B γ 3γ (τε ) − B 1 + νε γ 3γ (t1 ) , (62)
τε − tγ1 τε − tγ1
where µ∞ is the nonrelaxed shear modulus.

But according (57)
1 − Bτεγ (τεγ − tγ1 ) = 0,
−1
and
tγ 3νε
B 1 + νε γ 1 γ = , (63)
τε − t1 2(1 + ν∞ )A
and finally relationship (62) with due account for (63) takes the form
γ
3νε ∗ τε
µ
e = µ∞ 1 − 3γ . (64)
2(1 + ν∞ )A A
e1 , let us use the following formula:
To define operator λ
e νe
E 1 E e 1 E e
e=
λ = − . (65)
(1 − 2e
ν )(1 + νe) 3 1 − 2e
ν 3 1 + νe
The first operator in the right-hand side of (65), according to (51), is a constant, while
the second one within an accuracy of a constant coincides with formula (64). Thus,
γ
e 1 E∞ 1 E∞ 3νε τε
λ= − 1− 3∗γ ,
3 1 − 2ν∞ 3 1 + ν∞ 2(1 + ν∞ )A A
or
e (1 − 2ν∞ )νε ∗ τεγ
λ = λ∞ 1 + 3 , (66)
2(1 + ν∞ )A γ A
where λ∞ is the nonrelaxed magnitude of the second Lame parameter.
In conclusion of this Subsection, we will calculate the operator
Ee Ee 1 1

= + , (67)
1 − νe2 2 1 + νe 1 − νe
which within an accuracy of a constant coincides with the operator of cylindrical rigidity
and is of frequent use in the problems dealing with mechanical behavior of thin bodies.
Substituting (19), (53), and (54) in (67), multiplying the operators involving in the
enumerated relationships and using (21), we obtain
Ee E∞ γ γ
2
= 2
1 − m1 3∗γ (t1 ) − m2 3∗γ (t2 ) , (68)
1 − νe 1 − ν∞
where
3 B(1 − ν∞ ) 1 D(1 + ν∞ )
m1 = , m2 = . (69)
2 (1 − 2ν∞ ) 2 (1 − 2ν∞ )
In order to find an operator reverse to the operator (68), let us represent it in the form
1 − νe2 2
1 − ν∞
= 1 + n1 3∗γ (τ1γ ) + n2 3∗γ (τ2γ ) , (70)
Ee E∞
where τ1γ , τ2γ , and n1 , n2 are yet unknown constants which are determined from Eqs. (44)
and (45). It is convenient to rewrite these equations in another form. Thus, the equation for
defining the values τiγ has the form
x x γ
1 + m1 + m2 γ = 0, x = τi (i = 1, 2), (71)
tγ1 − x t2 − x
or with due account for (69)

3 B(1 − ν∞ ) x 1 D(1 + ν∞ ) x
1+ γ + γ = 0. (72)
2 (1 − 2ν∞ ) (t1 − x) 2 (1 − 2ν∞ ) (t2 − x)
A set of equations for finding n1 and n2 is the following:
 tγ γ
 γ 1 γ n1 + γt1 γ n2 = 1,
τ −t
1 1 τ −t 2 1
tγ2 tγ2
(73)
 n1 + n2 = 1.
τ1γ −tγ2 τ2γ −tγ2
It could be shown that if substitute τεγ instead of x in (71) or (72) , then the following
identity could be obtained:
τεγ τεγ
1 ≡ −m1 − m 2 . (74)
tγ1 − τεγ tγ2 − τεγ
In order to find the second root of Eq. (71) or (72), let us rewrite it in the form
(1 − m1 − m2 ) x2 − (tγ1 + tγ2 − m1 tγ2 − m2 tγ1 ) x + tγ1 tγ2 = 0. (75)
Due to the Viéte theorem, if the first root is x1 = τ1γ = τεγ , then the second root x2 = τ1γ
satisfies the relationship
γ tγ1 tγ2
τ ε x2 = . (76)
1 − m1 − m2
Calculations show that
tγ1 tγ2 τε2γ
= . (77)
1 − m1 − m2 1 − νε
But τεγ (1 − νε )−1 = τσγ , and from (77) it follows that x2 = τ2γ = τσγ .
Substituting the found τ1γ = τεγ and τ2γ = τσγ in the set of Eqs. (73), we obtain

 2(1−ν∞ ) n1 − 2(1−ν∞ ) n2 = −1,
(1−2ν∞ )νε νσ
2(1+ν∞ )
(78)
 n1 + 2 1+ν∞
n2 = 1,
(1−2ν∞ )νε 3 νσ
whence it follows
(1 − 2ν∞ )2 νε 3νσ
n1 = 2 )
, n2 = 2 )
. (79)
4(1 − ν∞ 4(1 − ν∞
Considering aforesaid, we have

1 − νe2 2
1 − ν∞ (1 − 2ν∞ )2 νε ∗ γ 3νσ ∗ γ
= 1+ 2 )
3γ (τε ) + 2 )
3γ (τσ ) . (80)
Ee E∞ 4(1 − ν∞ 4(1 − ν∞
Thus, we obtain two important mutually reciprocal operators (68) and (80) which are
widely used in different applications.
Besides, two useful formulas have been found
tγ1 tγ2 = τεγ τσγ (1 − m1 − m2 ), (81)
tγ1 (1 − m2 ) + tγ2 (1 − m1 ) = (τεγ + τσγ ) (1 − m1 − m2 ). (82)

4.2. The Main Case

In the previous Subsection we have not take the second, or bulk, viscosity into account,
in spite the fact that there are numerous evidence of its existence in scientific literature
[36, 39, 74, 75].
One of the first papers in the field was published by Meshkov and Pachevskaya [39],
wherein the attempt for considering the influence of the bulk relaxation on the internal fric-
tion phenomenon was carried out on the example of longitudinal harmonic vibrations of a
three-dimensional hereditary elastic rod under the conditions of homogeneous deformation.
A fractional exponential function has been used as a hereditary kernel. Investigation of the
frequency dependence of the tangent of the phase shift between the stress and deformation,
i.e., mechanical loss tangent, has revealed two peaks, namely, shear and bulk, in so doing
the peak due to the shear deformation is five times larger than that resulting from the bulk
deformation.
Assume that operators µe and K e are given in the form
γ
e = µ∞ 1 − νε1 3∗γ (τε1 ) ,
µ (83)

Ke = K∞ 1 − νε2 3∗ (τ γ ) , (84)
γ ε2
γ −γ γ −γ
where νε1 = 1 − µ0 µ−1 −1
∞ = 1 − τε1 τσ1 , νε2 = 1 − K0 K∞ = 1 − τε2 τσ2 , µ0 and µ∞ , K0
and K∞ are relaxed and nonrelaxed shear and bulk moduli, respectively, τε1 and τσ1 , τε2
and τσ2 are relaxation and retardation times, the subscript l relates to shear relaxation and
the subscript 2 to bulk relaxation.
Starting from formulas (83) and (84), reciprocal operators have the form
1 γ
= µ−1 ∗
∞ 1 + νσ1 3γ (τσ1 ) , (85)
µ
e
1 −1
γ
= K∞ 1 + νσ2 3∗γ (τσ2 ) , (86)
e
K
γ −γ γ −γ
where νσ1 = µ∞ µ−1 −1
0 − 1 = τσ1 τε1 − 1, and νσ2 = K∞ K0 − 1 = τσ2 τε2 − 1.
First we calculate the operator of compliance
e +µ
1 1 3 K e 1 1 1
Je = = = + . (87)
e
E 3 3K eµ
e 3 µ e 3K e
Substituting (85) and (86) in (87) yields

e 1 νσ1 E∞ ∗ γ 1 νσ2 E∞ ∗ γ
J = J∞ 1 + 3γ (τσ1 ) + 3γ (τσ2 ) ,
3 µ∞ 9 K∞
or
2 γ 1 γ
Je = J∞ 1 + (1 + ν∞ ) νσ1 3∗γ (τσ1 ) + (1 − 2ν∞ ) νσ2 3∗γ (τσ2 ) . (88)
3 3
Let us consider longitudinal harmonic vibrations of a three-dimensional hereditary elas-
tic rod under the conditions of homogeneous deformation when harmonic force acts in-
finitely long. If rod’s axis coincides with the z-axis, then its longitudinal strain is
εzz = p0 J ∗ eiωt , (89)
where p0 and ω are the amplitude and frequency of the external force.
Let us introduce the following operators:
Z t
3∗γ+ (τiγ ) x(t) = 3γ (−s/τi ) x(t − s)ds, (90)
−∞
Z t
γ (t − s)γ−1
I+ x(t) = x(s)ds. (91)
−∞ Γ(γ)
Then according to (13)
∞
X −γ(n+1) γ iωt
3∗γ+ (τiγ ) eiωt = (−1)n τi I+ e , (92)
n=0
but
∞
X ∞
X
−γ(n+1) γ iωt
(−1)n τi I+ e = (−1)n (iωτi)−γ(n+1) eiωt ,
n=0 n=0
and the sum in the right-hand side could be interpreted as an infinite decreasing geometrical
progression with the denominator d = −(iωt)−γ . In other words,
∞
X
iωt (iωτi )−γ eiωt eiωt
e (−1)n (iωτi)−γ(n+1) = = ,
1 − [−(iωτi)−γ ] 1 + (iωτi)γ
n=0
and
1
3∗γ+ (τiγ ) eiωt = eiωt
1 + (iωτi)γ
!
æ−γ
i + cos ψ sin ψ
= γ − i −γ eiωt , (93)
−γ
æi + æi + 2 cos ψ æi + æγi + 2 cos ψ
where æi = ωτi , and ψ = 21 πγ.

Considering (93), the complex compliance J(iω) has the form
(
00 1 æγ + æ−γ γ
ε1 + [1 + (æσ1 /æε1 ) ] cos ψ
J(iω) = J 0 (ω) − iJ (ω) = J∞ 2(1 + ν∞ ) σ1 γ
3 æ−γ
σ1 + æσ1 + 2 cos ψ
)
æγσ2 + æ−γ γ
ε2 + [1 + (æσ2 /æε2 ) ] cos ψ
+ (1 − 2ν∞ ) γ
æ−γ
σ2 + æσ2 + 2 cos ψ
( )
1 [(æσ1 /æε1 )γ − 1] sin ψ [(æσ2 /æε2 )γ − 1] sin ψ
−i J∞ 2(1 + ν∞ ) −γ γ + (1 − 2ν∞ ) −γ γ . (94)
3 æσ1 + æσ1 + 2 cos ψ æσ2 + æσ2 + 2 cos ψ
Now knowing the value J(iω), it is possible to find the tangent of the mechanical loss
angle characterizing the internal friction in a viscoelastic rod
00
J
tan δ = . (95)
J0
Meshkov and Pachevskaya [39] considered a numerical example with ν∞ = 0.3,

µ0 /µ∞ = K0 /K∞ = 0.8, τε1 /τε2 = 103 , and ω = 1 in order to establish whether, in
principle, a relaxation peak due to bulk deformation can occur. The ln æε2 -dependence of
tan δ and phase diagram for the compliance J 00 = f (J 0 ) were presented in [39] in Figures
1a and 1b, respectively, where the values of the fractional parameter γ were indicated by
figures near the corresponding curves. The limiting value γ = 1 in Figure 1a of [39] indi-
cates that the shear and bulk relaxations are described by the standard linear solid model,
and it leads to clear separation of the two peaks. If the bulk and shear moduli have equal de-
grees of relaxation, they make unequal contributions to the total effect defined by Poisson’s
ratio ν∞ , namely: the shear peak being about 5.5 times larger than the bulk one. Reduction
in γ corresponds to broadening of the relaxation spectrum [41], and the difference between
the peaks vanishes. The value γ = 0.5 may be considered in this case to correspond to the
lower limit for the bulk effect, i. e., bulk relaxation is not seen for all γ < 0.5, although it
exists. The possibility of observing bulk relaxation is thus dependent not only on the ratio of
the relaxation times for shear and bulk stresses, but also on the parameter that characterizes
the width of the relaxation spectrum.
Now we calculate the operator
e − 2e
1 3K µ
νe = . (96)
2 3Ke +µe
We will proceed from the operator
γ γ
e +µ
3K e = (3K∞ + µ∞ ) 1 − m1 3∗γ (τε1 ) − m2 3∗γ (τε2 ) , (97)
where m1 = 31 (1 − 2ν∞ )νε1 , and m2 = 23 (1 + ν∞ )νε2 .

Operator inverse to operator (97) has the form
−1
e +µ 1 γ γ
3K e = 1 + n1 3∗γ (tσ1 ) + n2 3∗γ (tσ2 ) , (98)
3K∞ + µ∞
where the values tγσ1 and tγσ2 are defined from the quadratic equation
γ γ γ γ
(1 − m1 − m2 )x2 − [τε1 (1 − m2 ) + τε2 (1 − m1 )] x + τε1 τε2 = 0, (99)
while the values n1 and n2 are determined from the set of two equations with due account
for the magnitudes of tγσ1 and tγσ2 found from Eq. (99)
 τγ γ
 γ ε1 γ n1 + γ τε1 γ n2 = −1,
τε1 −tσ1 τε1 −tσ2
γ γ (100)
 γ τε2 γ n + γ τε2 γ n = −1.
τ −t 1 τ −t 2
ε2 σ1 ε2 σ2
Then from (96) we find

γ γ γ γ
νe = ν∞ 1 − s1 3∗γ (τε1 ) + s2 3∗γ (τε2 ) 1 + n1 3∗γ (tσ1 ) + n2 3∗γ (tσ2 ) , (101)
where
3K∞νε2 2µ∞ νε1
s1 = , s2 = .
3K∞ − 2µ∞ 3K∞ − 2µ∞
Opening the brackets in (101) with due account for (21) yields

s1 n1 tγσ1 s2 n1 tγσ1
νe = ν∞ 1 + n1 + γ − γ 3∗γ (tγσ1 )
τε2 − tγσ1 τε1 − tγσ1

s1 n2 tγσ2 s2 n2 tγσ2
+ n2 + γ − γ 3∗γ (tγσ2 )
γ γ
s1 n1 τε2 s1 n2 τε2 γ
− s1 + γ + γ 3∗γ (τε2 )
γ γ
s2 n1 τε1 s2 n2 τε1 ∗ γ
+ s2 + γ + γ 3γ (τε1 ) . (102)
e it is necessary to rewrite operator Je in the form
In order to find operator E,
γ γ
Je = J∞ 1 + n1 3∗γ (τσ1 ) + n2 3∗γ (τσ2 ) , (103)
where n1 = 32 (1 + ν∞ )νσ1 , and n1 = 13 (1 − 2ν∞ )νσ2 .

Then operator Ee takes the form

Ee = E∞ 1 − m1 3∗γ (tγ ) − m2 3∗γ (tγ ) , (104)
ε1 ε2
where values tγε1 and tγε2

are determined from the following equation:
n1 x n2 x
γ + γ = 1, x = tγε i (i = 1, 2),
τσ1 − x τσ2 −x
or
γ γ γ γ
(1 + n1 + n2 )x2 − [τσ1 (1 + n2 ) + τσ2 (1 + n1 )] x + τσ1 τσ2 = 0, (105)
and values m1 and m2 are defined from the set of equations
 τγ γ
 γ σ1 γ m1 + γτσ1 γ m2 = 1,
τ −t τ −t
σ1 ε1 σ1 ε2
γ
τσ2 γ
τσ2
(106)
 m1 + m2 = 1.
τσ2 −tγε1
γ
τσ2−tγε2
γ
In conclusion of this Subsection we will show how to define the operator proportional
to the operator of cylindrical rigidity. With this purpose we will use the operators
1 e +µ
2 3K e 1 e +µ
2(3K e) 9Keµ
e
= , = , e=
E . (107)
1 + νe 9 Ke 1 − νe e
3K + 4e
µ e
3K + 4e
µ
Then considering (107) we have
Ee Ee Ee 9Keµ
e
= + = µ
e + =µ s−1 .
e+e (108)
1 − νe2 2(1 + νe) 2(1 − νe) e
3K + 4e
µ
First we will calculate the operator
e + 4e
3K µ 1 1 4 1
se = = +
9Keµe 3 µ e 3 K e

3K∞ + 4µ∞ 3K∞ νσ1 γ 3µ∞ νσ2 γ
= 1+ 3∗γ (τσ1 )+ 3∗γ (τσ2 ) .(109)
9K∞ µ∞ 3K∞ + 4µ∞ 3K∞ + 4µ∞
In order to find an operator reverse to operator (109), let us rewrite it in the form
γ γ
s = s∞ 1 + n1 3∗γ (τσ1
e ) + n2 3∗γ (τσ2 ) , (110)
where
3K∞ + 4µ∞ 3K∞ νσ1 3µ∞ νσ2
s∞ = , n1 = , n2 = .
9K∞µ∞ 3K∞ + 4µ∞ 3K∞ + 4µ∞
Then
9Keµ
e γ γ
se−1 = = s−1 ∗ ∗
∞ 1 − m1 3γ (tε1 ) − m2 3γ (tε2 ) , (111)
e + 4e
3K µ
where values tγε1 and tγε2 are determined from Eq. (105), and values m1 and m2 are defined
from the set of Eqs. (106).
Finally, to calculate operator (108), it is necessary to take formulas (83) and (111) into
account. As a result we obtain
e
E µ∞ s∞ + 1 µ∞ s∞ νε1 γ
2
= 1− 3∗γ (τε1 )
1 − νe s∞ µ∞ s∞ + 1

m1 m2
− 3∗γ (tγε1 ) − 3∗γ (tγε2 ) . (112)
µ∞ s∞ + 1 µ∞ s∞ + 1
The reverse operator has the form
1 − νe2 s∞ γ γ γ
= 1 + s1 3∗γ (Tσ1 ) + s2 3∗γ (Tσ2 ) + s3 3∗γ (Tσ3 ) . (113)
e
E µ∞ s∞ + 1
γ γ γ
Values Tσ1 , Tσ2 , and Tσ3 could be found from the following equation:
η1 x η2 x η3 x
γ + γ + γ = −1,
τε1 − x τε2 − x τε3 −x
or
γ γ γ γ γ γ
1 − η1 − η2 − η3 x3 + (τε2 + τε3 ) η1 + (τε3 + τε1 ) η2 + (τε1 + τε2 ) η3
γ γ γ γ γ γ γ γ γ
−τε1 − τε2 − τε3 x2 + τε2 τε3 (1 − η1 ) + τε3 τε1 (1 − η2 ) + τε1 τε2 (1 − η3 ) x
γ γ γ
−τε1 τε2 τε3 = 0, (114)
where
µ∞ s∞ νε1 m1 m2 γ γ γ γ
η1 = , η2 = , η3 = , τε2 = tε1 , τε3 = tε2 ,
µ∞ s∞ + 1 µ∞ s∞ + 1 µ∞ s∞ + 1
and values s1 , s2 , and s3 are determined from following set of equations
 s τγ γ
s2 τε1 γ
s3 τε1

 γ
1 ε1
γ + γ γ + γ γ = −1,

 τε1 −Tσ1 τε1−Tσ2 τε1−Tσ3
 γ γ γ
s1 τε2 s2 τε2 s3 τε2
γ γ + γ γ + γ γ = −1, (115)
 τ −T τ −T τ −T
 ε2
 s τγ
σ1 ε2 σ2 ε2 σ3
 γ γ
 γ 1 ε3γ + γs2 τε3γ + γs3 τε3γ = −1.
τ −T
ε3 σ1 τ −T
ε3 σ2 τ −T
ε3 σ3
Applying the Viéte theorem to Eq. (114), we have

γ γ γ
γ γ γ τε1 τε2 τε3
Tσ1 Tσ2 Tσ3 = ,
1 − η1 − η2 − η3
what completely corresponds to relationship (50) at n = 3.
5. Relationship of Rabotnov Fractional Operators with Other

Fractional Operators
Along with Rabotnov fractional operators there are exist other fractional operators
which are used to solve different problems of viscoelasticity. In this Section, we consider
those fractional operators which are somehow or other connected with Rabotnov operators.
5.1. Rzhanitsyn Operator

Let us consider Boltzmann-Volterra relationships with different kernels of relaxation
Kε and retardation Kσ
Z t
0 0 0
σ = E∞ ε − νε Kε (t − t )ε(t )dt , (116)
0
Z t
0 0 0
ε = J∞ σ + νσ Kσ (t − t )σ(t )dt . (117)
0
As a hereditary kernel we chose Rzhanitsyn kernel [71]

tγ−1 t
Ki = exp − (i = ε, σ), (0 < γ < 1), (118)
Γ(γ)τiγ τi
which could be both the kernel of relaxation and the kernel of retardation, although Rzhan-
itsyn himself used it only as the creep kernel [72].
Putting i = σ in (118) and substituting it in (117), we have
Z t γ−1
t t 0 0
ε(t) = J∞ σ(t) + νσ γ exp − σ(t − t )dt . (119)
0 Γ(γ)τσ τσ
In the Laplace domain, function Kσ has the form
1
K¯σ (p) = γ . (120)
p + τσ−1
Considering (120) from (119) we find

h −γ −γ i
ε̄ = J∞ 1 + νσ p + τσ−1 τσ σ̄, (121)

E∞
σ̄ = −γ ε̄. (122)
1 + νσ p + τσ−1 τσ−γ
In order to reverse from the frequency domain to the time domain in relationship (122),
it is necessary to calculate the integral
Z c+i∞
E∞ dp
K(t) = ept −γ . (123)
2πi c−i∞ 1+ νσ τσ−γ p + τσ−1
Carrying out the substitution
s = p + τσ−1 , ds = dp, c0 = c + τσ−1 ,
let us rewrite integral (123) in the form

Z c0 +i∞
E∞ −t/τσ st 1
K(t) = e e 1− ds, (124)
2πi c0 −i∞ 1 + sγ Tσγ
where Tσγ = τσγ νσ−1 .

Considering that
Z c0 +i∞ Z c0 +i∞
1 st 1 est ds
e ds = δ(t), =3γ (−t/Tσ ) ,
2πi c0 −i∞ 2πi c0 −i∞ 1 + sγ Tσγ
where δ(t) is the Dirac function, we obtain

t t t
K(t) = E∞ exp − δ(t) − exp − 3γ − . (125)
τσ τσ Tσ
Reversing from the frequency domain to the time domain in (122) with due account for
(125), finally we obtain
( Z t 0 0 γ−1 X ∞
)
t (t ) (−1)n νσn (t0 /Tσ )γn 0 0
σ(t) = E∞ ε(t) − νσ exp − ε(t − t )dt .
0 τσ Tσγ n=0 Γ[γ(n + 1)]
(126)
If we reverse from the image to the original in (121) considering the expansion of the
function (1 + pτσ )−γ in a series
γ(γ + 1) γ(γ + 1)(γ + 2)

(1 + pτσ )−γ = 1 − γ pτσ + (pτσ )2 − (pτσ )3 + ...,
2! 3!
then as a result we obtain
" #
d −γ
ε(t) = J∞ 1 + νσ 1 + τσ σ(t). (127)
dt
From (125) it is seen that the resolvent kernel

−t/τσ t
Kε (t) = e 3γ − (128)
Tσ
involves the fractional-exponential function, the attenuation of which with time is strength-
ened by decaying exponent.
If we take the Abel kernel as Kσ kernel
tγ−1
Kσ (t) = γ,
Γ(γ)τσ
which in the Laplace domain has the form
K̄σ (p) = (pτσ )−γ ,
then in this case

1
σ̄ = E∞ 1− ε̄,
1 + pγ Tσγ
and Z
t
t0 0 0
σ(t) = E∞ ε(t) − 3γ − ε(t − t )dt . (129)
0 Tσ
Relationship (129) is equivalent to the Maxwell model, wherein ordinary time-
derivatives are substituted by fractional derivatives.
Although Rzhanitsyn fractional operator (1 + τσ d/dt)−γ entering in (127) does not in-
volve the fractional derivative in the explicit form, as Rabotnov operator [1+τσγ (d/dt)γ )−1 ]
does, however, as it will be shown below on the example of a viscoelastic oscillator, pos-
sesses similar features.
Rzhanitsyn and Rabotnov kernels could be generalized with help of the operator
[1 + τiα (d/dt)α)]−β (0 < α, β < 1), (130)
the features of which have been studied in [59].

From (130) it is evident that the combinational Rabotnov-Rzhanitsyn operator (130)
at β = 1 goes over into the Rabotnov operator, while at α = 1 it transforms into the
Rzhanitsyn operator, respectively.
5.2. Koeller Viscoelastic Models

In 1986 Koeller suggested two viscoelastic models involving fractional derivatives with
several fractional parameters and several relaxation (retardation) times [34]
n
X n
X
iγ
ai D ε = bj D jγ σ, (131)
i=0 j=0
n
Y n
Y
γ
E∞ (D + βi ) ε = (D γ + γj ) σ, (132)
i=1 j=1
where ai , bj and βi, γj (i, j = 1, 2, ..., n) are certain coefficients, 0 < γ < 1 is the fractional
parameter, E∞ = an /bn, and

 Rt
 dmm 1 f (τ )dτ
, m − 1 < iα < m
dt Γ(m−iα) (t−τ )iα+1−m
D iαf (t) = 0

 dm
dtm f (t), iα = m
is the Riemann-Liouville fractional derivative [55].

The model (132) is equivalent to the model (131). In order to show this, it is necessary
to find the roots of equations
n
X n
X
j
bj Z = 0, ai Y i = 0 (133)
j=0 i=0
and suppose that all of them are simple real negative roots Zj = −t−α
j = −γj (j = 1, ..., n)
−α
and Yi = −τi = −βi (i = 1, ..., n), and they are different in magnitude. Knowing the
roots, it is possible to change the sums in (131) by the products.
However, the thermodynamic analysis of the models (131) and (132) was not carried
out, what did not allow Koeller [34] to determine the physical meaning of the involving
parameters, as well as to define the admissible boundaries of their domains.
Such analysis was performed in Rossikhin and Shitikova [62, 63, 65, 67], wherein it
has been shown that the both models (131) and (132) could be reduced to the generalized
Rabotnov model under the appropriate choice of entering constant values. Moreover, only
with such a choice of the constants the both models become to be thermodynamically ad-
missible. Really, the model (132) could be rewritten as
n
Q
(D γ + βi )
i=1
σ = E∞ n ε, (134)
Q
(D γ + γj )
j=1
and the fraction in the right-hand side of (134) could be expanded in simple fractions
n
Q
(D γ + βi ) n
X
i=1 Aj
n = (D γ + β1 )
Q Dγ + γj
(D γ + γj ) j=1
j=1
n
X n
X −1
Aj (γj − β1 ) γ
= Aj − =1− mj 1 + tj D γ , (135)
D γ + γj
j=1 j=1
where
n
Q
(βi − γj )
i=2 γ j − β1
Aj = n , mj = Aj .
Q γj
(γi − γj )
i=1
(i6=j)
n
P
The value Aj = 1, since the operator D γn with the coefficient equal to unit enters
j=1
into the numerator and denominator of the fraction in the left-hand side of formula (135).
Considering formula (10), from (135) we have
n
X Z t
σ = E∞ ε − E∞ mj 3γ −t0 /tj ε t − t0 dt0 , (136)
j=1 0
or  
n
X
σ = E∞ 1 − mj 3∗γ tγj  ε. (137)
j=1
As a result we are led to generalized Rabotnov’s rheological equation with the operator
(23).
Putting ε = ε0 = const in (137), we obtain the relationship
 
 X n 
σ = E∞ 1 − mj [1 − Eγ (− (t/tj )γ )] ε0 , (138)
 
j=1
which describes the stress relaxation, where

∞
X
γ (−1)n (t/tj )γn
Eγ [− (t/tj ) ] =
Γ(1 + γn)
n=0
is Mittag-Leffler function.
Tending t to ∞ in (138) and considering that in so doing Eγ [− (t/tj )γ ] tends to zero,
we have  
Xn
σ0 = E ∞  1 − mj  ε0 , (139)
j=1
or
σ0 = E0 ε0 , (140)
!
n
P
where E0 = E∞ 1− mj is the relaxed elastic modulus.
j=1
If we express ε in terms of σ from relationship (132)
n
Q
(D γ + γj )
j=1
ε= J∞ n σ (141)
Q
(D γ + βi )
i=1
and expand the fraction in the right-hand side of (141) in simple fractions
n
Q
(D γ + γj ) n n
j=1 X Bi X gi
= (D γ + γ1 ) = Bi + , (142)
n
Q D γ + βi 1 + τiγ D γ
(D γ + βi ) i=1 i=1
i=1
where
n
Q
(γj − βi ) n
j=2 γ 1 − βi X
Bi = n , gi = Bi , Bi = 1, J∞ = bn /an ,
Q βi
(βj − βi ) i=1
i=1
(i6=j)
then with due account for (10), we obtain relationship

" n
#
X γ
ε = J∞ 1 + gi 3∗γ (τi ) σ. (143)
i=1
Putting σ = σ0 = const in (143), we find the expression which describes the creep
( n
)
X γ
ε = J∞ 1 + gi [1 − Eγ (− (t/ti ) )] σ0 . (144)
i=1
Tending t to ∞ in (144), we have

n
!
X
ε0 = J∞ 1+ g i σ0 , (145)
i=1
or
ε0 = J0 σ0 , (146)
n

P
where J0 = J∞ 1 + gi is the relaxed compliance.
i=1
Thus, for the model (132) it has been possible to find the thermodynamic constrains on
the rheological parameters by reducing the given model involving fractional derivatives to
the known Rabotnov model, representing itself Boltzmann-Volterra relationships with the
sum of 3∗γ -operators which possess one fractional parameter and several relaxation (retar-
dation) times, namely: the values τi−γ and t−γ j should alternate with each other (32) as it is
shown in Fig. 1.
In other words, in order the models (137) and (143) are to be resolvent, as it has been
shown by Rabotnov [48], it is necessary that the relaxation times τi−γ and the retardation
times t−γ
j should obey the inequalities according to (32).
In conclusion it should be noted that if any two hereditarily elastic operators, K̃ and Ẽ,
or µ̃ and Ẽ, as examples, are given in terms of the Rabotnov-type operators, then it is rather
easy to determine all other operators of the model under consideration, which depend on
the two given viscoelastic operators.
6. Free Vibrations of Oscillators on the Basis of Fractional

Operator Viscoelastic Models
Assume that viscoelastic features of operators are described by the Bolzmann-Volterra
relationships with the relaxation kernel Kε (t) and retardation kernel Kσ (t). Then the equa-
tions of motion of such an oscillator could be written in two different but equivalent forms
[56] Z
t
2
ẍ + ω∞ x − ν K (t − t0 )x(t0 )dt0 = F δ(t), (147)
0
Z t
2
ẍ + ω∞ x + νσ Kσ (t − t0 )ẍ(t0 )dt0 = F [δ(t) + νσ Kσ (t)] , (148)
0
Figure 2. Contour of integration.
where x is the coordinate, F is the amplitude of force impulse per unit mass, ω∞ is the
frequency of elastic vibrations corresponding to the nonrelaxed magnitude of the elastic
modulus, and overdots denote time derivatives.
Applying the Laplace transformation to Eqs. (147) and (148) yields
F F [1 + νσ K̄σ (p)]
x̄(p) = = 2 . (149)
p2 + 2 [1
ω∞ − ν K̄ (p)] ω∞ + p2 [1 + νσ K̄σ (p)]
The solution of (149) in the time domain according to the inversion Mellin-Fourier
formula has the form Z c+i∞
1
x(t) = x̄(p)eptdp. (150)
2πi c−i∞
Functions x̄(p) considered here are multi-valued functions with the branch points p = 0
and p = −∞ or p = −s∗ , s∗ ≥ 0 and p = −∞. In other words, the inversion should be
carried out on the first sheet of Riemannian surface with a cut along the real negative semi-
axis from 0 to −∞ or from −s∗ to −∞. Figure 2 shows the closed contour of integration
for the function with the branch points p = 0 and p = −∞. If a function possesses the
branch points p = −s∗ and p = −∞, then the center of a small circumference should
locate at the point p = −s∗ . Due to the Jordan lemma, the curvilinear integrals taken along
the arcs cR tend to zero at R → ∞. For weakly singular kernels, the integral taken along
cρ also tends to zero when ρ → 0. Besides, the function x̄(p) possesses the ordinary poles
at the same magnitudes of p which vanish the denominator in the formula (149), i.e., they
are the roots of the characteristic equations:

p2 + ω∞2
1 − ν K̄ (p) = 0, (151)
2

ω∞ + p2 1 + νσ K̄σ (p) = 0. (152)
Representing the variable p in the form
p = r eiψ
and introducing notations
p =α [ ]
ρα = <2α [ ] + =2α [ ], tan χα = (α = 1, 2),
<α [ ]

<1 1 − νε K̄ε (p) = <1 [ ] , =1 1 − νε K̄ε (p) = =1 [ ] ,

<2 1 + νσ K̄σ (p) = <2 [ ] , =2 1 + νσ K̄σ (p) = =2 [ ] ,
we obtain
r 2 ρ−1
1 e
i(2ψ−χ1 ) 2
+ ω∞ = 0, (153)
r 2 ρ−1
2 e
i(2ψ+χ2 ) 2
+ ω∞ = 0. (154)
Separating the real and imaginary parts in Eqs. (153) and (154), we find
(
r 2 ρ−1 2
1 cos (2ψ − χ1 ) + ω∞ = 0,
(155)
r 2 ρ−1
1 sin (2ψ − χ1 ) = 0,
(
r 2 ρ−1 2
2 cos (2ψ + χ2 ) + ω∞ = 0,
(156)
r 2 ρ−1
2 sin (2ψ + χ2 ) = 0.
From the sets of Eqs. (153) and (154), we find
(
2ψ − χ1 = ±π,
(157)
r 2 ρ−1 2
1 = ω∞ ,
(
2ψ + χ2 = ±π,
(158)
r 2 ρ−1 2
2 = ω∞ .
From relationships (157) and (158), we find the values of r and ψ which are, respec-
tively, the modulus and argument of the root of characteristic Eqs. (151) and (152).
Using the Jordan lemma and the main theorem of the theory of residues, the solution to
Eqs. (157) and (158) may be written as
Z ∞ X
drift vibr 1
x(t) = x (t)+x (t) = x̄(se−iπ ) − x̄(seiπ ) e−st ds+ res x̄(pk )epk t ,
2πi 0
k
(159)
or
Z ∞ X
1
x(t) = x̄(se−iπ ) − x̄(seiπ ) H(s − s∗ )e−st ds + res x̄(pk )epk t , (160)
2πi 0
k
where H(s−s∗ ) is the Heaviside function, the summation is taken over all isolated singular
points (poles).
Since characteristic equations possess, as a rule, two complex conjugate roots
p1,2 = −α ± iω = r e±iψ , (161)
then relationships (159) and (160) take the form
x(t) = A0 (t) + A exp(−αt) sin(ωt − ϕ), (162)
where Z ∞
A0 (t) = τ −1 B(τ )et/τ dτ, (163)
0
and B(τ ) is the function of distribution of the relaxation parameters (retardation parame-
ters) of the dynamic system.
From Eq. (162) it is seen that the relationship describing vibrations of an oscillator
with the natural frequency ω and the damping factor α possesses two terms, one of which
describes the drift of the equilibrium position and is represented by the integral involving
the distribution function of dynamic and rheological parameters, while the other term is
the product of two time-dependent functions, exponent and sine, and it describes damped
vibrations around the drifting equilibrium position, in so doing the drift is defined by the
function A0 (t). The first term xdrift(t) is governed by an improper integral taken along two
sides of the cut along the negative real semi-axis of the complex plane (see Fig. 2), while
the second term xvibr(t) is determined by two complex conjugate roots of the characteristic
equation, which locate in the left half-plane of the complex plane.
The original method for solving characteristic equation with fractional powers without
its rationalization has been suggested in [40, 89], when the roots and model’s parameters
(relaxation or retardation times) are found at a time. It has been established that the charac-
teristic equation lacks real negative roots and possesses only two complex conjugate roots.
The behavior of the characteristic equation roots in the complex plane depends on the type
of relaxation or creep kernel involved in the model of an oscillator under consideration. The
interested reader is referred to Refs. [55, 60, 64] for a fully comprehensive description of
the mathematical formulation.
Note that 25 years later than it was published in [40, 89], this result was re-discovered
in [7].
Thus, in order to obtain the final solution, it is necessary to specify the type of the
relaxation or creep kernel. Below we consider a few such examples.
6.1. Rabotnov Model

If we use the Rabotnov model as a relaxation kernel (16), then the solution (159) takes
the form of (162) [57], where
1/2
A = 2F (a2 + b2 )−1 (τ−2γ + r 2γ + 2τ−γ r γ cos γψ) , (164)
τ−γ cos β + r γ cos(β − γψ) b ω

tan ϕ = − , β= , r 2 = ω 2 +α2 , tan ψ = − ,
τ−γ sin β + r γ sin(β − γψ) a α
a = (2 + γ)r 1+γ cos(1 + γ)ψ + 2τ−γ r cos ψ + γω∞
2 γ−1
r cos(γ − 1)ψ,
b = (2 + γ)r 1+γ sin(1 + γ)ψ + 2τ−γ r sin ψ + γω∞
2 γ−1
r sin(γ − 1)ψ,
and
sin γπ 2 F [θ (τ )]−1 [θ (τ )]−1 τ 3
ν ω∞ ∞ 0
B(τ, τ−γ ) = , (165)
π θ∞ (τ )[θ0 (τ )] τ τ + θ0 (τ )[θ∞ (τ )]−1 τ γ τγ + 2 cos πγ
−1 −γ −γ
θ∞ (τ ) = τ 2 ω∞
2
+ 1, θ0 (τ ) = τ 2 ω02 + 1,
and ω0 is the frequency of elastic vibrations corresponding to the relaxed magnitude of the
elastic modulus.
The behavior of the roots of the characteristic equation (151) as function of the pa-
rameter τ−γ is shown in Figs. 3 a-d for ω∞ = 1 and four values ξ = 0, 1/50, 1/9
and 1/6, respectively, where figures near curves denote the magnitudes of the value γ, and
ξ = E0 /E∞ = 1 − ν .
Figure 3. Behavior of the complex conjugate roots p1,2 = −α ± iω for an oscillator based
on the Rabotnov model: (a) ξ = 0, (b) ξ = 1/50, (c) ξ = 1/9, and (d) ξ = 1/6.
It is seen that the τ−γ -dependence of the two complex conjugate roots p1,2 = −α ± iω
at γ 6= 1 leave the points ±i and converge in the points ±iξ 1/2 when τ−γ changes from
0 to ∞; in so doing it does not meet the real negative semi-axes and remains inside the
curves for the τ−γ -dependencies of three roots of the characteristic equation with γ = 1
(the ordinary standard linear solid model). The behavior of the two roots of three at γ = 1
(the third root is the real root all the time and changes from 0 to ∞ as τ−γ varies from 0 to
∞) essentially depends on the magnitude of the value ξ. Thus, at the values ξ taken from the
interval (0, 1/9) the two complex conjugate roots first become real as τ−γ changes from 0
−γ
to ∞. But then during further increase in τ , they again become complex conjugate roots,
i.e., the domain of aperiodicity (the values of τ−γ wherein ω = 0) has finite dimensions
contracting with increase in ξ (Figs. 3a,b). At ξ = 0 the ordinary Maxwell model with a
semi-infinite domain of aperiodicity is obtained (Fig. 3a) [57], but at ξ = 1/9 the domain of
aperiodicity degenerates into a point (Fig. 3c). When ξ > 1/9, the domain of aperiodicity
completely disappears (Fig. 3d).
In other words, at γ 6= 1 the behavior character of the roots of the characteristic equa-
tion for the Rabotnov model is governed by the magnitudes of the value ξ, which may be
considered as the deficiency of the elastic modulus.
6.2. Rzhanitsyn Model

It has been already mentioned in Sect. 5.1 that the Rzhanitsyn kernel (118) [71] could
be utilized as the kernel of relaxation and the kernel of retardation, so we will consider these
two cases separately.
6.2.1. Rzhanitsyn After-effect Kernel

Hereditary elastic oscillator model with the Rzhanitsyn creep, or after-effect, kernel
(120) was studied in [56], wherein it had been shown that knowing the behavior of roots of
the characteristic equation (152) and considering that the branch points are s∗ = τσ−1 and
−∞, the solution (159) may be written in the form of (162) with
1/2
A = 2F (h2 + q 2 )−1 (νσ2 + 2Rγ νσ cos γΦ + R2γ ) , (166)
where
νσ sin χ + Rγ sin(γψ + χ) h
tan ϕ = − , χ= ,
νσ cos χ + Rγ cos(γψ + χ) q
R2 = 1 + 2τσ r cos ψ + τσ2 r 2 , tan Φ = τσ r sin ψ(1 + τσ r cos ψ)−1 ,
h = 2rRγ cos(ψ+Φ)+r 2Rγ−1 γτσ cos[2ψ+(γ−1)Φ]+τσω∞
2
γRγ−1 cos(γ−1)Φ+2rνσ cos ψ,
q = 2rRγ sin(ψ+Φ)+r 2 Rγ−1 γτσ sin[2ψ+(γ−1)Φ]+τσω∞
2
γRγ−1 sin(γ−1)Φ+2rνσ sin ψ,
and
2 2 2 −1 3
sin πγ F ω∞ (1 + ω∞ τ ) τ H(τσ − τ )
B(τ, τσ ) = −1
, (167)
π [Dσ (τ )] + Dσ (τ ) + 2 cos πγ
Dσ (τ ) = τ γ (1 + ω∞
2 2 −1
τ ) (τσ − τ )−γ νσ .
The root locus in the complex plane as function of the parameter τσ for three magnitudes
νσ = 49, 8, and 5 (ξ = 1/50, 1/9 and 1/6) at ω∞ = 1 was presented in Figs. 2a-c in [56].
Its analysis reveals that the behavior of the characteristic equation roots in the case under
consideration is identical to that of the oscillator with the Rabotnov relaxation kernel shown
in Figs. 3b-d for the corresponding magnitudes of the parameter ξ.
6.2.2. Rzhanitsyn Relaxation Kernel

Let us take the Rzhanitsyn kernel (118) at i = as the relaxation kernel in the
Boltzmann-Volterra relationships (116). Then the characteristic equation (151) takes the
form
(p2 + ω∞2
)(1 + pτ )γ − ω∞
2
ν = 0. (168)
The behavior of the roots of the characteristic equation (168) in the complex plane as
function of the parameter τ for four magnitudes ν = 0.98, 14/15, 8/9, and 5/6 (ξ =
1/50, 1/15, 1/9 and 1/6) at ω∞ = 1 is presented in Figs. 4(a),(b),(c), and (d), respectively,
wherein the magnitudes of γ are indicated by figures. Reference to Figs. 4 shows that as
τ changes from 0 to ∞, the curves for two complex conjugate roots p1,2 = −α ± iω at
any 0 < γ < 1 issue out of the points ±iξ 1/2 and converge at the points ±i. At ν = 0.98
and 0.12 ≤ γ < 1 (Fig. 4(a)) or ν = 14/15 and 0.47 ≤ γ < 1 (Fig. 4(b)) the domain of
aperiodicity is observed, which narrows with decrease in γ from 1 to 0.12 or from 1 to 0.47
and degenerates into a point at γ = 0.12 or γ = 0.47. This domain disappears completely at
0 ≤ γ < 0.12 or 0 < γ < 0.47, respectively. At ν = 8/9 (Fig. 4(c)) and 5/6 (Fig. 4(d)) the
domain of aperiodicity is entirely absent, although the domain of aperiodicity at ν = 8/9
and γ = 1 exists in the form of a point.
In other words, the structural parameter γ along with the parameter ξ influences not
only the dimensions of the aperiodicity domain, but the existence of this domain as well.
This circumstance profitably distinguishes the model under consideration from the previous
model, since it allows one to use this model for describing dissipative processes of high
intensity.
Now knowing the behavior of roots of the characteristic equation (168) and considering
that the branch points are s∗ = τσ−1 and −∞, the solution (162) is as follows [56]:
for the domains of vibration motions (one real and two complex conjugate roots)
x(t) = A0 (t) + A1 exp(−α1 t) + A2 exp(−α2 t) sin(ωt − ϕ), (169)
or (real root disappears)
x(t) = A0 (t) + A2 exp(−α2 t) sin(ωt − ϕ); (170)
for the domain of aperiodic motions (three real different roots)

3
X
x(t) = A0 (t) + Hi exp(−βi t); (171)
i=1
for the boundaries of the domain of aperiodic motions (one real root, for example, β1 is
the simple root, and the other real root β = β2 = β3 is one repeated root)
x(t) = A0 (t) + H1 exp(−β1 t) + B1 exp(−βt) + B2 t exp(−βt), (172)
or (in the case, if the domain of the aperiodic motions degenerates into a point, the real root
β ∗ = β1 = β2 = β3 becomes a 3-fold root)
x(t) = A0 (t) + B1∗ exp(−β ∗ t) + B2∗ t exp(−β ∗ t) + B3∗ t2 exp(−β ∗ t), (173)
Figure 4. Behavior of the complex conjugate roots p1,2 = −α ± iω for an oscillator based
on the fractional calculus model with the Rzhanitsyn relaxation kernel: (a) ξ = 1/50, (b)
ξ = 1/15, (c) ξ = 1/9, and (d) ξ = 1/6.
where α1 , βi (i = 1, 2, 3), β and β ∗ are the real roots of Eq. (168) which are located
1/γ
between τ−1 (1 − ν ) and the branch point τ−1 , and α2 ± iω are the complex conjugate
roots of Eq. (168).
The amplitudes Ai , Hi , Bi∗ (i = 1, 2, 3), B1 , and B2 , as well as ϕ are expressed in
terms of the damping coefficients α1 , α2 , βi , β, β ∗ and natural frequency ω as follows:
F (1 − α1 τ ) F (1 − βi τ )
A1 = , Hi == ,
α21 τ (2 + γ) − 2α1 + γτω∞
2 βi2 τ (2 + γ) − 2βi + γτω∞
2
A2 = 2F Rγ (h2 + q 2 )−1/2 , ϕ = −(γψ + χ), χ = h/q,
h = 2rRγ cos(ψ + γΦ) + γr 2 Rγ−1 τ cos[2ψ + (γ − 1)Φ] + γω∞

2
Rγ−1 τ cos(γ − 1)Φ,
R2 = 1 + 2τ r cos ψ + τ2 r 2 , tan Φ = τ r sin ψ(1 + τ r cos ψ)−1 ,
q = 2rRγ sin(ψ + γΦ) + γr 2Rγ−1 τ sin[2ψ + (γ − 1)Φ] + γω∞

2
Rγ−1 τ sin(γ − 1)Φ,
B1 = 2F γ(1 − βτ )τ l −1 , B2 = 2F (1 − βτ )2 l −1 ,
l = β 2 τ2 (2 + γ)(1 + γ) − 4βτ (1 + γ) + γ(γ − 1)τ2 ω∞

2
,
B1∗ = 3F γ(γ −1)(1−βτ)τ2 l1−1 , B2∗ = 6F γ(1−βτ)2 τ l1−1 , B3∗ = 3F (1−βτ)3 l1−1 ,
l1 = β 2 τ3 (2 + γ)(1 + γ)γ − 6βτ2 (1 + γ)γ + 4τ (1 + γ) + γ(γ − 1)(γ − 2)τ2 ω∞

2
.
The value A0 (t) is determined by Eq. (162), wherein the distribution function B(τ, τ)
of the relaxation and creep parameters of the dynamical system under consideration has the
form
sin πγ F (1 + ω∞ 2 τ 2 )−1 τ 3 H(τ − τ )

B(τ, τ ) = , (174)
π [D (τ )] + D (τ )τ + 2τ 2 cos πγ
−1 4
D (τ ) = τ γ (1 + ω∞
2 2 −1 2
τ ) (τ − τ )−γ ω∞ ν .
Comparison studies of damped vibrations of the hereditary elastic oscillators, whose

hereditary properties are described by the fractional calculus models with the weakly sin-
gular Rzhanitsyn kernel, allow us to make the following conclusions: (1) the peculiarity of
the vibrational process of the viscoelastic single-mass systems, which are modeled by the
fractional calculus models with the weakly singular Rzhanitsyn kernel as the creep kernel,
resides in impossibility of the transition from the vibrating motions to the aperiodic regime
in spite of the known fact that under the sufficiently large intensity of dissipative processes
real vibrating systems may experience the aperiodic regime; (2) the more complicated rhe-
ological model, i.e., the model with the weakly singular Rzhanitsyn kernel as the relaxation
kernel, utilized for describing the viscoelastic properties of a single-mass system, allows
one to trace the influence of the fractional operator parameter γ on the dynamic character-
istics of the system not only in the region of vibration, but in the domain of the aperiodic
motions as well. Moreover, it has been shown that the occurrence or vanishing of the region
of the aperiodic motions for the model put forward is governed not only by the magnitudes
of the value ν , but by the magnitudes of the value γ as well.
6.3. Combinational Rabotnov-Rzhanitsyn Model

The comparison between theoretical and experimental results in the field of applica-
tion of fractional calculus models in viscoelasticity carried out in [59] has shown that the
fractional models with more than one fractional parameters seem to be very flexible for
describing relaxation (retardation) behavior over wide frequency ranges and for different
materials. One distinct advantage of the fractional calculus models is their ability to de-
scribe real material behavior using only a small number of parameters.
However, the analysis of many of such models [59] revealed the fact that at some mag-
nitudes of rheological parameters they are not thermodynamically admissible, i.e., the loss
tangent takes on negative values in some frequency domain and the relaxation (retardation)
function becomes non-monotonic in some time intervals.
It has been proved in [59] that the rheological models involving fractional operators
(130) with two independent fractional parameters 0 < α, β ≤ 1, which represent the com-
bination of Rabotnov and Rzhanitsyn models
σ = E∞ [ε − νε (1 + τεα D α )−β ε], (175)
ε = J∞ [σ + νσ (1 + τσα D α)−β σ] (176)

are thermodynamically well-conditioned at all magnitudes of the rheological parameters.
Note that the models (175) and (176) at 0 < α, β < 1 are not resolvent despite of the
Rabotnov models at 0 < α < 1 and β = 1.
Havriliak and Negami [26, 27] used these models for the analysis of dielectric and me-
chanical dispersion for five polymers (polycarbonate and polyisophthalat on the basis of
bisphenol A, isotaclic polymethyl methacrylate, polymethylacrylate, and co-polymers of
phenylmethacrylate and acrylonitrile), while Hartmann et al. [25] utilized them for ap-
proximation of the loss factor for polymer relaxation on the basis of experimental data for
different polymers. It has been shown close agreement between theoretical and experimen-
tal findings.
6.3.1. Rabotnov-Rzhanitsyn Relaxation Kernel

Now let us first consider viscoelastic oscillators (147), the behavior of which is de-
scribed by the model (175), following Rossikhin and Shitikova [59]. For this case, Eq. (149)
takes the form
F [1 + (pτε )α ]β
x̄(p) = . (177)
2 ) [1 + (pτ )α ]β − ω 2 ν
(p2 + ω∞ ε ∞ ε
To determine the poles of the function x̄(p) (177), it is necessary to find the roots of the
characteristic equation
(p2 + ω∞
2
) [1 + (pτε )α]β − ω∞
2
νε = 0. (178)
The procedure of solving characteristic Eq. (178) is described in detail in [59].

2 =
The behavior of the roots in the complex plane as function of the parameter τε at ω∞
1 for three magnitudes of νε = 0.98, 8/9, 5/6 and β = 0.98 is presented in Figures
5 (a)-(c), respectively, where figures near curves denote the magnitudes of the fractional
Figure 5. Behavior of the complex conjugate roots p1,2 = −α±iω for an oscillator based on
the fractional calculus model with the Rabotnov-Rzhanitsyn relaxation kernel at ω∞ = 1:
(a) νε = 0.98, (b) ξ = 8/9, and (c) ξ = 5/6 when β = 0.98.
parameter α. From Figure 5 it is seen that the curves describing the roots behavior do not
cut the real axis, i.e., the fundamental possibility for the transition of the vibrational process
to an aperiodic regime is absent. Note that at α = 1 and fractional β this model, as it has
been discussed above in Sec. 6.2.2, behaves similarly to the standard linear solid model with
derivatives of an integer order, that is, the system may be both in vibrational and aperiodic
regimes depending on the magnitudes of the parameter τε .
Knowing the behavior of the characteristic equation roots, the solution (159) takes the
form of (162) [55], where
A = 2F Rβε (h2 + q 2 )−1/2 , (179)
h
tan ϕ = − tan(βΦε + χ), tan χ = ,
q
p r ατα sin αψ
R = 1 + 2r α τα cos αψ + r 2α τ2α, tan Φ = ,
1 + r α τα cos αψ
h = 2rRβε cos(ψ + βΦε ) + r 1+α αβτεα Rβ−1
ε cos[(β − 1)Φε + (α + 1)ψ]
2 α−1 α β−1
+ω∞ r αβτε Rε cos[(β − 1)Φε + (α − 1)ψ],
q = 2rRβε sin(ψ + βΦε ) + r 1+α αβτεα Rβ−1

ε sin[(β − 1)Φε + (α + 1)ψ]
2 α−1 α β−1
+ω∞ r αβτε Rε sin[(β − 1)Φε + (α − 1)ψ],
F π −1 s−1 (s2 + ω∞
2 )−1 sin βΦ ∗
ε
B(s) = ,
2 )R∗β ν −1 ω −2 + (s2 + ω 2 )−1 R∗−β ν ω 2 − 2 cos βΦ∗
(s2 + ω∞ ε ε ∞ ∞ ε ε ∞ ε
R∗ = R |ψ=π , Φ∗ = Φ |ψ=π .
The character of the solution x(t) is presented in Figure 6, where figures near curves
denote the magnitude of the fractional parameter α.
6.3.2. Rabotnov-Rzhanitsyn Retardation Kernel

Now let us consider viscoelastic oscillators (148), the behavior of which is described by
the model (176), following Rossikhin and Shitikova [59]. For this case, Eq. (149) takes the
form
F [1 + (pτσ )α]β + νσ
x̄(p) = 2 2 )[1 + (pτ )α ]β + p2 ν
. (180)
(p + ω∞ σ σ
To determine the poles of the function x̄(p) (180), it is necessary to find the roots of the
characteristic equation
(p2 + ω∞
2
) [1 + (pτσ )α ]β + p2 νσ = 0. (181)
The behavior of the roots in the complex plane as function of the parameter τσ is pre-
sented in Figs. 7 for three magnitudes of νσ = 49, 8 and 5, respectively, when β = 0.98.
Knowing the behavior of the characteristic equation roots, the solution (159) takes the
form of (162) [55], where
q
A = 2F (h2 + q 2 )−1 (R2β β
σ + 2Rσ νσ cos βΦσ + νσ ),
2 (182)
νσ sin χ + Rβσ sin(βΦσ + χ) h

tan ϕ = − , tan χ = ,
νσ cos χ + Rβσ cos(βΦσ + χ) q
p r α τσα sin αψ
Rσ = 1 + 2r ατσα cos αψ + r 2ατσ2α , tan Φσ = ,
1 + r ατσα cos αψ
h = 2rRβσ cos(ψ + βΦσ ) + r 1+α αβτσα Rβ−1
σ cos[(β − 1)Φσ + (α + 1)ψ]
2 α α−1 β−1
+ω∞ αβτσ r Rσ cos[(β − 1)Φσ + (α − 1)ψ] + 2rνσ cos ψ,
q = 2rRβσ sin(ψ + βΦσ ) + r 1+α αβτσα Rβ−1

σ sin[(β − 1)Φσ + (α + 1)ψ]
2 α α−1 β−1
+ω∞ αβτσ r Rσ sin[(β − 1)Φσ + (α − 1)ψ] + 2rνσ sin ψ,
2 −3 2 2 −1
F π −1 ω∞ s (s + ω∞ ) sin βΦσ∗
B(s) = ,
2 )R∗β ν −1 s−2 + (s2 + ω 2 )−1 R∗−β ν s2 + 2 cos βΦ ∗
(s2 + ω∞ σ σ ∞ σ σ σ
R∗σ = Rσ |ψ=π , Φ∗σ = Φσ |ψ=π .

The character of the solution x(t) is presented in Figure 8, where figures near curves
denote the magnitude of the fractional parameter α.
Figure 6. Character of the solution x(t) behavior for an oscillator based on the fractional
calculus model with the Rabotnov-Rzhanitsyn relaxation kernel at ω∞ = 1: (a) νε = 0.98,
(b) νε = 8/9, and (c) νε = 5/6 for β = 0.98 and τε = 1.
As it has been noted in Sec 6.2.1, when α = 1 and β = γ, the model under consideration
goes over into the oscillator model with the Rzhanitsyn retardation operator, for which the
behavior of the characteristic equation roots resembles the character of the roots behavior
for the generalized standard linear solid model with the only difference that in this case the
noticeable asymmetry is observed in the behavior of the roots.
Figure 7. Behavior of the complex conjugate roots p1,2 = −α±iω for an oscillator based on
the fractional calculus model with the Rabotnov-Rzhanitsyn retardation kernel at ω∞ = 1:
(a) νσ = 49, (b) νσ = 8, and (c) νσ = 5 when β = 0.98.
6.4. Generalized Rabotnov Model

The equation of motion of an oscillator, hereditary features of which are described by
the model (137) has the form [63]:
 
Xn
2 
ẍ + ω∞ 1− mj 3∗γ tγj  x = F δ(t). (183)
j=1
Applying the Laplace transformation to Eq. (183) yields
F
x̄(p) = Pn . (184)
p2 2 {1 −
+ ω∞ j=1 mj [1 + (ptj )γ ]−1 }
Figure 8. Character of the solution x(t) behavior for an oscillator based on the fractional
calculus model with the Rabotnov-Rzhanitsyn retardation kernel at ω∞ = 1: (a) νσ = 49,
(b) νσ = 8, and (c) νσ = 5 for β = 0.98 and τσ = 1.
The solution in the time domain is determined by the Mellin-Fourier inversion formula
(150). To calculate the integral in (150), it is necessary to determine all singular points of
the complex function (184). This function has the branch points p = 0 and p = ∞ and
simple poles at the same magnitudes of p which vanish the denominator of formula (184),
i.e., they are the roots of the characteristic equation
 
 Xn 
f (p) = p2 + ω∞ 2
1− mj [1 + (ptj )γ ]−1 = 0. (185)
 
j=1
Since for the multivalued functions possessing branch points the inversion formula is
valid only for the first sheet of the Riemann surface, then for calculating integral (150) we
shall use the closed contour presented in Fig. 2. Using the main theorem of the residue
theory, it could be written in the form of (159).
The value xdrift(t) in (159) is calculated immediately and has the form [63]
Z Pn
drift F ω∞2 ∞ mj R−1
j sin Φj e
j=1
−st ds
x (t) = Pn −1 Pn −1 ,
π 0 (s2 + ω∞
2 − ω2
∞
2 4
j=1 mj Rj cos Φj ) + ω∞ ( j=1 mj Rj sin Φj )
2
(186)
where
q
(stj )γ sin γπ
Rj = 1 + 2(stj )γ cos γπ + (stj )2γ , tan Φj = .
1 + (stj )γ cos γπ
In order to calculate xvibr(t), it is necessary to investigate the roots of the characteristic

equation (185). It has been proved in [63] that (185) lacks real roots and possesses for
each fixed magnitude of tj (j = 1, ..., n) only two complex conjugate roots p1,2 = r iψ =
−δ ± iω. For example, in the case of n = 2, the character of the root locus behavior as
function of Z1 (t1 ) = (rt1 )γ at fixed magnitudes of Z2 (t1 ) = (rt2 )γ is shown in Figs. 9 for
ξ = 1 − (m1 + m2 ) = 1/9 at m1 = m2 = 4/9. The values of the fractional parameter γ
are indicated by digits near the corresponding curves. In all Figs. 9 a-f only one from each
pair of complex conjugate roots is shown in the upper quadrant of the negative half-plane
of the complex plane.
Knowing the behavior of the characteristic equation roots, the function xvibr(t) could
be represented as
xvibr(t) = Ae−δt cos(ωt + ϕ), (187)
where
2F =f 0 reiψ
A= q , tan ϕ = ,
2 2 <f 0 (reiψ )
[<f 0 (reiψ )] + [=f 0 (reiψ )]
X n
f (p) 2
= f 0 (p) = 2p + ω∞ mj [1 + (ptj )γ ]−2 γpγ−1 ,
dp
j=1
n
X
<f 0 reiψ = 2r cos ψ + ω∞
2
mj tγj γr γ−1 R̄−2
j cos 2Φ̄j + (1 − γ)ψ ,
j=1
Figure 9. Behavior of the complex conjugate roots p1,2 = −δ ±iω for an oscillator based on
the fractional calculus model with the generalized Rabotnov relaxation kernel for ξ = 1/9
at m1 = m2 = 4/9 and ω∞ 2 = 1.
n
X
=f 0 reiψ = 2r sin ψ − ω∞
2
mj tγj γr γ−1 R̄−2
j sin 2Φ̄j + (1 − γ)ψ ,
j=1
q
(rtj )γ sin γψ
R̄j = 1 + 2(rtj )γ cos γψ + (rtj )2γ , tan Φ̄j = .
1 + (rtj )γ cos γψ
Reference to Figs. 9 shows that under the presence of two relaxation times the curves
on the complex plane, which characterize the relaxation time dependence of two complex
conjugate roots of the characteristic equation, at the fixed magnitude of t2 (0 < t2 < ∞)
are detached from the imaginary axis, so that even for the magnitudes t1 = 0 (Z1 = 0) or
t1 = ∞ (Z1 = ∞) the vibratory process remains dissipative one. In other words, if one
of the relaxation mechanisms ceases to work or works under a degenerate regime, then the
other mechanism begins to dominate.
However, when t2 = 0 (Z2 = 0) and t2 = ∞ (Z2 = ∞), the curves are attached to the
imaginary axis at their initial and terminal points (see the cases (a) and (f) in Fig. 9), since in
these cases the limiting values of the characteristic equation roots have the following form:
when t2 = 0 (Z2 = 0)
π
at t1 → 0 (Z1 → 0) : r02 = ω∞
2
(1 − m1 − m2 ), ψ0 = ± ,
2
2 2 π
at t1 → ∞ (Z1 → ∞) : r∞ = ω∞ (1 − m2 ), ψ∞ = ± ,
2
while t2 = ∞ (Z2 = ∞)
π
at t1 → 0 (Z1 → 0) : r02 = ω∞
2
(1 − m1 ), ψ0 = ± ,
2
2 2 π
at t1 → ∞ (Z1 → ∞) : r∞ = ω∞ , ψ∞ = ± .
2
Thus, if both relaxation processes go on very quickly or very slowly, then the given
hereditary elastic model behaves itself as the elastic one.
P
In a particular case, when ω0 = 0 or nj=1 mj = 1, the curves in the complex plane
π
issue from zero, (what is characteristic of the Maxwell-like model) at the angles ψ = ± 2−δ ,
when all Zj → 0, and arrive at the points ±iω∞ , when all Zj → ∞. For the case of n = 2,
this is evident in Figs. 10a and 10f, respectively.
Note that the procedure proposed for the construction and the analysis of the char-
acteristic equation roots’ locus can be utilized for an arbitrary number n of terms in the
governing equation (183) , though it was demonstrated for n = 2 in the examples provided
above, while the example of retaining three terms in (185) can be found in [68].
7. Stationary Shock Waves in Viscoelastic Media

with the Rabotnov Operator
For constructing the evolutionary equations describing the behavior of nonlinear waves
in one-dimensional media three methods of simplification - the iterative, the spectral and
the asymptotic - are used [18, 31, 44]. The ray method [52] which allows one to derive
evolutionary equations both for weak and strong (shock) waves is found to be the most
Figure 10. Behavior of the complex conjugate roots p1,2 = −δ ± iω for an oscillator based
on the fractional calculus model with the generalized Rabotnov relaxation kernel for ξ = 0
2 = 1.
at ω∞
efficient among the asymptotic methods. However, in some nonlinear media, shock waves
do not propagate in the form of strong discontinuity surfaces but exist only in the form of a
shock layer of finite thickness. Among such media are, as an example, linear and nonlinear
hereditary elastic media with weakly singular kernels of heredity [10, 24]. It turns out that
for such media the application of the ray method within the shock layer allows one to obtain
simplified equations describing evolution of shock waves of finite thickness, which admit
the solution in the form of a stationary shock wave.
Let us consider, following [53], a nonlinear hereditarily elastic half-space. Equations
describing the motion of such a half-space in Eulerian description in the rectangular Carte-
sian coordinate system x1 , x2 , x3 have the form
Z t
∂u ∂u 2 ∂u
σ=k +α −γ K(t − t0 ) (t0 )dt0 , (188)
∂x ∂x 0 ∂x
2
∂σ ∂ u ∂u ∂u2
= ρ0 +2 , (189)
∂x ∂t2 ∂x ∂t∂x
where σ = σ11 is the stress, u = u1 is the nonzero component of the displacement vector,
k = λ + 2µ, λ and µ are the equilibrium Lame constants, α = 3(l + m + n) − 7k/2, l,
m and n are the third-order elastic moduli, t is the time, x = x1 is the coordinate measured
along the normal to the boundary x = 0 of the hereditarily elastic half-space x > 0, K(t) is
the kernel of heredity, ρ0 is the density, and γ is a small parameter taking into consideration
the effect of viscosity.
Let beginning from the moment t = 0 the boundary x = 0 of the half-space x > 0 is
loaded so that its initial velocity and acceleration are equal to v0 and a0 , respectively.
As it has been shown in [10, 51], impact loading of the hereditarily elastic half-space
boundary gives rise to the shock wave in the half-space, i.e., the geometric surface on which
stresses and strains have a discontinuity, in so doing the shock wave velocity is as follows
G = c(1 + bκ1 ), κ1 = v0 {1 − (a0 b + β)t} , (190)

∂u +

∂u −
where κ1 = ∂u ∂t = ∂t − ∂t is the discontinuity of the value ∂u 2 −1
∂t , c = kρ0 ,
b = (α − k)k−1 (2c)−1 , the signs “ + ” and “ − ” denote that the value ∂u ∂t is calculated
immediately ahead of and behind the wave front, respectively, and β = K(0)(2k)−1.
Reference to Eq. (190) shows that at β ≤ −ba0 (b < 0) the value κ1 increases with time,
but when β > −ba0 , the value κ1 is damped out to zero in the finite time lapse between 0
and t∗ , where t∗ = (β + ba0 )−1 . If the kernel of heredity possesses weak singularity, i.e.,
β → ∞ [48], then damping of the discontinuity κ1 takes place in an infinitely short time
interval, and in that case G = c.
Thus, in the nonlinear hereditarily elastic media with weakly singular kernels of hered-
ity, the shock waves cannot propagate in the form of geometrical surfaces of strong discon-
tinuity.
Suppose that in such a medium the shock layer of thickness h propagates with the ve-
∂u
locity G = c, in so doing within the layer the functions σ, ∂x and ∂u
∂t change monotonically
∂u
+ ∂u
+
and uninterruptedly from the magnitudes σ + , ∂x and ∂t to the magnitudes σ − ,
∂u
− −
∂x and ∂u∂t .
To construct the solution within the shock layer, let us eliminate the stress σ from
Eqs. (188) and (189). As a result, we obtain

∂ 2u ∂u ∂ 2 u ∂ 2Φ ∂ 2u ∂u ∂ 2 u
+ 2κ − γ 0 = c−2 + 2 , (191)
∂x2 ∂x ∂x2 ∂x2 ∂t2 ∂t ∂t∂x
Rt
where γ0 = γk−1 , Φ = 0 K(t − t0 )u(t0 )dt0 , K(0) = ∞, and κ = αk−1 .
Using the conditions of compatibility which link derivatives in the two coordinate sys-
tems: the stationary coordinate system and the moving coordinate system traveling together
with the layer
∂u ∂u δu ∂u ∂u ∂ 2u ∂2u
= −c + , = , = ,
∂t ∂n δt ∂x ∂n ∂x2 ∂n2

∂ 2u ∂ 2u δ ∂u ∂ 2u 2
2 ∂ u δ ∂u δ2 u
= −c 2 + , = c − 2c + , (192)
∂x∂t ∂n δt ∂n ∂t2 ∂n2 δt ∂n δt2
where ∂/∂n is the derivative with respect to the normal to the wave layer, and δ/δt is the
Thomas δ-derivative [81], from Eq. (191) we obtain

∂ 2u ∂u ∂ 2 u ∂2Φ −2
2
2∂ u δ ∂u δ2 u
+ 2κ − γ 0 = c c − 2c +
∂n2 ∂n ∂n2 ∂n2 ∂n2 δt ∂n δt2

∂u δu ∂ 2u δ ∂u
+2 −c + −c 2 + . (193)
∂n δt ∂n δt ∂n
Let us reduce Eq. (193) to the dimensionless form introducing the following dimen-
−3/2 1/2
sionless values: w = v0−2 a0 u, θ = v0−1 a0 t, z = v0 c−1/2 a0 n, and = v0 c−1/2 . As a
result, we have

∂w ∂ 2 w ∂ 2 Φ∗ δ ∂w 2
2 δ w ∂w ∂ 2 w
2κ − ν = −2 + + 2
∂z ∂z 2 ∂z 2 δθ ∂z δθ2 ∂z ∂z 2
2

2 ∂w δ ∂w 2 δw ∂ w 3 δw δ ∂w
−2 − 2 + 2 , (194)
∂z δθ ∂z δθ ∂z 2 δθ δθ ∂z
Rθ
where Φ∗ = 0 K(θ − θ0 )w(θ0 )dθ0 , ν = γ0 τ , and τ = v0 a−1
0 .
Considering that is a small value, we seek the solution of Eq. (194) in the form
w = w0 + w1 + 2 w2 + . . . . (195)
Substituting (195) into Eq. (194) and limiting ourselves by the zeroth term of the series
(195) yields
Z θ
∂w0 ∂ 2 w0 δ ∂w0 ∂2
∗
(κ − 1) + =ν K(θ − θ0 )w0 (θ0 )dθ0 , (196)
∂z ∂z 2 δθ ∂z ∂z 2 0
where ν ∗ = ν(2)−1 .
Carrying out the change of the variable v = (κ − 1)∂w0 /∂z in Eq. (196) yields
Z θ
∂v δv ∗ ∂
v + =ν K(θ − θ0 )v(θ0 )dθ0 . (197)
∂z δθ ∂z 0
Equation (197) describes variations in the value v within the shock layer which moves
with the constant speed c. It could be rewritten in the form
Z θ
1 ∂v 2 δv ∂
+ = ν∗ K(θ − θ0 )v(θ0 )dθ0 . (198)
2 ∂z δθ ∂z 0
Assuming that the function v depends only on the variable η = θ − zG−1 , and intro-
ducing the change of variables v = 2Φ(η), we are led to the equation for determining the
structural shock wave
Z η
∂Φ2 ∂Φ ∗ ∂
− +G = −ν K(η − η 0)Φ(η 0)dη 0 , (199)
∂η ∂η ∂η 0
where G is a constant.
Integrating Eq. (199) yields
Z η
2 0 0 0
Φ (η) = G Φ(η) + e1 K(η − η )Φ(η )dη , (200)
0
where e1 = ν ∗ G−1 .
At η = 0, we have
Φ20 = GΦ0 ,
Φ0 = 0 or Φ0 = G, (201)
where Φ0 = Φ(0).
From two values of Φ0 = Φ(0) governed by (201) Φ0 = 0 should be chosen, since
discontinuities could not propagate in hereditarily elastic media with a singular kernel of
heredity [55].
In order to find the asymptotic magnitude of the desired function Φ at η → ∞, we
will consider Volterra operators K ∗ with a positive kernel K(t), which not only obeys the
condition of fading memory K(∞) = 0, but also satisfies the condition of boundedness of
the integral Z ∞
lim (K ∗ · 1) = K(t)dt = K∞ . (202)
t→∞ 0
We will name such integrals as bounded integrals.
Let us show that
Z ∞
∗ γ
lim 3γ (τi ) · 1 = 3γ (−t/τi ) dt = 1. (203)
t→∞ 0
Really, Laplace transformation of the product 3∗γ (τiγ ) · 1 has the form
τi−γ
3∗γ (τiγ ) · 1 = . (204)
p(pγ + τi−γ )
Applying the limiting theorem of the Laplace transform

h i τi−γ
lim p 3∗γ (τiγ ) · 1 = lim 3∗γ (τiγ ) · 1 = lim = 1, (205)
p→0 t→∞ p→0 p + τi−γ
results in relationship (203).

Further we will introduce a class of functions G such that if the function g(t) ∈ G, then
at any choice of the positive value ζ there exists such a number g∞ , and it is possible to
indicate such a magnitude t = t∗ that at t > t∗ the inequality
|g(t) − g∞ | < ζ (206)
is valid.
Now we will prove the following two theorems stated by Rabotnov [48].
Theorem 1. Let two bounded operators L∗ and M ∗ are specified, and their product
L∗ · M ∗ is defined, then at t → ∞ the following limiting equality is fulfilled:
(L∗ · M ∗ ) · 1 = (L∗ · 1) (M ∗ · 1) (t → ∞). (207)
To prove the theorem, let us represent the kernels of the operators L and M via Laplace
integrals Z Z
∞ ∞
L(t) = L̄(p)e−pt dp, M (t) = M̄ (q)e−qt dq. (208)
0 0
Integrating (208) over t from the left and from the right, we obtain
Z ∞ Z ∞
∗ L̄(p) −pt
∗ M̄ (q)
L ·1 = 1−e dp, M ·1= 1 − e−qt dq. (209)
0 p 0 q
Tending t to ∞ in formulas (209), we have

Z ∞ Z ∞
∗ L̄(p) ∗ M̄ (q)
L ·1= dp, M ·1= dq. (210)
0 p 0 q
Now we could calculate the kernel of the product of operators L∗ · M ∗ , which, as a rule,
has the form [48]:
Z t0
N (t − τ ) = L(t − s)τ M (s − τ )ds. (211)
τ
Considering (208), formula (211) could be written as follows:
Z tZ ∞Z ∞
N (t − τ ) = L̄(p)M̄(q)e−p(t−s)e−q(s−τ ) dpdqds. (212)
τ 0 0
Changing the order of integration in (212), i.e., integrating first over t, we reduce (212)
to the form
Z ∞Z ∞
1 −q(t−τ )
N (t − τ ) = L̄(p)M̄(q) e − e−p(t−τ ) dpdq. (213)
0 0 p−q
Putting τ = 0 in (213) and integrating the obtained relationship in terms of t, we find

Z t Z Z
∞ ∞
1 1 1
(L∗ · M ∗ ) · 1 = N (t)dt = L̄(p)M̄ (q) 1−e−qt − 1−e−pt dpdq.
0 0 0 p−q q p
(214)
Tending t to ∞ in (214) and considering (210), we obtain

Z ∞Z ∞ Z ∞ Z ∞
∗ ∗ L̄(p)M̄(q) L̄(p) M̄ (q)
lim (L · M ) · 1 = dpdq = dp dq
t→∞ 0 0 pq 0 p 0 q
= lim (L∗ · 1) · lim (M ∗ · 1) . (215)
t→∞ t→∞
Relationship (215) coincides with (207) what proves the stated theorem.
Theorem 2. If operator K ∗ is bounded and g(t) ∈ G, then at t → ∞ the following

equality is fulfilled:
K ∗ g = K∞ g∞ . (216)
To prove the theorem, we put
L∗ = K ∗ , M (t) = g 0 (t). (217)
Then the kernel of the operator L∗ · M ∗ , according to (211) and (217), takes the form
Z t
N (t − τ ) = K(t − s)g 0(s − τ )ds, (218)
τ
while relationship (L∗ · M ∗ ) · 1 with due account for (218) could be written as
Z t Z tZ t
(L∗ · M ∗ ) · 1 = N (t)dt = K(t − s)g 0 (s)dsdt. (219)
0 0 0
Making the change
t − s = τ, s = t − τ, ds = −dτ
in the double integral, we rewrite (219) as

Z tZ t Z t
(L∗ · M ∗ ) · 1 = K(τ )g 0(t − τ )dτdt = K(τ ) [g(t − τ ) − g(−τ )] dτ
0 0 0
Z t Z t
= K(τ )g(t − τ )dτ − g(0) K(τ )dτ. (220)
0 0
Since the function g(t) for negative magnitudes of the argument was determined, then
we have prolonged it in the negative domain putting g(−t) = g(0).
Using Theorem 1 and considering that
Z t
∗
L ·1= K(τ )dτ, M ∗ · 1 = g(t) − g(0),
0
as well as relationship (220), we have

Z t Z t Z t
lim K(t − τ )g(τ )dτ − g(0) K(τ )dτ = lim (g(t) − g(0)) lim K(τ )dτ,
t→∞ 0 0 t→∞ t→∞ 0
whence it follows equality (216), hence Theorem 2 has been proved.

Using formula (216) and tending the value η → ∞ in Eq. (200), we obtain
Φ∞ = G(1 + e1 K∞ ), (221)
where Φ∞ = Φ(∞).
If as a relaxation kernel we use the fractional-exponential function, then K∞ = 1, and
formula (221) takes the form
Φ∞ = G(1 + e1 ). (222)
Thus, formulas (201) and (222) define the limiting locations of the stationary shock
wave profile.
8. Application of Rabotnov’s Operators in Problems of Impact

Response of Thin Structures
The approach existing for studying the impact response of viscoelastic engineering
structures is considered below in this Section by the example of the dynamic response of
a viscoelastic Bernoulli-Euler beam, which Young’s modulus is the time-dependent opera-
tor and the bulk modulus is considered to be constant, transversely impacted by an elastic
sphere [67].
Following Timoshenko [80] who considered the problem on a transverse impact of an
elastic sphere upon an elastic Bernoulli-Euler beam, let us formulate the problem on the
transverse impact response of a viscoelastic beam, the viscoelastic features of which are
described by a certain viscoelastic operator. In this case, the equations of motion of an elas-
tic spherical impactor of radius R and the viscoelastic beam of length L have, respectively,
the form
mẅ2 = −P (t), (223)
4

e ∂ w1 + %Aẅ1 = P (t)δ x − L ,
EI (224)
∂x4 2
where m is the mass of the sphere, w2 is the displacement of the sphere, P (t) is the contact
e is a certain viscoelas-
force, w1 (x, t) is the displacement of the beam at the contact point, E
tic operator, I is the moment of inertia of the beam’s cross section, A is the beam’s cross
sectional area, % is its density, δ(x − L2 ) is the Dirac delta-function, x is the longitudinal
coordinate, and an overdot denotes partial time-derivative.
Equations (223) and (224) are subjected to the following initial conditions:
w1 (x, 0) = 0, ẇ1 (x, 0) = 0, w2 (0) = 0, ẇ2 (0) = V0 , (225)
where V0 is the initial velocity of the impactor at the moment of impact.

Integrating twice Eq. (223) yields

Z t
1
w2 (t) = − P (t0 )(t − t0 )dt0 + V0 t. (226)
m 0
Substituting the solution for a simply-supported Bernoulli-Euler beam

∞
X nπ
w1 (x, t) = Tn (t) sin x (227)
L
n=1
nπ

in Eq. (224), and considering the orthogonality condition for the eigenfunctions sin L x
on the segment from 0 to L, we are led to the infinite set of uncoupled equations
e n (t) = Fn P (t),
T̈n (t) + Ω2n ET (n = 1, 2, ...) (228)
each of which describes force driven vibrations of the viscoelastic oscillator, where
E1 I nπ 4 2 nπ
Ω2n = , Fn = sin ,
%A L %AL 2
and E1 is the elastic modulus.
For those operators E,e which will be considered below, the Green function Gn (t) for
each oscillator from (228) has the form [67]
Gn (t) = A0n (t) + An e−αn t sin(ωn t − ϕn ), (229)
where the index n indicates the ordinal number of the oscillator, and all values entering in
(229) have the same structure and the same physical meaning as the corresponding values
in Eq. (162) discussed in Sect. 6.
Thus, from (229) it is seen that the Green function possesses two terms, one of which,
A0n (t), describes the drift of the equilibrium position and is represented by the integral
involving the distribution function of dynamic and rheological parameters, while the other
term is the product of two time-dependent functions, exponent and sine, and it describes
damped vibrations around the drifting equilibrium position, where An is the amplitude,
αn is the damping coefficient, and ωn and ϕn are the frequency and phase, respectively.
The first term is governed by an improper integral taken along two sides of the cut along
the negative real semi-axis of the complex plane (see Fig. 2), while the second term is
determined for each n by two complex conjugate roots of the characteristic equation, which
locate in the left half-plane of the complex plane.
Knowing the Green functions, the solution of Eq. (224) takes the form
∞
X nπ Z t
w1 (x, t) = sin x Gn (t − t0 )P (t0 )dt0 . (230)
L 0
n=1
Let us introduce the value characterizing the relative approach of the sphere and beam,
i.e., penetration of the beam by the sphere, is

L
y(t) = w2 (t) − w1 ,t , (231)
2
which is connected with the contact force by the Hertz law
P (t) = ky 3/2 , (232)
where
4√
k= RE ∗ (233)
3
is the operator involving the geometry of the impactor and the elastic features of the im-
pactor and viscoelastic features of the target, which are described due to the Volterra corre-
spondence principle by the operator E ∗
1 ∗ 1 − ν22 1 − νe12
= J = + , (234)
E∗ E2 e1
E
ν2 and E2 are constant Poisson’s coefficient and Young’s modulus, respectively, for the
elastic sphere (impactor), while νe1 and E e1 are the operators for the viscoelastic beam (tar-
get).
Further in order to obtain the integro-differential equation for the values y(t) and P (t),
it is necessary to assign the form of the operator E e1 .
Phillips and Calvit [45] were probably the first to investigate the response of a viscoelas-
tic infinitely extended plate to impact of a rigid sphere. They used the Hertz’s contact law in
its hereditary form [45]. This problem is an immediate extension of Zener’s approach [90]
for the dynamic rigid spherical-indenter problem for the case of a thin elastic plate [61].
Assuming a falling body to be rigid, the function k (233) in the formula for the contact
force in the Hertzian contact law (232) has been represented in [45] by the integral operator
with a kernel of heredity based on the constant mechanical loss tangent model.
Ingman and Suzdalnitsky [29] considered dynamic response of a circular viscoelastic
plate, viscoelastic features of which are described using the fractional derivative Kelvin-
Voigt model, subjected to the impact of a falling elastic sphere following Timoshenko ap-
proach [80] but without considering viscoelastic features of the target in the Hertz’s contact
law. The authors of [29] referring to [92] wrote that
“experimental investigations with a steel sphere and a circular composite

barrier showed that Hertz’s law is applicable not only in the case of static con-
tact of elastic bodies, but also in problems of response of composite plates to a
low-velocity impact,”
and made the conclusion that this argument could justify the choice of Eq. (232) with a
constant stiffness coefficient k. However, firstly, the problems of viscoelasticity have not
been considered in [92] (only targets made of elastic isotropic or anisotropic materials with
and without account for plastic features in the contact region were utilized in this collective
monograph), and secondly, Hunter [28] and Phillips and Calvit [45] showed that the local
viscoelastic damping associated with deformation in the contact zone is essential at least on
the first stage of active loading, since its ignorance in reducing plate impact data results in
large error in the calculation of loss tangent and other characteristic values.
Following Rossikhin and Shitikova [67], we will use the governing equations of the
viscoelastic material of the target in terms of Rabotnov’s fractional exponential operators
(23) involving one term, i.e.,

Ẽ = E∞ 1 − νε 3∗γ (τεγ ) , (235)
utilizing the algebra of such operators presented in Sect 4.

Now we write Eq. (228) in terms of the Green function Gn (t) using operator (235) as
e As a result we obtain
the operator E.

G̈n (t) + Ω2∞n 1 − νε 3∗γ (τεγ ) Gn (t) = Fn δ(t), (236)

nπ 4
where Ω2∞n = E%A
∞I
L .
Applying the Laplace transform to Eq. (236) yields
Fn (æn + pγ )
Ḡn = , (237)
p2+γ + æn p2 + Ω2∞n pγ + Ω20n æn

nπ 4
where æn = τε−γ , and Ω20n = E%A
0I
L .
The solution of the characteristic equation
p2+γ + æn p2 + Ω2∞n pγ + Ω20n æn = 0, (238)
and the inversion of the expression (238) on the first sheet of the Riemannian surface via
Mellin-Fourier inversion formula (150), results in the desired relationship (229), where the
function A0n (t) is defined according to
Z ∞
A0n (t) = τ −1 Bn (τ, æn)e−t/τ dτ, (239)
0
with the following distribution of the relaxation-retardation parameters:
sin γπ ν Ω2∞n Fn [θ∞n (τ )]−1 [θ0n (τ )]−1 τ 3

Bn (τ, æn) = ,
π θ∞n (τ )[θ0n(τ )]−1 (τ /τ )−γ + θ0n (τ )[θ∞n (τ )]−1 (τ /τ )γ + 2 cos πγ
(240)
where
θ∞n (τ ) = τ 2 Ω2∞n + 1, θ0n (τ ) = τ 2 Ω20n + 1,
while the amplitude and phase of vibrations have, respectively, the following form:
1/2
An = 2Fn (a2n + b2n )−1 (æ2n + rn2γ + 2æn rnγ cos γψn) , (241)
æn cos βn + rnγ cos(βn − γψn ) bn

tan ϕn = − , tan βn = , (242)
æn sin βn + rnγ sin(βn − γψn) an
where
an = (2 + γ)rn1+γ cos(1 + γ)ψn + 2æn rn cos ψn + γΩ2∞n rnγ−1 cos(γ − 1)ψn,
bn = (2 + γ)rn1+γ sin(1 + γ)ψn + 2æn rn sin ψn + γΩ2∞nrnγ−1 sin(γ − 1)ψn,

rn2 = ωn2 + α2n , tan ψn = −ωn αn−1 .
If we omit the number n in (237)–(242), then all relationships will coincide with formu-
las (164) and (165) devoted to the vibrations of a viscoelastic oscillator based on Rabotnov
relaxation operator with the root locus at n = 1 presented in Fig. 4.
Assuming that the bulk modulus is a constant value, i.e., relationship (51) is fulfilled,
the operator E ∗ −1 is determined according to formula (234) with due account for (80). Thus
we are led to the following functional equation for determining the contact force P (t):
2/3 Z
3 1 − ν22 1 − ν∞ 2 (1 − 2ν∞ )2 νε t t − t0
√ + P (t) + 3γ − P (t0 )dt0
4 R E2 E∞ 4E∞ 0 τ ε
Z t 2/3 Z
3νσ t − t0 1 t
+ 3γ − P (t0 )dt0 =− P (t0 )(t − t0 )dt0 + V0 t
4E∞ 0 τσ m 0
X ∞ nπ Z t
− sin Gn (t − t0 )P (t0 )dt0 . (243)
2 0
n=1
9. Conclusion
Decoding of different viscoelastic operators has been carried out, and in particular, of
the operator of cylindrical rigidity and its reverse operator, in the case when the operator
Ẽ is given on the basis of the Rabotnov model, while the bulk operator K̃ = const. It is
also proposed how to decode different viscoelastic operators, and in particular, Poisson’s
operator ν̃, if operators µ̃ and K̃ are given by the Rabotnov model. It is shown that consid-
eration for the bulk relaxation results in the appearance of the second peak in the frequency
dependence of the tangent of mechanical loss angle.
It is shown that even if one proceeds from the simple Rabotnov model, then during
calculation of such intricate operators as an operator of cylindrical rigidity, it is a need to
utilize the generalized Rabotnov model, involving at least sums of two, and sometimes of
three or more fractional operators with one and the same fractional parameter. It is proved
that the Koeller model is physically admissible only when it is reduced to the generalized
Rabotnov model.
Using fractional operators it is possible to describe not only damped vibrations of me-
chanical systems, but aperiodic motions of these systems as well.
It is shown that stationary shock waves with null discontinuities could propagate in
nonlinear hereditary media with a weakly singular kernel of heredity.
Acknowledgments
This work has been supported by the Russian Foundation for Basic Research by Grant
No. 14-08-92008-HHC-a and by Grant No. 7.22.2014/K as a Government task from the
Ministry of Education and Science of the Russian Federation.
References
[1] B.N.N. Achar, J.W. Hanneken, T. Clarke, “Response characteristics of a fractional
oscillator,” Physica A, vol. 309, pp. 275–288, 2002.
[2] B.N.N. Achar, J.W. Hanneken, T. Clarke, “Damping characteristics of a fractional

oscillator,” Physica A, vol. 339, pp. 311–319, 2004.
[3] B.N.N. Achar, J.W. Hanneken, T. Enck, T. Clarke, “Dynamics of the fractional oscil-
lator,” Physica A, vol. 297, pp. 361–367, 2001.
[4] R.L. Bagley, P.J. Torvik, “A generalized derivative model for an elastomer damper,”
Shock and Vibration Bulletin, vol. 49(2), pp. 135–143, 1979.
[5] R.L. Bagley, P.J. Torvik, “On the fractional calculus model of viscoelastic behavior,”
Journal of Rheology, vol. 30(1), pp. 133–155, 1986.
[6] M.A. Belov, A.E. Bogdanovich, “Numerical inversion of Laplace transform by the
method of asymptotic extension of the interval in dynamic viscoelasticity problems,”
Mechanics of Composite Materials (Engl transl), vol. 12(5), pp. 762–768, 1976.
[7] H. Beyer, S. Kempfle, “Definition of physically consistent damping laws with frac-
tional derivatives,” ZAMM, vol. 75(8), pp. 623–635, 1995.
[8] G.W. Scott Blair, “Analytical and integrative aspects of the stress-strain-time prob-
lem,” Journal of Scientific Instruments, vol. 21(5), pp. 80–84, 1944.
[9] P.W. Buchen, F. Mainardi, “Asymptotic expansions for transient viscoelastic waves,”
Journal de Mecanique, vol. 14(4), pp. 597–608, 1975.
[10] A.A. Burenin, Yu.A. Rossikhin, “The effect of viscosity on the character of the prop-
agation of a plane longitudinal shock wave,” Journal of Applied Mechanics and Tech-
nical Physics (Engl. transl.), vol. 31(6), pp. 807–810, 1990.
[11] M. Caputo, “Linear models of dissipation whose Q is almost frequency independent

- II,” Geophysics Journal of the Royal Astronomy Society, vol. 13(5), pp. 529–539,
1967.
[12] M. Caputo, “Vibrations of an infinite viscoelastic layer with a dissipative memory,”

Journal of the Acoustical Society of America, vol. 56(3), pp. 897–904, 1974.
[13] M. Caputo, “Vibrations of an infinite plate with a frequency independent Q,” Journal
of Acoustical Society of America, vol. 60(3), pp. 634–639, 1976.
[14] M. Caputo, F. Mainardi, “A new dissipation model based on memory mechanism,”

Pure and Applied Geophysics, Vol. 91(1), 134–147, 1971.
[15] M. Caputo, F. Mainardi, “Linear models of dissipation in inelastic solids,” La Rivista

del Nuovo Cimento, vol. 1(2), 161–198, 1971.
[16] F. Dal, “Multiple time scale solution of an equation with quadratic and cubic nonlin-
earities having fractional-order derivative,” Mathematical and Computational Appli-
cations, vol. 16(1), pp. 301–308, 2011.
[17] D.D. Demir, N. Bildik, B.G. Sinir, “Linear vibrations of continuum with fractional
derivatives,” Boundary Value Problems, 2013:104, pp.1–15, 2013.
[18] J. Engelbrecht, Nonlinear Wave Processes of Deformation in Solids, Pitman, London,

1983.
[19] A. Gemant, “A method for analyzing experimental results obtained from elasto-
viscous bodies,” Physics, vol. 7, pp. 311–317, 1936.
[20] A.N. Gerasimov, “A generalization of linear laws of deformation and its application to
the problems of internal friction” (in Russian), Prikladnaya Matematika i Mekhanika,
vol. 12, pp. 251–260, 1948.
[21] B. Gross, “On creep and relaxation,” Journal of Applied Physics, vol. 18(2), pp. 212–
221, 1947.
[22] V.L. Gonsovkii, S.I. Meshkov, Yu.A. Rossikhin, “Impact of a viscoelastic rod onto
a rigid target,” Soviet Applied Mechanics (Engl. transl.), vol. 8(10), pp. 1109–1113,
1972.
[23] V.L. Gonsovskii, Yu.A. Rossikhin, “On the propagation of an impulsive load in a
viscoelastic medium” (in Russian), in: Proceedings of the Scientific-Research Institute
of Mathematics of Voronezh University, No. 6, pp. 63–66, 1972.
[24] V.L. Gonsovkii, Yu.A. Rossikhin, “Stress waves in a viscoelastic medium with a sin-
gular hereditary kernel,” Journal of Applied Mechanics and Technical Physics (Engl.
transl.), vol. 14(4), pp. 595–597, 1973.
[25] B. Hartmann, G.F. Lee, J.D. Lee, “Loss factor height and width limits for polymer
relaxations,” Journal of the Acoustical Society of America, vol. 95(1), pp. 226–233,
1994.
[26] S. Havriliak, S. Negami, “Analysis of α-dispersion in some polymer systems by the

complex variables method,” Journal of Polymer Science (C), vol. 14, pp. 99–117,
1966.
[27] S. Havriliak, S. Negami, “On the equivalence of dielectric and mechanical dispersions
in some polymers; e.g. poly(n-octyl methacrylate)’, Polymer, vol. 10(10), pp. 859–
872, 1966.
[28] S.C. Hunter, “The Hertz problem for a rigid spherical indenter and a viscoelastic half
space,” Journal of Mechanics and Physics of Solids, vol. 8, pp. 219–234, 1960.
[29] D. Ingman, J. Suzdalnitsky, “Response of viscoelastic plate to impact,” ASME Journal

of Vibration and Acoustics, vol. 130, pp. 011010-1–8, 2008.
[30] K. Iosida, Functional Analysis, 1967.
[31] A. Jeffrey, T. Kawahara, Asymptotic Methods in Nonlinear Wave Theory, Pitman, Lon-
don, 1982.
[32] S. Kempfle, “Causality criteria for solutions of linear fractional differential equations,”
Fractional Calculus Application and Analysis, vol. 1(4), pp. 351–364, 1998.
[33] R.C. Koeller, “Applications of fractional calculus to the theory of viscoelasticity,”

ASME Journal of Applied Mechanics, vol. 51, pp. 299–307, 1984.
[34] R.C. Koeller, “Polynomial operators, Stieltjes convolution, and fractional calculus in
hereditary mechanics,” Acta Mechanica, vol. 58(3-4), pp. 251–264, 1986.
[35] M.A. Krasnosel’skii, P.P. Zabreiko, E.I. Pystul’nik, P.E. Sobolevskii, Integral Opera-
tors in Spaces of Summing Functions (in Russian), Moscow, Nauka, 1966.
[36] V.M. Mal’kov, “The viscoelasticity equations of an elastomer layer,” Journal of Ap-
plied Mathematics and Mechanics (Engl. transl.), vol. 59(2), pp. 209–216, 1995.
[37] R.P. Meilanov, M.S. Yanpolov, “Features of the phase trajectory of a fractal oscillator,”
Technical Physics Letters (Engl transl), vol. 28(1), pp. 30–32, 2002.
[38] S.I. Meshkov, “Description of internal friction in the memory theory of elasticity using
kernels with a weak singularity,” Journal of Applied Mechanics and Technical Physics
(Engl. transl.), vol. 8(4), pp. 100–102, 1967.
[39] S.I. Meshkov, G.N. Pachevskaya, “Allowance for bulk relaxation by the method of
internal friction,” Journal of Applied Mechanics and Technical Physics (Engl. transl.),
vol. 8(2), pp. 47–48, 1967.
[40] S.I. Meshkov, G.N. Pachevskaya, V.S. Postnikov, Yu.A. Rossikhin, “Integral repre-
sentation of 3γ -functions and their application to problems in linear viscoelasticity,”
International Journal of Engineering Science, vol. 9, pp. 387–398, 1971.
[41] S.I. Meshkov, G.N. Pachevskaya, Y.D. Shermergor, “Internal friction described with
the aid of fractionally-exponential kernels,” Journal of Applied Mechanics and Tech-
nical Physics (Engl. transl.), vol. 7(3), pp. 63–65, 1966.
[42] S.I. Meshkov, Yu.A. Rossikhin, “Propagation of acoustic waves in a hereditarily elas-
tic medium,” Journal of Applied Mechanics and Technical Physics (Engl. transl.),
vol. 9(5), pp. 589–592, 1968.
[43] S.I. Meshkov, Yu.A. Rossikhin, “Temperature dependence of the damping coefficients
for a dynamical system with a singular kernel,” Journal of Engineering Physics and
Thermophysics (Engl. transl.), vol. 21(2), pp. 1090, 1971.
[44] E. Pelinovski, V. Fridman, J. Engelbrecht, Nonlinear Evolution Equations (in Rus-

sian), Valgus, Tallinn, 1984.
[45] J.W. Phillips, H.H. Calvit, “Impact of a rigid sphere on a viscoelastic plate,” ASME
Journal of Applied Mechanics, vol. 34, pp. 873–878, 1967.
[46] Yu.N. Rabotnov, “Equilibrium of an elastic medium with after-effect” (in Russian),
Prikladnaya Matematika i Mekhanika, vol. 12(1), pp. 53–62, 1948.
[47] Yu.N. Rabotnov, Creep Problems in Structural Members, (Engl. transl. by Transcripta
Service LTD, London, F.A. Leckie, ed.), North-Holland Series in Applied Mathe-
matics and Mechanics, 7, North-Holland Publishing, Amsterdam-London. Originally
published in Russian in 1966 as Polzuchest’ Elementov Konstruktsii, Nauka, Moscow,
1969.
[48] Yu.N. Rabotnov, Elements of Hereditary Solid Mechanics, Nauka, Moscow (Engl.
transl. by Mir Publishers, Moscow in 1980) 1977.
[49] A.R. Rhzanitsyn, Some Problems of the Mechanics of Systems Deformable with Time
(in Russian), Gostekhizdat, Moscow, 1949.
[50] Yu.A. Rossikhin, “Reflections on two parallel ways in the progress of fractional calcu-
lus in mechanics of solids,” Applied Mechanics Reviews, vol. 63(1), pp. 010701-1–12,
2010.
[51] Yu.A. Rossikhin, M.V. Shitikova, “On the structure of waves in a nonlinear elastic
medium,” Facta Universitatis. Series “Mechanics, Automatic Control and Robotics”,
vol. 1(4), pp. 437–449, 1994.
[52] Yu.A. Rossikhin, M.V. Shitikova, “Ray method for solving dynamic problems con-
nected with propagation of wave surfaces of strong and weak discontinuities,” Applied
Mechanics Reviews, vol. 48(1), pp. 1–39, 1995.
[53] Yu.A. Rossikhin, M.V. Shitikova, “Derivation of nonlinear evolutionary equations for
a shock layer of finite thickness,” in: Nonlinear Acoustics in Perspective, Proc. 14th
Int. Symp. on Nonlinear Acoustics (R.J. Wei, Ed.), Nanjing University Press, Nanjing,
pp. 270–274, 1996.
[54] Yu.A. Rossikhin, M.V. Shitikova, “Application of fractional derivatives for the analy-
sis of nonlinear damped vibrations of suspension bridges,” in: Proc. 1997 Int. Symp.
on Nonlinear Theory and its Applications, vol. 1, pp. 541–544, Honolulu, USA, Nov
29 - Dec 2, 1997.
[55] Yu.A. Rossikhin, M.V. Shitikova, “Applications of fractional calculus to dynamic

problems of linear and nonlinear hereditary mechanics of solids,” Applied Mechan-
ics Reviews, vol. 50(1), 15–67, 1997.
[56] Yu.A. Rossikhin, M.V. Shitikova, “Application of fractional operators to the analy-
sis of damped vibrations of viscoelastic single-mass systems,” Journal of Sound and
Vibration, vol. 199(4), pp. 567–586, 1997.
[57] Yu.A. Rossikhin, M.V. Shitikova, “Application of fractional derivatives to the analysis
of damped vibrations of viscoelastic single-mass systems,” Acta Mechanica, vol. 120,
pp. 109–125, 1997.
[58] Yu.A. Rossikhin, M.V. Shitikova, “Application of fractional calculus for analysis of
nonlinear damped vibrations of suspension bridges,” ASCE Journal of Engineering
Mechanics, vol. 124 (9), pp. 1029–1036, 1998.
[59] Yu.A. Rossikhin, M.V. Shitikova, “Analysis of rheological equations involving more
than one fractional parameters by the use of the simplest mechanical systems based
on these equations,” Mechanics of Time-Dependent Materials, Vol. 5, pp. 131–175,
2001.
[60] Yu.A. Rossikhin, M.V. Shitikova, “Analysis of the viscoelastic rod dynamics via mod-
els involving fractional derivatives or operators of two different orders,” The Shock
and Vibration Digest, vol. 36(1), pp. 3–26, 2004.
[61] Yu.A. Rossikhin, M.V. Shitikova, “Transient response of thin bodies subjected to im-
pact: Wave approach,” Shock and Vibration Digest, vol. 39, pp. 273–309, 2007.
[62] Yu.A. Rossikhin, M.V. Shitikova, “Comparative analysis of viscoelastic models in-
volving fractional derivatives of different orders,” Fractional Calculus & Applied
Analysis, vol. 10, pp. 111–121, 2007.
[63] Yu.A. Rossikhin, M.V. Shitikova, “Free damped vibrations of a viscoelastic oscilla-
tor based on Rabotnov’s model,” Mechanics of Time-Dependent Materials, vol. 12,
pp. 129–149, 2008.
[64] Yu.A. Rossikhin, M.V. Shitikova, “Application of fractional calculus for dynamic
problems of solid mechanics: Novel trends and recent results,” Applied Mechanics
Reviews, vol. 63(1), pp. 010801-1–52, 2010.
[65] Yu.A. Rossikhin, M.V. Shitikova, “Analysis of free vibrations of a viscoelastic oscil-
lator via the models involving several fractional parameters and relaxation/retardation
times,” Computes and Mathematics with Applications, vol. 59, pp. 1727–1744, 2010.
[66] Yu.A. Rossikhin, M.V. Shitikova, “On fallacies in the decision between the Caputo
and Riemann-Liouville fractional derivatives for the analysis of the dynamic response
of a nonlinear viscoelastic oscillator,” Mechanics Research Communications, vol. 45,
pp. 22–27, 2012.
[67] Yu.A. Rossikhin, M.V. Shitikova, “Two approaches for studying the impact response
of viscoelastic engineering systems: An overview,” Computes and Mathematics with
Applications, vol. 66, pp. 755–773, 2013.
[68] Yu.A. Rossikhin, M.V. Shitikova, T.A. Shcheglova, “Control of dynamic response of
a fractional oscillator via rheological parameters variation,” in: Proceedings of the 4th
European Conference of Structural Control (A.K. Belyaev and D.A. Indeitsev, eds),
St. Petersburg, Russia, Sept. 8-12, 2008, vol. 2, pp. 719–726, 2008.
[69] Yu.A. Rossikhin, M.V. Shitikova, T.A. Shcheglova, “Forced vibrations of a nonlin-
ear oscillator with weak fractional damping,” Journal of Mechanics of Materials and
Structures, vol. 4(9), pp. 1619–1636, 2009.
[70] M.I. Rozovskii, S.E. Sinaiskii, “Vibrations of an oscillator with residual creep,” Jour-
nal of Applied Mathematics and Mechanics (Engl. transl.), vol. 30(3), pp. 696–703,
1966.
[71] A.R. Rzhanitsyn, Some Questions of Mechanics of Systems Deformable in Time (in
Russian), Moscow: Gostekhizdat, 1949.
[72] A.R. Rzhanitsyn, Theory of Creep (in Russian), Moscow: Stroiizdat, 1968.
[73] S. G. Samko, A. A. Kilbas and O. I. Marichev, Fractional Integrals and Derivatives.

Theory and Applications (in Russian), Nauka i Tekhnika, Minsk 1988 (Engl. transl.
by Gordon and Breach Science Publ.)
[74] M.F. Selivanov, Effective properties of a linear viscoelastic composite, International

Applied Mechanics, vol. 45, pp. 1084–1091, 2009.
[75] M.F. Selivanov, Y.O. Chornoivan, Computational optimization of characteristics for

composites of viscoelastic components, Journal of Engineering Mathematics, vol. 74,
91–100, 2012.
[76] D.T. Shermergor, “On the use of fractional differentiation operators for the descrip-
tion of elastic-aftereffect properties of materials,” Journal of Applied Mechanics and
Technical Physics (Engl. transl.), vol. 7(6), pp. 85–87, 1966.
[77] G.I. Slonimsky, “Laws of mechanical relaxation processes in polymers,” Journal of

Polymer Science, Part C vol. 16, pp. 1667–1672, 1967.
[78] W. Smit, H. de Vries, “Rheological models containing fractional derivatives,” Rheo-

logica Acta, vol. 9, pp. 525–534, 1970.
[79] A.A. Stanislavsky, “Fractional oscillator,” Physical Review E, vol. 70, pp. 051103-1–
6, 2004.
[80] S.P. Timoshenko, “Zur Frage nach der Wirkung eines Stosses auf einen Balken” (in
German), Zeitschrift für Mathematik und Physik, vol. 62, pp. 198–209, 1913.
[81] T.Y. Thomas, Plastic Flow and Fracture in Solids, Academic Press, New York, 1961.
[82] A. Tofighi, “The intrinsic damping of the fractional oscillator,” Physica A, vol. 329,
pp. 29–34, 2003.
[83] P.J. Torvik, R.L. Bagley, “On the appearance of the fractional derivative in the be-
havior of real materials,” ASME Journal of Applied Mechanics, vol. 51, pp. 294–298,
1984.
[84] P.J. Torvik, R.L. Bagley, “A different view of viscous damping,” Shock and Vibration
Bulletin, vol. 55(3), pp. 81–84, 1985.
[85] T. Usuki, “Dispersion curves of viscoelastic plane waves and Rayleigh surface wave
in high frequency range with fractional derivatives,” Journal of Sound and Vibration,
vol. 332, pp. 4541–4559, 2013.
[86] S.Z. Vulfson, “Thermal stresses in concrete solid masses with due account for creep in
concrete” (in Russian), Izvestija Akademii Nauk SSSR Mekhanika i Mashinostroenie,
Issue 1, pp. 162–165, 1960.
[87] G.M. Zaslavsky, A.A. Stanislavsky, M. Edelman, “Chaotic and pseudochaotic attrac-
tors of perturbed fractional oscillator,” Chaos, vol. 16, pp. 013102-1–6, 2006.
[88] V.M. Zelenev, S.I. Meshkov, Yu.A. Rossikhin, “Effect of the 3-function singularity
parameters on the damped vibrations of elastic systems with aftereffect,” Mechanics
of Solids (Engl. transl.), vol. 5(3), pp. 92-94, 1970.
[89] V.M. Zelenev, S.I. Meshkov, Yu. A. Rossikhin, “Damped vibrations of hereditary-
elastic systems with weakly singular kernels,” Applied Mechanics and Technical
Physics (Engl. transl.), vol. 11(2), pp. 290–293, 1970.
[90] C. Zener, “The intrinsic inelasticity of large plates,” Physics Reviews, vol. 59, pp. 669–
673, 1941.
[91] C. Zener, Elasticity and Anelasticity of Metals, the University of Chicago Press,
Chicago, 1948.
[92] J.A. Zukas, T. Nicholas, H.F. Swift, L.B. Greszczuk, D.R. Curran, Impact Dynamics,
Wiley, New York, 1982.
[93] D.R. Bland, The Theory of Linear Viscoelasticity, International series of Monographs
on Pure and Applied Mathematics, vol. 10, (I.N. Sneddon and S. Ulam, General Edi-
tors), Pergamon Press, Oxford, 1960.
Chapter 9
T HEORY OF D IFFUSIVE S TRESSES B ASED

ON THE F RACTIONAL A DVECTION -D IFFUSION
E QUATION
Yuriy Povstenko∗
Institute of Mathematics and Computer Science
Jan Długosz University in Czȩstochowa, Poland
Abstract
Conventional theory of diffusive stresses is based on the principles of the classical the-
ory of diffusion, specifically on the classical Fick law, which relates the matter flux
to the concentration gradient. In combination with the balance equation for mass, this
law leads to the classical diffusion equation. We study time-nonlocal generalizations
of the diffusive flux governed by the Fick law and of the advection flux associated
with the velocity field. The nonlocal constitutive equations with the long-tail power
memory kernel result in the time-fractional advection-diffusion equation. The nonlo-
cal constitutive equations with the middle-tail memory kernel expressed in terms of the
Mittag-Leffler function lead to the fractional advection-diffusion equation of the Cat-
taneo type. The theory of diffusive stresses based on the fractional advection-diffusion
equation is formulated.
Keywords: Fractional calculus, advection-diffusion, diffusive stresses, Mittag-Leffler

function
1. Introduction
The classical Fourier law constitutes the relation between the heat flux q and the tem-
perature gradient
q = −k grad T. (1)
In combination with a law of conservation of energy, equation (1) leads to the parabolic
heat conduction equation
∂T
= aT ∆T (2)
∂t
∗
E-mail address: j.povstenko@ajd.czest.pl
228 Yuriy Povstenko
with aT being the heat diffusivity coefficient.

From the mathematical point of view, the Fourier law in the theory of heat conduction
corresponds to the Fick law in the theory of diffusion
j = −a grad c, (3)
where j is the matter flux, c is the concentration. In combination with the balance equation
for mass, equation (3) results in the classical diffusion equation:
∂c
= a ∆c (4)
∂t
with a being the diffusivity coefficient.
The classical heat conduction and diffusion equations based on the Fourier and Fick
laws, respectively, are quite acceptable for different physical situations. However, many
theoretical and experimental studies testify that in media with complex internal structure
the standard parabolic equations are no longer accurate enough. In nonclassical theories,
the Fourier and Fick laws as well as the parabolic heat conduction and diffusion equations
are replaced by more general equations. Some of these theories were formulated in terms
of the theory of heat conduction, other in terms of diffusion. For an extensive bibliography
on this subject and further discussion see Chandrasekharaiah (1986, 1998), Ignaczak and
Ostoja-Starzewski (2010), Joseph and Preziosi (1989), Metzler and Klafter (2000, 2004),
Zaslavsky (2002) and references therein.
Gurtin and Pipkin (1968) proposed the general time-nonlocal dependence between the
heat flux vector and the temperature gradient. Nigmatullin (1984a, 1984b) considered the
following general form of such an equation:
Z t
q(t) = −k K (t − τ ) grad T (τ ) dτ (5)
0
resulting in the heat conduction equation with memory

Z t
∂T
= aT K (t − τ ) ∆T (τ ) dτ. (6)
∂t 0
Nigmatullin (1984a) and Green and Naghdi (1993) proposed the constitutive equation
of heat conduction in the case of constant kernel (full memory with no memory decay)
Z t
q (t) = −k grad T (τ )dτ . (7)
0
The wave equation for temperature
∂ 2T
= aT ∆T (8)
∂t2
obtained from Eq. (7) is a constituent part of thermoelasticity without energy dissipation
developed by Green and Naghdi (1993).
Theory of Diffusive Stresses ... 229
Cattaneo (1948) introduced the generalized constitutive equation for the heat flux which
can be rewritten in a non-local form with the “short-tail” exponential time-nonlocal kernel
Z
k t t−τ
q (t) = − exp − grad T (τ ) dτ, (9)
ζ 0 ζ
where ζ is a nonnegative constant. This equations leads to the telegraph equation for tem-
perature
∂T ∂2T
+ ζ 2 = aT ∆T . (10)
∂t ∂t
Based on this equation, Lord and Shulman (1967) proposed the generalized theory of ther-
moelasticity.
The time-nonlocal dependences between the heat flux vector and the temperature gra-
dient with the “long-tail” power kernel (see Povstenko (2005, 2009a, 2013))
Z t
k ∂
q (t) = − (t − τ )α−1 grad T (τ ) dτ, 0 < α ≤ 1, (11)
Γ (α) ∂t 0
Z t
k
q (t) = − (t − τ )α−2 grad T (τ ) dτ, 1 < α ≤ 2, (12)
Γ (α − 1) 0
can be interpreted in terms of fractional calculus
q (t) = −k D 1−α grad T (t), 0 < α ≤ 1, (13)
q (t) = −k I α−1 grad T (t), 1 < α ≤ 2, (14)

and result in the time-fractional heat conduction equation
∂αT
= aT ∆T, 0 < α ≤ 2, (15)
∂tα
or in terms of diffusion
∂ αc
= a ∆c, 0 < α ≤ 2, (16)
∂tα
with the particular cases corresponding to subdiffusion (weak diffusion) (0 < α < 1);
normal diffusion (α = 1); superdiffusion (strong diffusion) (1 < α < 2), and ballistic
diffusion (α = 2).
d α f (t)
In Eqs. (13)–(15) I αf (t), D α f (t) and dtα are the Riemann-Liouville fractional
integral, Riemann-Liouville fractional derivative and the Caputo fractional derivative, re-
spectively (Kilbas et al. (2006), Podlubny (1999)):
Z t
1
I α f (t) = (t − τ )α−1 f (τ ) dτ, α > 0, (17)
Γ(α) 0
Z t
α dm 1 m−α−1
D f (t) = m (t − τ ) f (τ ) dτ , m − 1 < α < m, (18)
dt Γ(m − α) 0
Z t
d αf (t) 1 m
m−α−1 d f (τ )
= (t − τ ) dτ, m− 1 < α < m. (19)
dtα Γ(m − α) 0 dτ m
230 Yuriy Povstenko
The general fractional Cattaneo-type equations were considered by Povstenko (2012,

2013):
Z
k t (t − τ )β−α
q (t) = − (t − τ )β−2 Eβ−α, β−1 − grad T (τ ) dτ, (20)
ζ 0 ζ
where Eα,β (z) is the two-parameter Mittag-Leffler function (see Kilbas et al. (2006), Pod-
lubny (1999))
∞
X zn
Eα,β (z) = , α > 0, β > 0, (21)
Γ (αn + β)
n=0
being the generalization of the exponential function. The constitutive equation (20) leads
to the fractional telegraph equation for temperature
∂ αT ∂βT
+ ζ = aT ∆T . (22)
∂tα ∂tβ
Several particular casers of Eqs (20) and (22) corresponding to the different choise of α and
β were analyzed by Povstenko (2011) (see also Compte and Metzler (1997)) and were used
in construction of the corresponding theories of thermoelasticity (Yussef (2010), Sherief et
al. (2010), Povstenko (2011)). Analysis of space-fractional generalizations of the Fourier
law and the heat conduction equation as well as of the thermoelasticity theories based on
these equations can be found in Povstenko (2009a, 2009b, 2012, 2013).
2. Time-fractional Advection-diffusion Equation

Consider the following constitutive equation for the matter flux (see, for example, Ka-
viany (1995)):
j = −a grad c + vc, (23)
where v is the velocity vector.
In combination with the balance equation for mass we get
∂c
= a ∆c − div (vc) (24)
∂t
or
∂c
= a ∆c − (div v) c − v · grad c. (25)
∂t
Supposing v = const (or div v = 0 for generality), we obtain the standard advection-
diffusion equation
∂c
= a ∆c − v · grad c (26)
∂t
which has several physical interpretations in terms of Brownian motion, diffusion or heat
transport with external force or with additional velocity field, diffusion of charge in the elec-
trical field on comb structure, transport processes in porous media, groundwater hydrology,
etc. (Bird et al. (2002), Feller (1971), Kaviany (1995), Nield and Bejan (2006), Risken
(1989), Scheidegger (1974), Van Kampen (1981)).
In the case of one spatial coordinate x, equation (26) has the following form
∂c ∂2c ∂c
=a 2 −v . (27)
∂t ∂x ∂x
The constitutive equation (23) as well as the advection-diffusion equation (26) can be
generalized like the Fick law (3) and the standard diffusion equation (4) similarly to equa-
tions considered in the previous section. We will restrict our consideration to the case
v = const. For example, the constitutive equation
Z t
j(t) = K (t − τ ) [−a grad c(τ ) + vc(τ )] dτ (28)
0
leads to the advection-diffusion equation with the general memory kernel:

Z t
∂c
= K (t − τ ) [a ∆c(τ ) − v · grad c(τ )] dτ. (29)
∂t 0
Equations describing the ballistic advection-diffusion have the following form:

Z t
j (t) = [−a grad c(τ ) + vc(τ )] dτ , (30)
0
and
∂ 2c
= a ∆c − v · grad c. (31)
∂t2
The time-nonlocal constitutive equation with the “short-tail” exponential kernel
Zt
1 t−τ
j (t) = exp − [−a grad c(τ ) + vc(τ )] dτ (32)
ζ ζ
0
results in the telegraph advection-diffusion equation
∂c ∂ 2c
+ ζ 2 = a ∆c − v · grad c. (33)
∂t ∂t
The time-nonlocal constitutive equations for the matter flux with the “long-tail” power
kernel
j (t) = D 1−α [−agrad c(t) + vc(t)] , 0 < α ≤ 1, (34)
j (t) = I α−1 [−agrad c(t) + vc(t)] , 1 < α ≤ 2, (35)
give the time-fractional advection-diffusion equation
∂αc
= a ∆c − v · grad c, 0 < α ≤ 2. (36)
∂tα
Time-fractional generalizations of the advection-diffusion equation were considered by
many authors (Barkai (2001), Barkai et al. (2000), Huang and Cao (2013), Huang and
Liu (2005), Jumarie (1992), Karatay and Bayramoglu (2012), Liu et al. (2003), Metzler
et al. (1999), Zheng and Wei (2010)). In majority of the abovementioned papers, the
232 Yuriy Povstenko
fractional generalizations of the one-dimensional equation (27) were considered. It should

be noted that interesting results were also obtained for fractional reaction-diffusion systems
(see Datsko and Gafiychuk (2012), Gafiychuk and Datsko (2010)).
Lastly, the constitutive equation
Z
1 t β−2 (t − τ )β−α
j (t) = (t − τ ) Eβ−α, β−1 − [−a grad c(τ ) + vc(τ )] dτ (37)
ζ 0 ζ
produces the following fractional telegraph equation with the advection term:
∂ αc ∂βc
+ ζ = a ∆c − v · grad c. (38)
∂tα ∂tβ
3. Theory of Diffusive Stresses

The theory of diffusive strain and stress is governed by a system of partial differential
equations consisting of the equations of elasticity with gradient of concentration taken into
account. The problem can be treated as an uncoupled in the case when the influence of the
time change of strain on the matter flux is neglected. The first theoretical investigation of
diffusive stresses dates back to Podstrigach (1961, 1963, 1965). For additional references,
further generalizations and discussion see Nowacki (1974), Nowacki and Olesiak (1991),
Podstrigach and Povstenko (1985), among others.
Povstenko (2005) proposed the theory of thermoelasticity based on the time-fractional
heat conduction equation. A comprehesive review of further generalizations of fractional
thermoelasticity (fractional theory of diffusive stresses) can be found in Povstenko (2013).
The stressed-strained state of a solid is governed by the equilibrium equation in terms
of displacements
µ∆u + (λ + µ)grad divu = βc Kgradc, (39)
the stress-strain-concentration relation
σ = 2µe + (λ tr e − βc Kc)I (40)
and the corresponding equation describing diffusion. In this paper we will use the time-
fractional advection-diffusion equation
∂ αc
= a ∆c − v · grad c. (41)
∂tα
Here u is the displacement vector, σ the stress tensor, e the linear strain tensor, λ and µ are
Lamé constants, K = λ + 2µ/3, βc is the diffusion coefficient of volumetric expansion, I
denotes the unit tensor.
As in the case of classical thermoelasticity or the standard theory of diffusive stresses
we can use the representation of stresses in terms of the displacement potential Φ (Parkus
(1959))
σ = 2µ (∇∇Φ − I ∆Φ) , (42)
where ∇ stands for the gradient operator. The displacement potential is determined from
the equation
1 + ν βc
∆Φ = mc, m= (43)
1−ν 3
with ν being the Poisson ratio.
3.1. The Time-fractional Advection-diffusion Equation in a Plane

Consider the time-fractional advection diffusion equation
2
∂ αc ∂ c ∂2c ∂c ∂c
α
=a 2
+ 2 −v −v , −∞ < x < ∞, −∞ < y < ∞, (44)
∂t ∂x ∂y ∂x ∂y
under initial condition

t = 0 : c = p0 δ(x) δ(y). (45)
Here δ(x) is the Dirac delta function. In (45) we have introduced the constant multiplier p0
to obtain the nondimensional quantities displayed in Figures.
The zero conditions at infinity are also imposed:
lim c(x, y, t) = 0, lim c(x, y, t) = 0. (46)

x→±∞ y→±∞
In this paper, we will use the integral transform technique. Recall that the Caputo
fractional derivative for its Laplace transform rule requires the knowledge of the initial
values of the function f (t) and its integer derivatives of order k = 1, 2, . . ., m − 1 (see
Kilbas et al. (2006), Podlubny (1999)):
m−1
X
dα f (t) α ∗
L = s f (s) − f (k) (0+ )sα−1−k , m − 1 < α < m, (47)
dtα
k=0
where s is the transform variable, the asterisk denotes the Laplace transform.
The exponential Fourier transform
Z ∞
e 1
F {f (x)} = f (ξ) = √ f (x) eixξ dx (48)
2π −∞
is used in the domain −∞ < x < ∞ and has the inverse
n o Z ∞
−1 e 1
F f (ξ) = f (x) = √ fe(ξ) e−ixξ dξ . (49)
2π −∞
The Fourier transform of the mth derivative of a function has the form
m
d f (x)
F = (−iξ)mfe(ξ). (50)
dxm
Applying to Eq. (44) the Laplace transform with respect to time t and the double Fourier
transform with rspect to the space coordinates x and y gives
ee p0 sα−1
c∗ (ξ, η, s) = . (51)
2π sα + a(ξ 2 + η 2 ) − iv(ξ + η)
or after inversion of transforms

Z ∞Z ∞
p0
c(x, y, t) = 2
Eα − a ξ 2 + η 2 − iv(ξ + η) tα e−ixξ−iyη dξ dη, (52)
4π −∞ −∞
234 Yuriy Povstenko
where the following formula (Kilbas et al. (2006), Podlubny (1999))

α−1
s
L−1
= Eα (−btα ) (53)
sα + b
has been used with Eα(z) being the Mittag-Leffler function in one parameter α
∞
X zk
Eα(z) = , α > 0, z ∈ C. (54)
Γ(αk + 1)
k=0
The substitution
v v
ξ0 = ξ − i , η0 = η − i
2a 2a
allows us to get
Z ∞ Z ∞
p0 v(x + y) 2 2 v2 α
c(x, y, t) = exp Eα − a(ξ + η ) + t
4π 2 2a −∞ −∞ 2a
(55)
× cos(xξ) cos(yη) dξ dη,
where the primes have been omitted for brevity sake.

Next we introduce the polar coordinates in the (ξ, η)–plane:
ξ = ρ cos θ, η = ρ sin θ,
thus obtaining:
Z ∞ Z 2π
p0 v(x + y) 2 v2 α
c(x, y, t) = exp Eα − aρ + t
4π 2 2a 0 0 2a
(56)
× cos(xρ cos θ) cos(yρ sin θ) ρ dρ dθ.
Due to periodic properties of the integrand

Z 2π Z π/2
cos(xρ cos θ) cos(yρ sin θ) dθ = 4 cos(xρ cos θ) cos(yρ sin θ) dθ. (57)
0 0
Changing variables w = sin θ and taking into account the following integral (Prudnikov
et al. (1981))
√
Z 1 cos p 1 − x2
π p 2
√ cos(qx) dx = J0 p + q2 , (58)
0 1 − x2 2
where Jn is the Bessel function of the first kind of the order n, we get
Z ∞ p
p0 v(x + y) 2 v2 α
c(x, y, t) = exp Eα − aρ + t J0 x2 + y 2 ρ ρ dρ.
2π 2a 0 2a
(59)
0.10
v̄ = 0 v̄ = 1
0.08 J
J
J
J^

0.06 v̄ = 2

c̄

0.04

0.02
0.00
−3.0 −2.0 −1.0 0.0 1.0 2.0 3.0
x̄
Figure 1. Dependence of the fundamental solution to the Cauchy problem on distance (the
classical advection-diffusion equation, α = 1).
Let us analyze several particular cases of the solution (59). In the case of standard
advection-diffusion equation (α = 1 and E1 (−x) = e−x ), evaluating the integral (Prud-
nikov et al. (1983))
Z ∞ 2
−px2 1 q
xe J0 (qx) dx = exp − , p > 0, q > 0, (60)
0 2p 4p
we arrive at
p0 (x − vt)2 + (y − vt)2
c(x, y, t) = exp − . (61)
4πat 4at
In the case α = 1/2 the Mittag-Leffler function can be represented as
Z ∞
2 2
E1/2(−x) = √ e−u −2xu du (62)
π 0
and the corresponding solution reads
Z ∞ √ √
p0 1 −u2 (x − 2v tu)2 + (y − 2v tu)2
c(x, y, t) = √ e exp − √ du. (63)
4a tπ 3/2 0 u 8a tu
The results of numerical computations of concentration for y = 0 are presented in Fig. 1
for α = 1 and in Fig. 2 for α = 0.5. The following nondimensional quantities
atα tα/2 x y
c̄ = c, v̄ = √ v, x̄ = √ α/2 , ȳ = √ (64)
p0 a at atα/2
have been introduced.
236 Yuriy Povstenko
0.5
0.4
0.3
c̄
0.2
v̄ = 4 v̄ = 4 v̄ = 1
@
v̄ = 1 @ v̄ = 0
@ R
@
0.1 v̄ = 0
@
@ R
@
@ ?
R
@
0.0
−3.0 −2.0 −1.0 0.0 1.0 2.0 3.0
x̄
Figure 2. Dependence of the fundamental solution to the Cauchy problem on distance (the
time-fractional advection-diffusion equation, α = 0.5).
4. Diffusive Stresses
The double Fourier transform applied to Eq. (43) gives:
e
e =− m e
Φ c
e (65)
ξ2 + η2
and after inverting the transforms
Z Z
mp0 ∞ ∞ e−ixξ−iyη
Φ=− 2 Eα − a ξ 2 + η 2 − iv(ξ + η) tα dξ dη. (66)
4π −∞ −∞ ξ2 + η2
The corresponding stress components are calculated according to (42) and have the
following form:
Z Z
2µmp0 ∞ ∞ 2 2
α η 2 e−ixξ−iyη
σxx = − E α − a ξ + η − iv(ξ + η) t dξ dη,
4π 2 −∞ −∞ ξ2 + η2
Z Z (67)
2µmp0 ∞ ∞ 2 2
α ξ 2 e−ixξ−iyη
σyy = − Eα − a ξ + η − iv(ξ + η) t dξ dη,
4π 2 −∞ −∞ ξ2 + η2
Z Z (68)
2µmp0 ∞ ∞ ξηe−ixξ−iyη
σxy = Eα − a ξ 2 + η 2 − iv(ξ + η) tα dξ dη. (69)
4π 2 −∞ −∞ ξ2 + η2
0.00
A
K
A
−0.01 A
A
A
v=2
−0.02
σ̄xx
A
K
−0.03 A
A
A
A
−0.04 v=1

v=0
−0.05
−3.0 −2.0 −1.0 0.0 1.0 2.0 3.0
x̄
Figure 3. Dependence of the stress component σxx on distance ((the classical advection-
diffusion equation, α = 1).
In the particular case of diffusive stresses caused by the solution to the standard
advection-diffusion equation (α = 1) we obtain
(
µmp0 1 (y − vt)2 (x − vt)2 + (y − vt)2
σxx = − exp −
π 2at (x − vt)2 + (y − vt)2 4at
) (70)
(x − vt)2 − (y − vt)2 (x − vt)2 + (y − vt)2
+ 1 − exp − .
[(x − vt)2 + (y − vt)2 ]2 4at
(
µmp0 (x − vt)(y − vt) 1 (x − vt)2 + (y − vt)2
σxy = exp −
π (x − vt)2 + (y − vt)2 2at 4at
) (71)
2 (x − vt)2 + (y − vt)2
− 1 − exp − .
(x − vt)2 + (y − vt)2 4at
σyy = −2µmc − σxx . (72)
To get the stress components (70)–(71) the polar coordinates in the (ξ, η)-plane have
been inroduced and the following integrals (Prudnikov et al. (1981))
Z 1p p
1 − x2 cos (bx) cos c 1 − x2 dx
0
" # (73)
π c2 p b 2
− c 2 p
= J0 b 2 + c2 + J1 b 2 + c2
2 b 2 + c2 (b2 + c2 )3/2
238 Yuriy Povstenko
0.00
v=1
A A
K
A
A A
A A
A A
−0.05
AA
U A
A
v=4
v=4
σ̄xx

A
K
v=1 A
−0.10 A
A
A
v=0
v=0
−0.15
−2.0 −1.0 0.0 1.0 2.0

x̄
Figure 4. Dependence of the the stress component σxx on distance (the time-fractional
advection-diffusion equation, α = 0.5).
and Prudnikov et al. (1983)

Z ∞ 2
−px2 1 q
e J1 (qx) dx = 1 − exp − , p > 0, q > 0, (74)
0 q 4p
have been used.
Additionally,
Z 1 p
x sin (bx) sin c 1 − x2 dx
0
(75)
π bc p 2 p
= − J0 2 2
b +c + √ J1 2
b +c 2 .
2 b 2 + c2 b 2 + c2
Another particular case α = 1/2 results in the stress component

Z * √ 2
2µmp0 ∞
−u2 y − 2v tu
σxx = − 3/2
π
e √ h √ 2 √ 2 i
0 4a tu x − 2v tu + y − 2v tu
" √ 2 √ 2 #
x − 2v tu + y − 2v tu
× exp − √
8a tu
√ 2 √ 2 (76)
x − 2v tu − y − 2v tu
+h √ 2 √ 2 i2
x − 2v tu + y − 2v tu
( " √ 2 √ 2 #) +
x − 2v tu + y − 2v tu
× 1 − exp − √ .
8a tu
Dependence of the nondimensional quantity

atα
σ̄xx = σxx
2µmp0
on the nondimensional coordinate x̄ for ȳ = 0 is shown in Fig. 3 for α = 1 and in Fig. 4
for α = 0.5.
5. Conclusion
We have analyzed the time-nonlocal generalizations of the constitutive equation for the
matter flux resulting in fractional advection-diffusion equation. The theory of diffusive
stresses associated with fractional advection-diffusion equation has been proposed. The
fundamental solution to the Cauchy problem for such an equation has been considered in the
case of two space variables, which has been solved using the integral transform technique.
It should be emphasized that the fundamental solution to the Cauchy problem in the case
0 < α < 1 has the logarithmic singularity at the origin:
r p !
p0 v(x + y) v 2 tα x2 + y 2
c(x, y, t) ∼ − exp ln 1+ √ α/2 . (77)
2πΓ(1 − α)atα 2a 2a at
This result is similar to the case of the time-fractional diffusion equation when v = 0
(see Povstenko (2005) and Schneider (1990)). Such a singularity disappears only for the
classical advection diffusion equation (α = 1). Due to singularity of the solution at the
origin, in the case 0 < α < 1 drift caused by the quantity v is less noticeable than in the
case α = 1 (compare Figs. 1 and 2 and Figs. 3 and 4).
References
Barkai, E., (2001). Fractional Fokker-Planck equation, solution, and application. Phys.
Rev. 63, 046118.
Barkai, E., Metzler, R., Klafter, J., (2000). From continuous time random walks to the
fractional Fokker-Planck equation. Phys. Rev. E 61, 132-138.
Bird, R.B., Stewart, W.E., Lightfoot, E.N., (2002). Transport Phenomena, 2nd ed. John
Wiley & Sons, Inc., New York.
Cattaneo, C., (1948). On the conduction of heat. Atti Sem. Mat. Fis. Univ. Modena, 3,
83-101.
Chandrasekharaiah, D.S., (1986). Thermoelasticity with second sound: a review. Appl.
Mech. Rev. 39, 355-376.
Chandrasekharaiah, D.S., (1998). Hiperbolic thermoelasticity: a review of recent literature.
Appl. Mech. Rev. 51, 705-729.
Compte, A., Metzler, R., (1997). The generalized Cattaneo equation for the description of
anomalous transport processes. J. Phys. A: Math.Gen. 30, 7277-7289.
Datsko, B., Gafiychuk, V., (2012). Complex nonlinear dynamics in subdiffusive activator-
inhibitor systems. Commun. Nonlinear Sci. Numer. Simulat. 17, 1673-1680.
240 Yuriy Povstenko
Feller, W., (1971). An Introduction to Probability Theory and Its Applications, 2nd ed. John
Wiley & Sons, New York.
Gafiychuk, V., Datsko, B., (2010). Mathematical modeling of different types of instabilities
in time fractional reaction-diffusion systems. Comput. Math. Appl. 59, 1101-1107.
Green A.E., Naghdi, P.M., (1993), Thermoelasticity without energy dissipation. J. Elast.
31, 189-208.
Gurtin, M.E., Pipkin, A.C., (1968). A general theory of heat conduction with finite wave
speeds. Arch. Rational Mech. Anal. 31, 113-126.
Huang, H., Cao, X., (2013). Numerical method for two dimensional fractional reaction
subdiffusion equation. Eur. Phys. J. Special Topics 222, 1961-1973.
Huang, F., Liu, F., (2005). The time fractional diffusion equation and the advection-
dispersion equation. ANZIAM J. 46, 317-330.
Ignaczak, J., Ostoja-Starzewski, M., (2010). Thermoelasticity with Finite Wave Speeds.
Oxford University Press, Oxford.
Joseph, D.D., Preziosi, L., (1989). Heat waves. Rev. Modern Phys. 61, 41-73.
Jumarie, G., (1992). A Fokker-Planck equation of fractional order with respect to time.
J. Math. Phys. 33, 3536-3542.
Karatay, I., Bayramoglu, S.R., (2012). An efficient scheme for time fractional advection
dispersion equations. Appl. Math. Sci. 6, 4869-4878.
Kaviany, M., (1995). Principles of Heat Transfer in Porous Media, 2nd ed. Springer, New
York.
Kilbas, A.A., Srivastava, H.M., Trujillo, J.J., (2006). Theory and Applications of Fractional
Differential Equations. Elsevier, Amsterdam.
Liu, F., Anh, V., Turner, I., Zhuang, P., (2003). Time-fractional advection-dispersion equa-
tion. J. Appl. Math. Comput. 13, 233-245.
Lord, H.W., Shulman, Y., (1967). A generalized dynamical theory of thermoelasticity. J.
Mech. Phys. Solids 15, 299-309.
Metzler, R., Klafter, J., (2000). The random walk’s guide to anomalous diffusion: a frac-
tional dynamics approach. Phys. Rep. 339, 1-77.
Metzler, R., Klafter, J., (2004). The restaurant at the end of the random walk: recent
developments in the description of anomalous transport by fractional dynamics. J. Phys.
A: Math. Gen. 37, R161-R208.
Metzler, R., Barkai, E., Klafter, J., (1999). Anomalous diffusion and relaxation close to
thermal equilibrium: a fractional Fokker-Planck equation approach. Phys. Rev. Lett.
82, 3563-3567.
Nield, D.A., Bejan, A., (2006). Convection in Porous Media, 3rd ed. Springer, New York.
Nigmatullin, R.R. (1984a). To the theoretical explanation of the “universal response”. Phys.
Stat. Sol. (b) 123, 739-745.
Nigmatullin, R.R. (1984b). On the theory of relaxation for systems with “remnant” mem-
ory. Phys. Stat. Sol. (b) 124, 389-393.
Nowacki, W., (1974). Dynamical problems of thermodiffusion in solids. Bull. Acad. Polon.
Sci., Sér. Sci. Techn. 23, 55-64, 129-135, 257-266.
Nowacki, W., Olesiak, Z.S., (1991). Thermodiffusion in Solids. Polish Scientific Publishers
(PWN), Warsaw. (In Polish).
Parkus, H., (1959). Instationäre Wärmenspannungen. Springer, Wien.
Podlubny, I., (1999). Fractional Differential Equations. Academic Press, San Diego.
Podstrigach, Ya.S., (1961). Differential equations of thermodiffusion problem in isotropic
deformable solid. Dop. Ukrainian Acad. Sci. 2, 169-172. (In Ukrainian).
Podstrigach, Ya.S., (1963). Differential equations of the diffusive strain theory of a solid.
Dop. Ukrainian Acad. Sci. 3, 336-339. (In Ukrainian).
Podstrigach, Ya.S., (1965). Diffusion theory of inelasticity of metals. J. Appl. Mech. Techn.
Phys. 2, 67-72. (In Russian).
Podstrigach, Ya.S., Povstenko, Y.Z., (1985). Introduction to Mechanics of Surface Phenom-
ena in Deformable Elastic Solids. Naukova Dumka, Kiev. (In Russian).
Povstenko, Y., (2005). Fractional heat conduction equation and associated thermal stresses.
J. Thermal Stresses 28, 83-102.
Povstenko, Y., (2009a). Thermoelasticity which uses fractional heat conduction equation.
J. Math. Sci. 162, 296-305.
Povstenko, Y., (2009b). Theory of thermoelasticity based on the space-time fractional heat
conduction equation. Phys. Scr. T 136, 014017.
Povstenko, Y., (2011). Fractional Cattaneo-type equations and generalized thermoelasticity.
J. Thermal Stresses 34, 97-114.
Povstenko, Y., (2012). Theories of thermal stresses based on space-time-fractional tele-
graph equations. Comp. Math. Appl. 64, 3321-3328.
Povstenko, Y., (2013). Fractional thermoelasticity. In: Encyclopedia of Thermal Stresses,
vol. 4. R.B. Hetnarski (ed.), Springer, Berlin, pp. 1778-1787.
Prudnikov, A.P., Brychkov, Yu.A., Marichev, O.I., (1981). Integrals and Series. Elementary
Functions. Nauka, Moscow. (In Russian).
Prudnikov, A.P., Brychkov, Yu.A., Marichev, O.I., (1983). Integrals and Series. Special
Functions. Nauka, Moscow. (In Russian).
Risken, H., (1989). The Fokker-Planck Equation. Springer, Berlin.
Scheidegger, A.E., (1974). The Physics of Flow Through Porous Media, 3rd ed. University
of Toronto Press, Toronto.
Schneider, W.R., (1990). Fractional diffusion. In: R. Lima, L. Streit and R. Vilela
Mendes (Eds.), Dynamics and Stochastic Processes, Lecture Notes in Physics, vol. 355,
pp. 276–286. Springer, Berlin.
Sherief, H.H., El-Sayed, A.M.A., Abd El-Latief, A.M., (2010). Fractional order theory of
thermoelasticity. Int. J. Solids Struct. 47, 269-275.
Van Kampen, N.G., (1981). Stochastic Processes in Physics and Chemistry. North-
Holland, Amsterdam.
Youssef, H.M., (2010). Theory of fractional order generalized thermoelasticity. J. Heat
Transfer 132, 061301.
Zaslavsky, G.M., (2002). Chaos, fractional kinetics, and anomalous transport. Phys. Rep.
371, 461-580.
Zheng, G.H., Wei, T., (2010). Spectral regularization method for a Cauchy problem of the
time fractional advection-dispersion equation. J. Comput. Appl. Math. 233, 2631-2640.
Chapter 10
MODELLING DRUG EFFECT USING

FRACTIONAL CALCULUS
Clara M. Ionescu*
Ghent University, Department of Electrical Energy,
Systems and Automation, Zwijnaarde, Belgium
Abstract
Closed loop control of depth of anesthesia (DOA) implies a good knowledge of the patient
pharmaco-kinetics and -dynamics, i.e., the availability of a reliable model of the patient. This
is necessary since prediction of drug effect in the body is essentially the main component of
regulating DOA. Hitherto, this is done in a complex clinical environment, involving the
anesthesiologist as a main decision-maker element. Obviously, the decision of administrating
hypnotic and opioid drugs is a difficult one, since over- and under-dosing are a peril for
patient’s wellbeing and recovery. The expertise of the anesthesiologist and human factor make
this decision a subjective one and may be difficult to justify in a mathematical framework.
This chapter introduces emerging tools available on the 'engineering market' imported from
the area of fractional calculus. Drug diffusion compartmental models are introduced and a
novel interpretation of the classical drug-effect curve given. By employing tools from
fractional calculus, model nonlinearity is avoided, allowing linear control strategies for
automatic control of DOA systems.
Keywords: Analgesia, nonlinear dynamics, closed loop anesthesia, multivariable control,

depth of anesthesia, drug effect interaction, fractional order impedance
Introduction
The emerging concepts of fractional calculus (FC) in biology and medicine have shown a
great deal of success, explaining complex phenomena with a startling simplicity (Magin,
2006; Ionescu, 2013). It is clear that a major contribution of the concept of FC has been and
remains still in the field of biology and medicine (West, 1990; Machado et al., 2013).
*
Corresponding Author address: Email: claramihaela.ionescu@ugent.be.
244 Clara M. Ionescu
Fractional calculus generously allows integrals and derivatives to have any order, hence the
generalization of the term fractional-order to that of general-order. Of all applications in
biology, diffusion is the one most appealing from modeling point of view, since it allows
characterizing a relatively complex process with parsimonious models.
Modeling drug dynamics in the body using compartmental models is perhaps one of the
most popular modeling approaches (Jacques, 1985). These models are based on mathematical
characterization of molecular biochemistry and transport phenomena in the body (Macheas &
Iliadis, 2006). As such, diffusion plays an important role in drug assimilation, transport and
clearance, and it is a physiological process, which can be well described by means of
fractional calculus (Trujillo, 2006).
In this chapter, a fresh view upon the models for drug release and their effect will be
given in the application of depth of anesthesia regulation. Although in this chapter only
hypnosis will be discussed, general anesthesia consists of three components acting
simultaneously on the patient’s vital signals, leading to an overall state of a) hypnosis b)
neuromuscular blockade and c) analgesia (i.e., pain relief level).
Hypnosis is a general term indicating unconsciousness and the level of hypnosis is related
to the infusion of hypnotic drug. Hypnosis is relatively well characterized and sensors to
measure it by means of electroencephalogram (EEG) data are currently employed in standard
clinical practice (Liu et al., 2011; Struys et al. 1998) and have been proven useful in closed-
loop control of anesthesia (Ionescu et al. 2008; Ionescu et al. 2011).
Neuromuscular blockade ensures that the patient remains paralyzed during surgical
procedures and is also a relatively well-characterized process with standard sensors available
(EMG-electromyography) (Ozcan et al., 2012). Analgesia represents the loss of pain
sensation that results from a interruption in the nervous system pathway between the sense
organ and the brain. Finally, sedation refers to a combined effect of hypnosis and analgesia
(Kress et al., 2002). From the point of view of detection and measurement, analgesia is the
sole component remaining to be de-mystified.
Figure 1. A personal view upon the high interdisciplinarity required to obtain a comprehensive picture
of the anesthesia paradigm.
Modelling Drug Effect Using Fractional Calculus 245
If one would like to have optimal drug infusion rates into a patient with avoiding over-
and under-dosing, then accurate patient models are necessary (although not sufficient, since a
good control methodology is also required) (Absalom et al., 2011). Accurate models may
hold a realistic dynamic perspective for average populations datasets, but they are by far sub-
optimal when patient-individualized control is envisaged. This is due to the fact that inter-
patient variability has a strong impact on the perception and effectiveness of the drug into the
body of the patient (i.e., drug effect, patient sensitivity to the drug, etc.). A schematic
representation of the anesthesia paradigm in terms of its components is given in Figure 1. The
hypnosis and neuromuscular blockade are well characterized, yet the analgesia component
remains challenging for control purposes because no direct output measure is available.
Hence, no specific models are yet available for analgesia effect in the human body during
general anesthesia (i.e., unconsciousness).
Figure 2. An overview of inputs and outputs in the complex dynamic system of the human body during
surgery and general anesthesia.
The absence of more objective measures and decision-making elements in the DOA
regulating system is justified by the lack of direct measures of hypnosis and analgesia, as
depicted in Figure 2. Hypnosis is a state of (un)consciousness and this is currently measured
via an indirect measure, the electroencephalogram. Analgesia is a state of insensibility to pain
stimuli (mechanical, thermal, etc.) and is measured via indirect measures such as facial stress,
reaction to speech, reaction to electrical stimulation in the hand of the patient, etc. Clearly, of
these two components, analgesia is the most subjective measurement. Efforts are being made
to develop new sensor techniques and new horizons are explored for modeling this intricate
process and to provide mathematical models to replace these subjective evaluations.
The pharmacokinetic (PK) models available in the literature are linear in terms of model
parameters and dynamics (Schnider et al, 1998; 1999). Their frequency response is quasi-
identical, less for a scaling factor in the gain (this accounts in part for the sensitivity to the
drug with respect to the body mass index of the patient). The pharmacodynamic (PD) models
are usually represented by nonlinear Sigmoid (Hill) curves and represent the relationship of
the drug concentration to the drug effect in each patient. From patient-individualized control
point of view, PD models are the most challenging part of the patient model andpose most
challenges for control (i.e., highly nonlinear characteristic).
Fractional calculus offers tools to model such nonlinear characteristics as those of the PD
models with less degree of nonlinearity in model parameters (Ionescu, 2013). The original
purpose of this chapter is to put forward a fresh view to these models and to suggest how they
can be integrated in the control paradigm of depth of anesthesia regulation.
The chapter is structured as follows: next, a brief introduction on the fractional calculus
tools for modeling drug release and its effect is given. Section III depicts the classical
methodology for PK-PD models in anesthesia and section IV presents the proposed PD
modeling approach and some simulation results. Current limitations for the use of these
models in a closed loop control paradigm are given and a conclusion section summarizes the
main outcome of this paper.
Methods
Emerging Tools in the Field of Bio-Engineering
Pure empirical models such as input - output models are rather descriptive with loss of
physiological insight into the drug absorption, transport, diffusion and release. However, from
control point of view they are the most simple to obtain and to deliver information to the
controller for deciding the optimal drug infusion rate necessary for the specific patient whose
model is available. On the other hand, physiologically and bio-chemically based models are
difficult to attain due to the extraordinary complexity of this process. For instance, the
pathway of pain signaling into the body is depicted in Figure 3. It can be observed the various
flows and complex bio-chemical reactions necessary to be modeled.
Figure 3. A diagram of the pathways of pain signaling in the human body.
In general, pain is a complicated process that involves an intricate interaction between a

number of important chemicals found naturally in the brain and spinal cord. These chemicals,
called neurotransmitters, transmit nerve impulses from one cell to another. The complex
process of pain awareness can be divided in three main phases: (i) transduction, (ii)
transmission and (iii) perception. The first phase, transduction, begins when a painful or
noxious stimulus causes tissue damage. The damaged cells release substances that lead to the
generation of an electric action potential. In the next phase (i.e., transmission of the pain
signal) the electric action potential continues from the site of damage to the spinal cord,
where it ascends up the spinal cord to higher centers in the brain. The third phase, perception,
is the conscious awareness of pain. Treatment of pain-related suffering requires knowledge of
how pain signals are initially interpreted and subsequently transmitted and perpetuated.
Therefore, it is mandatory to first understand how an analgesic drug diffuses into the human
body and how we can build models of pain signaling and transmission.
An intermediate solution is perhaps the use of semi-empirical models. Such a fairly
simple model to employ in characterizing drug release is the power-law model:
Mt
 kt n (1)
M
where Mt and M∞ are cumulative amounts of drug released at time t and infinite time,
respectively; and k is a constant reflecting the structural and geometrical characteristics of the
system, and n is the release exponent (Higuchi, 1961). For purely diffusion-controlled release,
n=0.5, while other values of n may be indicative of various diffusion conditions. These
features of this model are of special interest in our study since we would like to look at the
inter-patient variability from this standpoint.
Following the fundamental law of physics which applies to well stirred, homogeneous
systems, it follows that the mean square displacement of the walker, <x2> in the random walk
model is proportional to time (Kopelman, 1988; Rinaki et al., 2003):
 x 2  t (2)
However, in disordered systems, such as most biological environments, this is no longer

proportional with time, but with the fractal walk dimension of the walker:
Dw
 x 2  t (3)
with Dw≠ 2. This property implies that scaling laws such as power laws are associated with
kinetics of various processes taking place in the (biological) environment (i.e., tissue). Most
biological tissues can be approximated in their properties by polymers. As such, the spatio-
temporal porosity of a dynamically changing polymer is close to the percolation threshold for
non-classical diffusion effects impinging on release kinetics.
Fractional calculus has shown in several (non)biological applications that classical
diffusion as well as non-classical diffusion can be characterized by fractional order differ-
integrals models (Machado et al, 2013; Trujillo, 2006). It is possible to derive further the
model from (1) in its Laplace equivalent, as indicated in (Magin, 2006). This result has
regained the attention of the research community and current efforts are being directed
towards providing pharmacokinetic models with fractional order differ-integrals

(Dokoumetzidis and Macheras, 2009; Kytariolos et al, 2010; Dokoumetzidis et al., 2010;
Popovic et al., 2011; Hennion and Hanert, 2013; Popovic et al, 2013; Copot et al., 2013;
Ionescu et al., 2013), and their equivalent Laplace models of non-integer order, coined in the
literature as FOIMs (fractional order impedance models) (Ionescu, 2013).
Classical PK-PD Models for Drug Diffusion
In order to control the depth of sedation based on predictive control strategies, a model
capturing the dynamics of the patient is required. The selection of the model’s input and
output variables is crucial for optimal control. Propofol is a commonly used hypnotic drug to
induce general anesthesia (Absalom et al., 2011; Bailey & Haddad, 2005) and is also used to
avoid awareness of pain. Sometimes is used together with Remifentanil, a strong opioid, to
emphasize analgesia (i.e., pain relief). The two drugs have a synergic effect, obeying exactly
the same behavior as that discussed in this section. For sake of simplicity, only the effects of
one drug (i.e., Propofol) will be considered hereafter.
Using the electroencephalogram (EEG), several derived, computerized parameters like
the BIS have been tested and validated as a promising measure of the hypnotic component of
anesthesia. BIS values lie in the range of 0-100; whereas 90-100 range represents fully awake
patients; 60-70 range and 40-60 range indicate light and moderate hypnotic state, respectively
(Struys et al., 2003).
PK-PD blocks denote compartmental models. The PK-PD models most commonly used
for propofol and remifentanil are the 4th order compartmental model described by Schnider et
al., (1998, 1999) and they have the structure depicted in Figure 4:
Figure 4. General compartmental model of the patient, where PK denotes the pharmacokinetic model
and PD denotes the pharmacodynamic model of an infused drug.
In this figure x1 [mg] denotes the amount of drug in the central compartment. The blood
concentration is expressed by x1/V1. The peripheral compartments 2 and 3 model the drug
exchange of the blood with well and poorly diffused body tissues. The masses of drug in fast
and slow equilibrating peripheral compartments are denoted by x2 and x3, respectively. The
parameters kji, for i≠j, denote the drug transfer frequency from the jth to the ith compartment
and u(t) [mg/s] is the infusion rate of the anesthetic drug into the central compartment. An
additional hypothetical effect compartment was proposed to represent the lag between drug
plasma concentration and drug response. The parameters kij of the PK models depend on age,
weight, height and gender and the relations can be found in (Schnider 1998; 1999). The PK-
PD model is represented by the following equations:
u (t )
x1 (t )    k10  k12  k13  x1 (t )  k21 x2 (t )  k31 x3 (t ) 
V1
x2 (t )  k12  x1 (t )  k21  x2 (t ) (4)
x3 (t )  k13  x1 (t )  k31  x3 (t )
xe (t )  ke 0  Ce (t )  k1e  x1 (t )
The effect compartment receives drug from the central compartment by a first-order
process and it is regarded as a volumeless additional compartment. Therefore, the drug
transfer frequency from the central compartment to the effect-site compartment is equal to the
frequency of drug removal from the effect-site compartment: ke0=k1e=0.456 [min-1] (Struys et
al. 2003). Drug concentration in the effect site compartment is denoted by the variable Ce.
The parameters kij of the PK models depend on age, weight, height and gender and can be
calculated for Propofol:
V1  4.27 l  ; V3  2.38 l  ; V2  18.9  0.391 (age  53)  l 

Cl1  1.89  0.0456(weight  77)  0.0681(lbm  59)  0.0264(height 177)  l / min 
Cl 2  1.29  0.024(age  53) l / min  ; Cl 3  0.836 l / min 
C C C
k10  l1  min 1  ; k12  l 2  min 1  ; k13  l 3  min 1  ;
V1   V 1   V1  
Cl 2  C
k21  min 1  ; k31  l 3  min 1 
V2   V3  
where Cl1 is the rate at which the drug is cleared from the body, and Cl2 and Cl3 are the rates
at which the drug is removed from the central compartment to the other two compartments by
weight 2
distribution. The lean body mass (lbm) for women and men are: 1.1 weight  128  and
height 2
weight 2
1.07  weight  148  , respectively.
height 2
The relation between the effect site concentration Ce and the BIS is given by a nonlinear
sigmoid Hill curve:
Ce (t ) (5)
BIS (t )  E0  Emax 
Ce (t )  C50

where E0 is the BIS value when the patient is awake; Emax is the maximum effect that can be
achieved by the infusion of Propofol; C50 is the Propofol concentration at half of the
maximum effect and γ is a parameter which together with the C50 determines the patient
sensitivity to the drug. Usually, E0 and Emax are usually considered equal to 100.
As explained earlier, the main challenge for control standpoint is the nonlinearity of the
Hill curve given by (5) and the inherent inter-patient variability. To illustrate this to the
reader, a realistic set of typical and atypical patients has been used, with the parameter values
given in Table 1. The resulted Hill curves by simulating the PKPD model of these patients is
shown in Figure 5.
Figure 5. Example of Hill curves from various patients illustrating the high-degree of interpatient
variability.
For instance, Patient 3 and patient 9 require the same amount of effect site concentration
before they start to react, but they have significantly different sensitivity to the drug effect.
On the other hand, patient 6 requires less effect site concentrations to initiate the reaction in
BIS, but the response is very slow. Moreover, it can be generalized that below a certain
threshold value for the effect site concentration, the BIS(t) does not vary significantly. Such a
threshold interval is used in practice to apply target-controlled infusion, which is in fact an
open loop control policy. This specific intial strategy provides a desired effect site
concentration in the patient, but does not take into account the actual measured BIS.
Obviously, this is only useful in the induction phase of anesthesia, i.e., until a certain ΔBIS is
achieved.
Table 1. Characteristic variables for each of the virtual realistic 12 patients used in this
study as from (Ionescu et al, 2008)
It is obvious that all these differences in reactions from patients to the same amount of
drug infused in their body require a high degree of robustness from the controller. The
drawback is a conservative control, which may not always deliver the expected performance,
e.g., when disturbances need to be removed fast to avoid awareness in the patient.
Consequently, a significant strain is put on the controller’s task to keep the closed loop
performance within the same specifications for all types of patients.
Hybrid Models emerging from Classical PK-PD and FOIMs
The interwoved thread connecting these models is based on the theory of fractional
calculus and fractal walk dynamics (West, 1990; Magin, 2006). An interesting feature of
systems characterized by fractal dynamics is the following: represented on a log-log plot,
their characteristic becomes linear. This feature is of special interest for us, especially if the
reader may recall the Hill curve from Figure 5. Of course, it would be very nice if the strong
nonlinearity and inter-patient variability depicted in this figure would be diminished, or even
better, simply vanish into thin air. In this section, some ideas will be put forward which will
lead the reader through this new approach of viewing Hill curves from a different perspective.
Based on (1) one can write the same relationship for the Hill curve:
Ce (t)
= k × tn (6)
BIS(t)
where k and n are varying on the patient PK-PD characteristics. If one compares (5) with (6)
it can be recognized the resemblance in the power term and observe in fact a simplification of
the model from (5) in terms of parameter number. From a structural point of view, there is no
difference between the models, since both are semi-empirical models. The term Ce(t)/BIS(t)
denotes the concentration-to-effect ratio (CER) and its units are [mg/ml/%].
The profile of Ce(t) may differ depending on the type of depth of anesthesia regulation
that can occur in practice. Usually, in target-controlled infusion systems, drug is infused in
the patient in open loop (i.e., no feedback from the patient is explicitly taken into account!)
and is the most widely applied in clinical practice. Averaged population models are used to
predict the concentration in the blood and thus in the effect site compartment from (4) and
their corresponding BIS effect. If one represents this in time, it appears as a train of steps of
different values; i.e., the anesthesiologist changes the value of the targetted concentration
depending on the reactions in biological signals from patient (e.g., heart rate, respiratory rate,
blood pressure, cardiac output, oxigen concentration, etc.).
In order to verify the validity of the above assumptions, a simulation study has been
performed. Values for the effect site concentration Ce(t) have been given in the range from
0.01-20, linear distribution of 2000 points. BIS(t) has been calculated with parameters from
Table 1 and formula from (5). The purpose of k is that of a scaling factor, so a common value
has been identified for all patients: 5.45-10. The identified values for the parameter n are
illustrated in Figure 4. From Figure 5 one may recall the specificity of patients 3, 6 and 9.
Their corresponding values are n=3.88, n=2.72 and n=4.45, respectively.
The results of the model identification from (5) are given in figures 6-11. Notice that
somewhat less accurate model is identified for patient 9 and patient 10. Looking back in
Table 1, these patients have the highest, respectively the lowest values for the Gamma
parameter, which give the slopes of the Hill curve in Figure 5. Correlating this with the results
of the identification from Figure 12, we observe that also the highest, respectively the lowest
value for the n parameter is obtained for the same patients. We conclude that the n parameter
holds a similar significance in (6) as that of the Gamma parameter in (5).
Figure 6. Illustration of identification result for Patient 1 and Patient 2 from Table 1.
Discussion
It may be possible to replace (5) by some form (6), since the fractional order impedance
can account for time-varying dynamics by having a variable fractional order value n. This
idea is a good incentive for further thought.
It is obvious that i) closed loop control is necessary and ii) an averaged patient model will
deliver sub-optimal results since they require a robust, conservative controller able to deal
with the inter-patient variability. Moreover, it is also clear that the inter-patient variability
may vary quite significantly and that one single controller (without online adaptation) will
never suffice in practice.
A first problem is to find online adaptation algorithms which may adapt the model
parameters k and n to the patient characteristics. Identification from logarithmically sampled
data has been proposed in (De Keyser et al., 2011) and represents a good framework for
developing the online identification algorithm.
Another problem is to find a solution to integrate the model from (6) into a closed loop
control system taking into account the requirement for a logarithmic sample time (i.e., in
order to maintain linearity). It has already been shown in various examples that a Riemann
sampling rate (i.e., linear periodic) may be outperformed by a Lebesque sampling rate (i.e.,
event-triggered) in several applications (Goodwin et al., 2013). Since the Lebesque sampling
rate is an event trigerred rate used successfully in practice in closed loop control (e.g.,
networked control, sensor networks), it may be revealing to look into the possibility of a
logarithmic sampling rate.
Figure 12. Identified values of the n parameter from the power law model in (6) for all patients given in
Table 1.
Conclusion
This chapter presented the available tools emerging from fractional calculus to model the
nonlinear characteristics of the pharmacokinetic and pharmacodynamic patient models.
Advantages and challenges have been discussed. Results suggest that the high degree of inter-
patient variability and nonlinearity may be avoided, leading to linear control techniques
instead of advanced, complex control techniques.
This fresh view upon the patient models enables further optimization of the multiple
input (i.e., two drugs) single output (i.e., their synergic effect) until the analgesia sensor
becomes available for a fully multivariable input-output control of anesthesia.
Acknowledgment
The author acknowledges the financial support of the Research Foundation Flanders
(FWO).
References
Absalom, A., De Keyser, R., Struys, M., (2011). Closed Loop Anesthesia: Are We Getting
Close to Finding the Holy Grail? Anesthesia and Analgesia, 112(3), 364-381.
Bailey, J.M., Haddad, W.M., (2005). Drug dosing control in clinical pharmacology, IEEE
Control Systems Magazine, 25(2), 35-51.
Copot, D., Chevalier, A., Ionescu, C., De Keyser, R., (2013). A two-compartment fractional
derivative model for propofol diffusion in anesthesia, Proc. Conference on Systems and
Control, Hyderabad, India, 28-31 August, 264-269.
De Keyser, R., Ionescu, C., Festila, C., (2011). A one-step procedure for frequency response
estimation based on switch-mode transfer function analyzer, Proc. IEEE-CDC Conf on
Decision and Control joint with ECC-European Control Conf., December 12-15,
Orlando, Florida, 1189-1194.
Dokoumetzidis A., Macheras P., (2009). Fractional kinetics in drug absorption and
disposition processes, J. Pharmacokinet Pharmacodyn., 36, 165-178.
Dokoumetzidis A., Magin R., Macheras P., (2010), Fractional kinetics in multi-
compartmental systems, J. Pharmacokinet. Pharmacodyn., 37, 507-524.
Goodwin, G., Aguero, J.C., Cea-Garrido, M.E., Salgado, M.E., Yuz, J.I., (2013). Sampling
and sampled-data models, IEEE Control Systems Magazine, 33(5), 34-53.
Hennion M., Hanert E. (2013), How to avoid unbounded drug accumulation with fractional
pharmacokinetics, J. Pharmacokinet Pharmacodyn, 40, 691-700.
Higuchi, T., (1961). Rate of release of medicaments from ointment bases containing drugs in
suspension, J. Pharm Sci, 50, 874-875.
Ionescu, C., De Keyser, R., Torrico, B., De Smet, T., Struys M., Normey-Rico, J., (2008).
Robust Predictive Control Strategy Applied for Propofol Dosing using BIS as a
Controlled Variable during Anesthesia, IEEE Transactions on Biomedical Engineering,
55, 2161-2170.
Ionescu C., Hodrea R., De Keyser R., (2011), Variable time-delay estimation for anesthesia
control during intensive care, IEEE Transaction on Biomedical Engineering, 58(2), 363-369.
Ionescu, C., (2013). The human respiratory system: an analysis of the interplay between
anatomy, structure, breathing and fractal dynamics Springer, Series in BioEngineering
2013, Print ISBN 978-1-4471-5387-0, On- line ISBN 978-1-4471-5388-7.
Ionescu, C.M., Copot, D., De Keyser, R., (2013). Three compartmental model for propofol
diffusion during general anesthesia, Discontinuity, Nonlinearity and Complexity, 2(4),
357-368.
Jacques, J.A., (1985). Compartmental analysis in biology and medicine, University of
Michigan Press.
Kopelman, R., (1988). Fractal reaction kinetics, Science, 241, 1620-1626.
Kress JP, Pohlman AS, Hall JB, (2002), Sedation and analgesia in the intensive care unit,
American Journal of Respiratory Critical Care Medicine; 166, 1024-8.
Kytariolos J., Dokoumetzidis A., Macheras P., (2010). Power law IVIVC: an application of
fractional kinetics for drug release and absorption, Eur. J. of Pharmaceutica.l Sci., 41,
299-304.
Liu N, Chazot T, Genty A, Landais A, Restoux A, McGee K, Laloe PA, Trillat B, Barvais L,
Fischler M, (2011), Closed-loop co-administration of Propofol and Remifentanil guided
by Bispectral index: a randomized multicenter study, Anesthesia and Analgesia, 112(3),

546-557.
Macheas, P., Iliadis, A., (2006). Modeling in Biopharmaceutics, Pharmacokinetics, and
Phamacodynamics: Homogeneous and Heterogeneous Approaches. Springer Verlag,
New York.
Magin, R. (2006). Fractional Calculus in Bioengineering. Begell House Publishers.
Ozcan A, Ozcan N, Gulec H, Yalcin F, Basar H, (2012), Comparison of the effects of
fentanyl, remifentanil, and dexmedetomidine on neuromuscular blockade, Journal of
Anesthesia, 26(2), 196-199.
Popovic J., Dolicanin D., Rapaic M., Popovic S., Pilipovic S., Atanackovic T., (2011), A
nonlinear two compartmental fractional derivative model, Eur. J. Drug Metab.
Pharmacokinet., 36, 189-196.
Popovic J., Posa M., Popovic K., Popovic D., Milosevic N., Tepavcevic V., (2013),
Inidividualization of a pharmacokinetic model by fractional and nonlinear fit
improvement, Eur. J. Drug Metab. Pharmacokinet., 38, 69-76.
Rinaki, E., Valsami, G., Macheras, P., (2003). The power law can describe the ’entire’ drug
release curve from HPMC-based matrix tablets: a hypothesis, Int. J. of Pharmaceutics,
255, 199-207.
Schnider, T.W., Minto, C.F., Gambus, P.L. , Andresen, C, Goodale, DB , Youngs, EJ, (1998).
The influence of method of administration and covariates on the pharmacokinetics of
Propofol in adult volunteers, Anesthesiology, 88: 1170-1182.
Schnider, TW., Minto, CF, Shafer, SL, Gambus, PL, Andresen, C, Goodale, DB, Youngs, EJ,
(1999). The influence of age on Propofol pharmacodynamics, Anesthesiology, 90:1502-1516.
Struys M, Versichelen L, Byttebier G, Mortier E, Moerman A, Rolly G, (1998), Clinical
usefulness of the bispectral index for titrating propofol target effect-site concentration,
Anaesthesia, 53, 4-12.
Struys, M., Vereecke, H., Moerman, A., Jensen, E., Verhaeghen, D., De Neve, N., Dumortier
F., Mortier E., (2003). Ability of the bispectral index, autoregressive modelling with
exogenous input-derived auditory evoked potentials, and predicted propofol
concentrations to measure patient responsiveness during anesthesia with propofol and
remifentanil, Anesthesiology, 99, 802-814.
Tenreiro Machado, J.A., Kiryakova, V., Mainardi, F., (2013). Recent history of fractional
calculus, Communications in Nonlinear Science and Numerical Simulation, 16(3), 1140-
1153.
Trujillo, J.J., (2006). Fractional models: Sub and super-diffusives, and undifferentiable
solutions, Innovation in Engineering Computational Technology, pp. 371-401.
West, B.J., (1990). Fractal physiology and chaos in medicine, Studies of nonlinear
phenomena in life sciences vol.1. Singapore: World Scientific.
Chapter 11
F UZZY F RACTIONAL PID C ONTROLLERS :

A NALYSIS , S YNTHESIS AND I MPLEMENTATION
Ramiro S. Barbosa∗and Isabel S. Jesus†
GECAD − Knowledge Engineering and Decision Support Research Center
Institute of Engineering / Polytechnic of Porto (ISEP/IPP)
Dept. of Electrical Engineering, Porto, Portugal
Abstract
Fuzzy logic controllers (FLC) have developed greatly in recent years. They offer easy
and robust solutions to complex problems, allowing human reasoning to be applied
to the control of systems. This chapter introduces the concepts of fractional calcu-
lus in FLC. The resulting fuzzy fractional PID controllers are investigated in terms of
their structures and respective digital implementations. In the first part of chapter, a
simple and effective tuning methodology is proposed and compared with traditional
approaches. The methodology for tuning the fuzzy controllers is based on the prior
knowledge of integer/fractional-order control strategy, making the procedure adequate
to replace an existent controller in order to improve the system control performance.
In the second part of chapter, are devised optimal fuzzy fractional PID controllers by
using a particle swarm optimization algorithm. A comparative study between fuzzy
integer and fractional PID controllers is presented. The feasibility and effectiveness
of the proposed controllers are tested on several benchmark systems that are represen-
tative of industrial processes. The simulation results show the better performance of
fractional controllers over the conventional PID or fuzzy PID controllers.
PACS: 05.45-a, 52.35.Mw, 96.50.Fm
Keywords: fuzzy control, PID controller, fractional PID control, fractional calculus, fuzzy
fractional PID control, optimization
1. Introduction
In recent years, fractional-order (FO) PID controllers have been a fruitful field of re-
search [3, 50, 51]. However, no effective and simple tuning rules still exist for these con-
∗
E-mail address: rsb@isep.ipp.pt
†
E-mail address: isj@isep.ipp.pt
260 Ramiro S. Barbosa and Isabel S. Jesus
trollers as those given for the integer PID controllers [2]. The FO-PID controller involves
an integrator/differentiator of real, irrational or, even, of complex order [11, 50]. It has
been demonstrated the good performance of this type of controller, in comparison with the
conventional PID algorithms. The FO-PID extends the capabilities of the classical counter-
part and, thus, have a wider domain of application, such as in suspension systems, robotics,
signal processing, control and diffusion [44, 50, 51, 33]. On the other hand, the FLC have
also been successfully applied in the control of many physical systems, particularly those
with uncertainty, unmodelled, disturbed and/or nonlinear dynamics [36, 37, 18, 27].
Fuzzy control emerged on the foundations of Zadeh’s fuzzy set theory [43, 12, 22].
This kind of control is based on the ability of a human being to find solutions for partic-
ular problematic situations. It is well know from our experience, that humans have the
ability to simultaneously process a large amount of information and make effective de-
cisions, although neither input information nor consequent actions are precisely defined.
Through multivalent fuzzy logic, linguistic expressions in antecedent and consequent parts
of IF-THEN rules describing the operator’s actions can be efficaciously converted into a
fully-structured control algorithm. The fuzzy method offer a systematic procedure to de-
sign controllers for many kind of systems, that often leads to a better performance than that
of the conventional PID controller. It is a methodology of intelligent control that mimics
human thinking and reacting by using a multivalent fuzzy logic and elements of artificial
intelligence. Many applications using FLC can be found in [40, 36, 48, 53, 27].
It has been proved that the use of the fuzzy fractional controllers improved the results
for many kind of systems, since it gives additional flexibility to the design. In this line of
thought, many applications using this type of controllers were developed in the last few
years [6, 3, 28, 4, 5, 17, 16, 47, 29, 30]. In [6, 3], the authors proposed the use of fuzzy
fractional PD and PID controllers in the control of integer plants. They proved their effec-
tiveness and superior robustness when compared with conventional integer or fuzzy PID
controllers. In [4, 5] the fuzzy FO-PID controllers are tested on several fractional-order
plants showing their superior control system performance. In [29, 30] are devised optimal
fuzzy FO-PID controlers for nonlinear control systems. The above studies are devoted to
discrete fractional-order PID controllers and their digital implementation. Other studies
using continuous fuzzy FO-PID controllers are found in [18, 17, 16, 47].
In the first part of chapter, are combined the features of fuzzy controllers with those of
fractional controllers of PID-type. The resulting fuzzy FO-PID controller is investigated
in terms of its digital implementation and robustness. The combined advantages of the
two controllers results in a better controller with superior robustness and wider domain of
application. The tuning methodology of these controllers is based on the prior knowledge of
integer or fractional-order control. First, a fractional-order controller is built and tuned (or
used one already implemented). Then, we replace it with a linear fuzzy fractional controller
displaying exactly the same step response. After, we make the controller nonlinear and fine
tune it in order to get better control of the system. The fuzzy fractional controller will give,
at least, the same performance of its linear counterpart.
In the second part of chapter, are designed optimal fuzzy FO-PID controllers by us-
ing particle swarm optimization (PSO) algorithm. PSO is one of the latest evolutionary
techniques inspired by social bahavior of bird flocking or fish schooling [20, 21, 35]. A
fractional nonlinear control system with saturation in actuator is analyzed in terms of step
Fuzzy Fractional PID Controllers: Analysis, Synthesis and Implementation 261
responses, variation of controller parameters and time domain specifications as the frac-
tional order of the fuzzy controller is changed. The influence of the order of the discrete
rational approximations in control system performance is also studied.
The chapter is organized as follows. Section 2 presents the fundamentals of continuous
and discrete fractional-order PID controllers. Section 3 outlines a tuning methodology for
the design of fuzzy FO-PID controllers. Two decomposed structures for fuzzy FO-PID
controllers are presented. In section 4, are devised optimal fuzzy FO-PID controllers using
the PSO algorithm. The proposed fuzzy FO-PID controllers are tested and assessed their
effectiveness and robustness in the control of integer and fractional-order plants commonly
found in industry. Finally, section 5 draws the main conclusions.
2. Basics of Fractional PID Controllers

The fractional-order controller of PID-type, usually named PIλ Dµ controller, may be
given as [51, 3]:
U (s) Ki
C (s) = = Kp + λ + Kd sµ (1)
E (s) s
where λ and µ are the orders of the fractional integrator and differentiator, respectively. The
parameters Kp , Ki and Kd are correspondingly the proportional, integral and derivative
gains of the controller. Clearly, taking (λ, µ) ≡ {(1, 1) , (1, 0) , (0, 1) , (0, 0)} we get
the classical {PID, PI, PD, P}-controllers, respectively. Other PID controllers are possible,
namely: PDβ controller, PIα controller, PIDβ controller, and so on. The PIλ Dµ -controller is
more flexible and gives the possibility of adjusting more carefully the dynamical proprieties
of a control system [51, 6].
The time domain equation of the PIλ Dµ controller is:
u (t) = Kp e (t) + KiD −λ e (t) + Kd D µ e (t) (2)
where D (∗) (≡ 0 Dtα ) denotes the differential operator of integration and differentiation
(differintegral) to a fractional-order α = {−λ, µ} ∈ <.
The two most commonly used definitions for the differintegral are the Riemann-
Liouville definition and the Grünwald-Letnikov definition [44, 42]. The Riemann-Liouville
definition of a fractional derivative is (α > 0):
1 dn t f (τ )
Z
α
D f (t) = dτ (3)
Γ (n − α) dt 0 (t − τ )α−n+1
n
where n − 1 < α < n, n is an integer, f (t) is the applied function, and Γ(x) represents the
Gamma function of x.
The Grünwald-Letnikov definition is (α ∈ <):
[t/h]
α 1 X j α
D f (t) = lim α (−1) f (t − jh) (4a)
h→0 h j
j=0

α Γ (α + 1)
= (4b)
j Γ (j + 1) Γ (α − j + 1)
where h is the time increment and [v] means the integer part of v.
The ”memory” effect of these operators is demonstrated by above equations, where the
convolution integral in (3) and the infinite series in (4), reveal the unlimited memory of
these operators, ideal for modeling hereditary and memory properties in physical systems
and materials.
The fractional integrals and derivatives can also be defined in the s-domain. Consider-
ing null initial conditions, they are given by the simple form (α ∈ <):
D α f (t) = L−1 {sα F (s)} (5)
where L represents the Laplace operator and F (s) = L {f (t)}. Definition (5) gives an
easily interpretation of the fractional-order operator in the frequency domain. In fact, the
open-loop Bode diagrams of amplitude and phase of sα have correspondingly a slope of
20α dB/dec and a constant phase positionated at απ/2 rad over the entire frequency domain
[46, 7, 15].
From a control (and signal) processing perspective, approach (4) seems to be the most
useful and intuitive, particularly for a discrete-time implementation [10, 39]. Thus, using
(4), a discrete fractional PIλ Dµ control equation can be obtained from (2) as (h ≈ T, T is
the sampling period):
u (k) = Kp e (k) + KiD −λ e (k) + Kd D µ e (k) (6)
with
k
α 1 X j α
D e (k) ≈ α (−1) e (k − j) (7)
T j
j=0
Replacing (7) in (6), the discrete PIλ Dµ controller is then expressed as:
k k
Ki X j −λ Kd X j µ
u (k) = Kp e (k) + −λ (−1) e (k − j) + µ (−1) e (k − j) (8)
T j T j=0 j
j=0
The control equation (8) shows that the current value of control signal u (k) depends
on all previous values of error e (k), making the computation too heavy as time increases
and so unsuitable for a practical implementation of these algorithms. This fact illustrates
the global character (i.e., unlimited memory) of the fractional-order operators. For practical
implementation of fractional integral and derivative (7) we often apply the ”short memory
principle” [50], resulting in expression:
k k
Ki X (−λ) Kd X (µ)
u (k) = Kp e (k) + −λ cj e (k − j) + µ cj e (k − j) (9)
T T
j=v j=v
where v = 0 for
k < L/T or v = k − L/T for k > L/T ; L is the memory length and
(α) α
cj = (−1)j are the binomial coefficients which may be calculated recursively as:
j

(α) (α) 1 + α (α)
c0 = 1; cj = 1 − cj−1 , j = 1, 2, · · · (10)
j
l(t)
r(t) + e(t) Fuzzy u(t) + y(t)
Fractional G(s)
+
− Controller
Figure 1. Fuzzy fractional PID controlled system.
Note that (9) is given in the form of a FIR filter [24]. Other discrete-time approxima-
tions in the form of IIR filters are also possible [54, 8, 9, 10, 13, 14, 45]. In section 4 is
presented a method to obtain digital rational approximations of fractional-order derivatives
and integrals.
3. Tuning Methodology for Fuzzy Fractional PID Controllers

In this section, we will apply fuzzy logic control (FLC) for the design of fuzzy FO-PID
controlled systems [36, 48]. Despite of variety of possible fuzzy controller structures, the
controller is usually arranged in cascade with the system being controlled. This type of
arrangement is shown in Figure 1 and will be used in this study. Other fuzzy schemes may
be adopted such as fuzzy supervisory control and adaptive fuzzy control [48, 40, 26].
The main idea here is to explore the fact that the FLC, under certain conditions, is equiv-
alent to a PID controller [43, 37, 26] and to extrapolate this fact for the case of fractional-
order controllers. In a certain sense, the fuzzy PID controllers are a special case of the more
general fuzzy FO-PID controllers, in which are involved two extra tuning parameters: the
fractional orders (λ, µ) of controller equation (6).
The basic form of a fuzzy controller is illustrated in Figure 2 [48], consisting of a
fuzzy rule base, an inference mechanism, and fuzzification and defuzzification interfaces.
In general, the mapping between the inputs and the outputs of a fuzzy system is nonlinear
[23, 26]. However, it is possible to construct a rule base with a linear input-output mapping
[43, 26]. For that, the following conditions must be fulfilled:
• Use triangular input sets that cross at the membership value 0.5
• The rule base must be complete AND combination (cartesian product) of all input
families
• Use the algebraic product (*) for the AND connective
• Use output singletons, positioned at the sum of the peak positions of the input sets
• Use sum-accumulation and centre of gravity for singletons (COGS) defuzzification
It seems reasonable to start with the design of a conventional integer/fractional PID

controller and from there to proceed to a fuzzy control design. In this way, the linear fuzzy
Figure 2. Structure of a fuzzy controller.
controller may be used in a design procedure based on integer/fractional PID control, as

follows [26, 6, 3]:
1. Build and tune an integer/fractional PID controller
2. Replace it with an equivalent linear fuzzy controller
3. Make the fuzzy controller nonlinear
4. Fine-tune it
With the above procedure, the design of fuzzy FO-PID controllers will be greatly sim-
plified, particularly if the controller was already implemented and it is desirable to enhance
its performance. Moreover, this new type of controllers extends the potentialities of both
fuzzy and fractional controllers and performs, at least, as well as its linear counterpart
[26, 6, 3].
3.1. Fuzzy Fractional PD Controller

The time-domain equation of a fractional PDµ -controller is given by (Ki = 0 in (2)):
u (t) = Kpe (t) + Kd D µ e (t) (11)
The corresponding discrete-time fractional PDµ -controller is:
u (k) = Kpe (k) + Kd D µ e (k) (12)
Figure 3 illustrates the block diagram of the fuzzy fractional PDµ (FF-PDµ ) controller.
As can be seen, the controller acts on the error, e(k), and on the fractional change of error,
e E
Ke
u U
FF-PDµ
Ku
FE Rule base
µ
D Kfe
Figure 3. Fuzzy fractional PDµ -controller.
D µ e(k). The control signal is U . The controller has three tuning gains, Ke and Kf e ,
corresponding to the inputs and Ku to the output. The following relations are thus verified:
E = Ke e, F E = Kf e D µ e, U = Ku u (13)
where E and F E represent the terms of error and fractional change of error, respectively.
The control signal U is generally a nonlinear function of E and FE:
U = f (E, F E)Ku = f (Ke e, Kf e D µ e) Ku (14)
With a proper choice of design, a linear approximation can be obtained as:
f (Ke e (k) , Kf e D µ e (k)) ≈ Ke e (k) + Kf eD µ e (k) (15)
and
U (k) = (Ke e (k) + Kf eD µ e (k)) Ku = Ke Ku e (k) + Kf e Ku D µ e (k) (16)
Comparing (16) with (12), it yields the relation between the gains of the conventional
and fuzzy PDµ controllers:
Ke Ku = Kp
Kf e Ku = Kd (17)
The linear FF-PDµ -controller provides all the advantages of the conventional fractional
PDµ -controller.
For an equivalent linear FF-PDµ -controller, the conclusion universe should be the sum
of the premise universes and the input-output mapping should be linear. Table 1 lists a
linear rule base for the FF-PDµ controller composed of four rules. There are only two fuzzy
labels (Negative and Positive) used for the fuzzy input variables and three fuzzy labels
(Negative, Zero and Positive) for the fuzzy output variable. This rule base should satisfy
above mentioned conditions in order to provide a linear mapping.
Scaling the input gains may be necessary to preserve the linearity of the fuzzy controller.
However, that should be made without affecting the tuning [6, 3]. For example, considering
(16), and a scale factor M (M > 0):
Ku
U (k) = (Ke e (k) + Kf eD µ e (k)) Ku = (M Ke e (k) + M Kf e D µ e (k)) (18)
M
This scaling has some advantages, as it will avoid saturation and will provide a simpler
design, since the universes ranges of inputs and outputs are normalized to a prescribed
interval, say percentage of full scale [−100, 100].
Table 1. Rule base for the FF-PDµ controller.
Rule 1 If E is N and F E is N then u is N

Rule 2 If E is N and F E is P then u is Z
Rule 3 If E is P and F E is N then u is Z
Rule 4 If E is P and F E is P then u is P
3.2. Fuzzy Fractional PID Controller

The inclusion of an integral action is necessary whenever the closed-loop system ex-
hibits a steady-state error. The fuzzy fractional PDµ +Iλ (FF-PDµ +Iλ ) controller combine
the fractional-order integral action with a fuzzy PDµ -controller, as illustrated in Figure 4.
The control signal U is generally a nonlinear function of error E, fractional change of
error FE, and fractional integral of error FIE:

µ −λ
U = (f (E, F E) + F IE) Ku = f (Ke e (k) , Kf eD e (k)) + Kf ie D e (k) Ku
(19)
Adopting the linear approximation (15) it yields the discrete control equation:

U (k) ≈ Ke e (k) + Kf e D µ e (k) + Kf ie D −λ e (k) Ku
= Ku Ke e (k) + Ku Kf e D µ e (k) + Ku Kf ie D −λ e (k) (20)
Comparing (20) with the discrete fractional PIλ Dµ -controller (6), it yields the relation
between the gains of the conventional and fuzzy fractional PID controllers:
Ke Ku = Kp
Kf ie Ku = Ki
Kf e Ku = Kd (21)
The linear FF-PDµ +Iλ controller provides all the advantages of the conventional frac-
tional PIλ Dµ -controller.
3.3. Illustrative Examples

In this section we apply the proposed fuzzy FO-PID controllers in the control of integer
and fractional-order plants. In all experiments, the simulation parameters are: absolute
memory computation of approximation (7), scale factor M = 0.1, and sampling period
T = 0.05 s.
3.3.1. Example 1: Integer-Order Plant

Consider the normalized transfer function of a double integrator plant, which serves as
a model of many dynamic systems:
1
G1 (s) = (22)
s2
e E
Ke
u + U
FF-PDµ
Ku
FE Rule base
Dµ Kfe +
FIE
D−λ Kfie
Figure 4. Fuzzy fractional PDµ + Iλ controller.
For the control of plant (22) a fractional PDµ controller was designed in [6]:
CP Dµ (s) = KP + KD sµ = 0.5 + 0.5s0.7 (23)
Let us design an equivalent linear FF-PDµ -controller. By configuring the fuzzy infer-
ence system (FIS) and selecting three scaling factors, we obtain a FF-PDµ -controller that
reproduces the exact control performance as the fractional PDµ -controller. We first fix
Ke = 100, since the error universe is chosen to be percentage of full scale [−100, 100],
and the maximum error to a unit step is 1. The values of Kf e and Ku are obtained using
expressions (17). Figure 5 shows the input families and the linear control surface obtained
by using the rule base of Table 1 while satisfying the conditions of linear mapping; E is
the error, F E is the fractional derivative of error and u is the output of the fuzzy PDµ con-
troller. Note that this result represents the step 2 – replace the conventional controller with
an equivalent linear fuzzy controller – of the design procedure.
In order to enhance the performance of the control system we now proceed to step 3
of the design – make the fuzzy controller nonlinear. Thus, after verifying that the linear
FF-PDµ -controller is properly designed, we may adjust the FIS settings such as its style,
membership functions and rule base to obtain a desired nonlinear control surface. In this
work, we choose to change the fuzzy rule base, as illustrated in Table 2. This rule base is
commonly used in fuzzy control systems and consists of 49 rules with 7 linguistic terms
(NL – Negative large, NM – Negative medium, NS – Negative small, ZR - Zero, PS –
Positive small, PM – Positive medium and PL – Positive large). The membership functions
for the premises and consequents of the rules are shown in Figure 6a). With two inputs
and one output the input-output mapping of the fuzzy logic controller is described by the
nonlinear surface of Figure 6b). For the defuzzication process we use the centroid method.
Figure 7 shows the step responses of both linear and nonlinear fuzzy PDµ (µ = 0.7)
controllers. We verify that the nonlinear controller improves the system control perfor-
mance, namely the overshoot and settling time.
For comparison purposes, we also compute the integral of the absolute error (IAE):
Z t
IAE = |e (t)|dt (24)
0
For t = 20 s we get IAE (linear) = 2.84 and IAE (nonlinear) = 2.00.

0.9
200 0.8
0.7
Degree of membership
100
0.6
0
u
0.5
−100 0.4
0.3
−200
100
0.2
50 100
0 0.1
0
−50
0
FE −100 −100 −100 −50 0 50 100
E
Input family: Neg and Pos
(a) (b)
Figure 5. Linear surface (a) with the corresponding input families (b).
Table 2. Fuzzy control rules.
E/F E NL NM NS ZR PS PM PL
NL NL NL NL NL NM NS ZR
NM NL NL NL NM NS ZR PS
NS NL NL NM NS ZR PS PM
ZR NL NM NS ZR PS PM PL
PS NM NS ZR PS PM PL PL
PM NS ZR PS PM PL PL PL
PL ZR PS PM PL PL PL PL
The system performance can be further improved. For that, we go to step 4 (and last)
of the design procedure – fine tune the controller. The nonlinear fuzzy controller will be
adjusted by changing the parameter values of Ke , Kf e, and Ku . In this study we propose
the use of a genetic algorithm (GA) to fine tune the gains of the controller. A GA is a search
technique used in computer science to find approximate solutions in optimization and search
problems. GA are a particular class of evolutionary algorithms that use techniques inspired
by evolutionary biology such as inheritance, mutation, natural selection, and crossover,
established by the Darwin’s theory of evolution [19, 38, 41, 32]. The advantage of GA is
in their parallelism. GA is travelling in a search space using more individuals than other
methods. However, GA also have disadvantages, namely the computational cost, because
many times these algorithms are slower than other methodologies.
The GA fitness function corresponds to the minimization of the integral absolute error
(IAE) criterion (as defined in (24)):
Z ∞
IAE(Ke, Kf e, Ku ) = |e (t)|dt (25)
0
where (Ke , Kf e , Ku ) are the fuzzy fractional PDµ controller parameters to be optimized.
NL NM NS ZR PS PM PL
1
100
0.8
50
0.6
0
u
0.4 −50
0.2 −100
100
50 100
0 50
0
0
−50 −50
−100 −50 0 50 100 E −100 −100
FE
E,FE,u
(a) (b)
Figure 6. Memberships functions for E, FE and u (a) Nonlinear control surface (b).
In Figure 7, the comparison of the unit-step responses of the closed-loop system with
G1 (s) controlled by the linear and nonlinear PDµ -controllers, and with the optimal nonlin-
ear FF-PDµ (µ = 0.7) controller is given. We establish the following values for the GA
parameters: population size P = 20, crossover probability C = 0.8, mutation probability
M = 0.05 and number of generations Ng = 50. The interval of the FLC parameters used in
the GA optimization are defined around 10% of the nominal parameters obtained with the
linear controller. In this case IAE (optimal) = 0.075. As can be seen, the optimal nonlinear
controller improves significantly the control system performance, in terms of overshoot,
rise time, and settling time, when compared with other fuzzy controllers. In this way, the
fuzzy fractional controller provides greater flexibility than the integer/fractional controller
and can be used to better adjust the dynamical properties of a control system.
3.3.2. Example 2: Fractional-Order Plant

Many real dynamical processes are modeled by fractional-order transfer functions [50,
44, 25]. Here we consider the fractional-order plant model given in [51]:
1
G2 (s) = (26)
0.8s2.2 + 0.5s0.9 + 1
An integer-order PD controller and a fractional-order PDµ -controller were designed in
[51]:
CP D (s) = KP + KD s = 20.5 + 2.7343s (27)
CP Dµ (s) = KP + KD sµ = 20.5 + 3.7343s1.15 (28)

Figure 8 shows the unit-step responses of the closed-loop fractional-order system with
the conventional PD-controller and with the PDµ -controller for G2 (s). The comparison
shows that for satisfactory feedback control of the fractional-order system is better to use
a fractional-order controller instead of a classical integer-order controller. Note, however,
1.6
Optimum
Linear
1.4
Nonlinear
1.2
1
Plant output
0.8
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18 20
Time [s]
Figure 7. Unit step responses of system with G1 (s) and with the linear, nonlinear and
optimum fuzzy fractional PDµ (µ = 0.7) controllers.
that the control system presents a steady-state error, since no integral action is employed.
In this case, for t = 6 s: IAE (PD) = 0.86 and IAE(PDµ ) = 0.55.
Let us now design an equivalent linear FF-PDµ controller. As in previous example, we
fix Ke = 100. The values of Kf e and Ku are obtained using expressions (17). The input
families and the linear control surface are illustrated in Figure 5. This result represents the
step 2 of the design procedure. In order to enhance the performance of the control system
we proceed to step 3 of the methodology.
For obtaining the desired surface control, we apply again the nonlinear fuzzy rule base
of Table 2. The corresponding membership functions and surface control are shown in
Figures 6.
In Figure 8, the comparison of the unit-step responses of the closed-loop system with
G2 (s) controlled by the linear and nonlinear FF-PDµ (µ = 1.15) controllers is given. We
verify that the nonlinear controller improved the overshoot, settling time, and steady-state
error, when compared with the linear fuzzy controller. In this case IAE (nonlinear) = 0.40,
which is smaller than the linear counterpart.
To further improve the system performance we fine tune the controller (step 4, and last,
of the design procedure). The fuzzy controller parameters (Ke , Kf e, Ku ) are tuned using
a GA while minimizing the IAE index (25). Figure 8 shows the unit-step responses of the
closed-loop system with G2 (s) controlled with the optimum nonlinear FF-PDµ (µ = 1.15)
controller. The GA parameters are: population size P =20, crossover probability C = 0.8,
mutation probability M = 0.05 and number of generations Ng = 25. The interval of
the FLC parameters used in the GA optimization are defined around 10% of the nominal
parameters obtained with the linear controller. In this case IAE (optimal) = 0.037. Clearly,
1.6
PD−controller
1.4 PDµ−controller
Nonlinear PDµ−controller
µ
1.2 Optimum PD −controller
1
Plant output
0.8
0.6
0.4
0.2
0
0 1 2 3 4 5 6
Time [s]
Figure 8. Unit step responses of system with G2 (s) with PD-controller, and with the linear,
nonlinear and optimal fuzzy fractional PDµ (µ = 1.15) controllers.
the optimal nonlinear controller produced better results than other fuzzy controllers, since
the transient response (namely the overshoot, rise time, settling time, and steady-state error)
and the error IAE are smaller.
Now, let us consider the FF-PDµ +Iλ -controller. In order to test the robustness of the
fuzzy controller, we introduce a load disturbance of amplitude l = −2 after 8 seconds in
system of Figure 1. We use the same (Kp, Kd ) parameters of the linear FF-PDµ (µ =
1.15) controller and tuned the (Ki , λ) for a satisfactory control response. The final tuned
parameters are (Ki , λ) = (10, 0.8). With Ke = 100, and using (21) we obtain Kf e, Ku ,
and Kf ie of the fuzzy controller.
Figure 9 shows the step and load responses of closed-loop system with FF-PDµ +Iλ
controller, (µ, λ)=(1.15, 0.8), for the linear and nonlinear control surfaces. We observe the
better response of the fuzzy controller to the reference and disturbance inputs with the non-
linear rule base (Table 2) compared to their linear counterpart (Table 1). If further improved
is needed, we can fine tune it using the GA while minimizing the IAE criterion (25), as done
in previous examples. Once more, we demonstrate the robustness and effectiveness of these
controllers.
4. Optimal Fuzzy Fractional PID Controllers

In this section we design optimal fuzzy fractional PID controllers by using PSO algo-
rithm [39, 31, 29, 30, 28].
1.6
Linear
Nonlinear
1.4
1.2
Plant output
0.8
0.6
0.4
0.2
0
0 5 10 15 20
Time [s]
Figure 9. Unit-step and load responses of the fractional control system with G2 (s) and with
the linear and nonlinear FF-PDµ +Iλ controllers.
4.1. Approximations by Rational Functions

In this study we adopt discrete integer-order approximations to the fundamental element
α
s (α ∈ <) of a fractional-order control (FOC) strategy. The usual approach for obtaining
discrete equivalents of continuous operators of type sα adopts the Euler, Tustin and Al-
Alaoui generating functions [10, 54, 14].
It is well known that rational-type approximations frequently converge faster than
polynomial-type approximations and have a wider domain of convergence in the complex
domain. Thus, by using the Euler operator w(z −1 ) = (1−z −1 )/Tc , and performing a power
series expansion of [w(z −1 )]α = [(1− z −1 )/Tc]α gives the discretization formula corre-
sponding to the Grünwald-Letnikov definition (4):
α ∞ ∞
1 − z −1 1 α

α −1
X
k α −k
X
hα (k) z −k

D z = = (−1) z = (29)
Tc Tc k
k=0 k=0
where Tc is the sampling period and hα (k) is the impulse response sequence.
A rational-type approximation can be obtained through a Padé approximation to the
impulse response sequence hα (k), yielding the discrete transfer function:
∞
b0 + b1 z −1 + . . . + bm z −m X
H z −1 = = h (k) z −k (30)
1 + a1 z −1 + . . . + an z −n
k=0
where m ≤ n and the coefficients ak and bk are determined by fitting the first m + n +
1 values of hα (k) into the impulse response h(k) of the desired approximation H(z −1 ).
Thus, we obtain an approximation that matchs the desired impulse response hα(k) for the
first m + n + 1 values of k [10]. Note that the above Padé approximation is obtained by
considering the Euler operator but the determination process will be exactly the same for
other types of discretization schemes [10].
4.2. Fuzzy Fractional Control System

Figure 10 presents the block diagram of the fuzzy control system to be analyzed, where
G(s) is the process transfer function. The gains of the fuzzy fractional PDβ +I controller
will be tuned through the application of a PSO algorithm in order to achieve a superior
control performance of system of Figure 10. Note the presence of a saturation nonlinearity.
The generalized PIα Dβ controller is given here as [51]:
U (s) Ki
Gc (s) = = Kp + α + Kd sβ , α, β > 0 (31)
E(s) s
where the fractional-order operators are implemented by rational approximations of type

(30).
In the system of Figure 10, we apply a fuzzy logic control (FLC) for the PDβ actions
and the integral of the error is added to the output in order to find a fuzzy PDβ + I controller
[6]. In fact, we can have a fuzzy PDβ +Iα controller, i.e. an added fractional integral action,
as presented in [5]. This is the FLC of Figure 4 with an integer integral action in the
feedforward path (α = 1). The block diagram of Figure 11 illustrates the configuration
of the proposed fuzzy controller. In this controller, the control actions are the error e, the
fractional derivative of e and the integral of e. The U represents the controller output. Also,
the controller has four gains to be tuned, Ke , Kie , Kce corresponding to the inputs and Ku
to the output.
The control action U is generally a nonlinear function of error E, fractional change of
error CE, and integral of error IE:
h i
U (k) = [f (E, CE) + IE] Ku = f Ke e (k) , KceD β e (k) + Kie Ie (k) Ku (32)
where D β is the discrete fractional derivative implemented as rational approximation (30)

using the Euler scheme (29); the integral of error is calculated by rectangular integration:
Tc
I z −1 =

(33)
1 − z −1
To further illustrate the performance of the fuzzy PDβ +I a saturation nonlinearity is
included in the closed-loop system of Figure 10, and inserted in series with the output of
the fuzzy controller. The saturation element is defined as:

u, |u| < δ
n (u) = (34)
δ sign (u) , |u| ≥ δ
where u and n are respectively the input and the output of the saturation block and sign(u)
is the signum function.
Figure 10. Block diagram of the fuzzy control system.
Figure 11. Fuzzy PDβ + I controller.
Here we give an emphasis of the proposed FLC presented in Figure 11. The basic
structure for FLC is illustrated in Figure 2. The fuzzy rule base, which reflects the collected
knowledge about how a particular control problem must be treated, is one of the main com-
ponents of a fuzzy controller. The other parts of the controller perform make up the tasks
necessary for the controller to be efficient. The rule base for the fuzzy PDβ +I controller is
presented in Table 2. The membership functions for the premises and consequents of the
rules are shown in Figure 12a). With two inputs and one output the input-output mapping
of the fuzzy logic controller is described by a nonlinear surface, presented in Figure 12b).
The fuzzy controller will be adjusted by changing the parameter values of Ke , Kce ,
Kie and Ku . The fuzzy inference mechanism operates by using the product to combine the
conjunctions in the premise of the rules and in the representation of the fuzzy implication.
For the defuzzification process we use the centroid method.
4.3. Particle Swarm Optimization

In this study we use a particle swarm optimization (PSO) algorithm for the tuning of
a fractional order controller, applied to a system. The PSO is one of the latest evolution-
ary techniques developed by Dr. Eberhart and Dr. Kennedy in 1995, inspired by social
behaviour of bird flocking or fish schooling. The PSO scheme optimizes searching by
virtue of the swarm intelligence produced by the cooperation and competition among the
particles of a species [20, 35, 21]. The social system is discussed through the collective be-
haviours of simple individuals interacting with their environment and each other. Examples
of this are the bird flock or fish school. Some applications of PSO are found in the field of
nonlinear dynamical systems, data analysis, electrical engineering, function optimization,
artificial neural network training, fuzzy control and many others in real world applications
[20, 21, 49, 55, 28].
It is important to mention that a reliable execution and analysis of a PSO usually re-
NL NM NS ZR PS PM PL
1
1
0.8
0.5
0.6
0
v
0.4 −0.5
0.2 −1
1
0.5 1
0 0.5
0
0
−0.5 −0.5
−1 −0.5 0 0.5 1 CE −1 −1
E
(a) (b)
Figure 12. Memberships functions for E, CE and v (a) Nonlinear control surface (b).
quires a large number of simulations to provide that stochastic effects have been properly
considered. In the PSO algorithm, all particles represent a potential solution to a problem,
which is performed by adjusting their position taking into account both personal and group
experiences [21, 28, 52]. In each iteration, the velocity (v) is actualized by expression
(35a). The new particle position (cpk+1 ) is found by adding their actual position with the
new velocity, as shown in (35b).
vk+1 = w ∗ vk + c1 ∗ r1 (lbp − cpk ) + c2 ∗ r2 ∗ (gbpk − cpk ) (35a)
cpk+1 = cpk + vk+1 (35b)
where ω is the inertia weight, c1 and c2 are the individual and sociality weights coefficients
for modelling attractive forces from the local and global best, respectively; r1 and r2 are
aleatory numbers between [0, 1], lbp is the local best position, cp is the current position,
and gbp is the global best position [28]. The inertia term, forces the particle to move in
the same direction as before by adjusting the old velocity. The cognitive term (personal
best), forces the particle to go back to the previous best position. On the other hand, the
social learning term (global best), forces the particle to move to the best previous position of
its neighbors. The PSO optimizes an objective function by iteratively improving a swarm
of solutions, called particles, based on special management of memory. Each particle is
modified by referring to the memory of individual swarm’s best information. Due to the
collective intelligence of these particles, the swarm is able to repeatedly improve its best
observed solution and converging to an optimum.
4.4. Objective Function for Optimization

The optimization fitness function corresponds to the minimization of the integral time
absolute error (ITAE) criterion, that measure the response error as defined as [29]:
Z∞
J (Ke , Kce , Kie, Ku ) = t |r (t) − c (t)| dt (36)
0
where (Ke , Kce , Kie , Ku ) are the PDβ +I controller parameters to be optimized. Other
criteria are possible like the integral time square error (ITSE), integral absolute error (IAE),
integral square error (ISE), among others [31, 34, 17, 16].
4.5. Case Studies
We analyze the closed-loop system of Figure 10 with a fuzzy fractional PDβ + I con-
troller (Figure 11). In all the experiments, the fractional order derivative Dβ in scheme
of Figure 11 is implemented by using a 4th order Padé discrete rational transfer function
(m = n = 4) of type (30). It is used a sampling period of Tc = 0.1 s. The PDβ +I con-
troller is tuned through the minimization of the ITAE (36) using a PSO. We use δ = 15.0.
We establish the following values for the PSO parameters: population number P N = 25,
c1 = c2 = 1, ω = 0.9 and a maximum number of iterations IM ax = 50. For guarantee that
stochastic effects are properly considered, the experiments consist on executing the PSO
10 times, and we get the best result, i.e. the simulation that leads to the smaller J, and
consequently a combination of controller gains, that leads to a better transient and steady
state responses.
4.5.1. Example 1. High-Order Transfer Function
In the first case, we compare a fuzzy fractional PDβ +I controller which leads to the
lower error (β = 0.9), with a fuzzy integer PID controller (β = 1). Figure 13a) shows the
unit step responses of both controllers. The plant system G3 (s) used is represented by a
high−order transfer function [1]:
1
G3 (s) = (37)
(s + 1)(1 + 0.5s)(1 + 0.52s)(1 + 0.53 s)
The controller parameters, corresponding to the minimization of the ITAE index, lead
to the values for the fuzzy integer PID controller: {Ke , Kce , Kie , Ku } ≡ { 1.0705, 3.8194,
0.4206, 4.4784 }, with J = 2.3763, and for the fuzzy fractional PDβ +I controller to the
following values: {Ke , Kce , Kie , Ku } ≡ {0.7193, 0.3551, 0.2702, 3.2029}, with J =
0.8409. These values lead us to conclude that the fuzzy fractional order controller produced
better results than the integer one, since the transient response (namely, the settling time,
rise time and overshoot) and the error J are smaller, as can be seen in Figure 13a). Figure
13b) shows the ITAE error as function of β. The graph shows that lowest error is produced
for β = 0.9. We also see that the fractional controller is always better than its integer
version (considering 0 < β ≤ 1).
Figure 14 illustrates the variation of FLC parameters (Ke , Kce , Kie , Ku ) as function
of the order’s derivative β, while Figure 15 shows the variation of the transient response
parameters, namely the settling time ts , rise time tr , peak time tp and overshoot ov(%)
versus β, for the closed-loop step response, when minimizing the ITAE index. The variation
of FLC parameters and the transient response parameters reveal a smooth variation with β.
0.8
c (t)
J
0.6
0.4
0
0.2 10
PIDβ
PID
0
0 2 4 6 8 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time [s] β
(a) (b)
Figure 13. Step responses of the closed-loop system with a fuzzy PD+I and PDβ +I (β =
0.9) controllers (a) Error J versus β for G3 (s) (b).
4.5.2. Example 2. Non-Minimum Phase System
In a second experiment, we consider a fuzzy PDβ +I controller which leads for lower
error to β = 0.8, applied to a process G4 (s) represented by a non−minimum system given
by the transfer function (38) [1].
1 − 5s
G4 (s) = (38)
(s + 1)3
Once again, we consider for comparison the corresponding integer version (β = 1).
Figure 16a) shows the unit step responses of both controllers.
to the values for the fuzzy integer controller: {Ke, Kce , Kie , Ku } ≡ { 0.9953, 1.3058,
0.4360, 0.1991 }, with J = 70.9016, and for the fuzzy fractional controller: {Ke , Kce ,
Kie , Ku }≡ {0.0873, 0.0569, 0.0417, 2.5740}, with J = 35.4189. These values lead us
to remain the previously conclusions drawn for G3 (s), namely that the fuzzy fractional
order controller produced better results than the integer one, since the transient response
(in particular the settling time, rise time and overshoot) and the error J are smaller. Figure
16b) shows the ITAE error as function of β. The graph shows a lowest error for β = 0.8.
Also, the fractional controller (for 0 < β ≤ 1) is always better than the conventional fuzzy
PID controller.
parameters, namely ts , tr , tp and ov(%) versus β, for the closed-loop step response, when
minimizing the ITAE index. The variation of FLC parameters and the transient response
parameters reveal a smooth variation with β.
1
10
0
10
0
Kce
−1 10
Ke
10
−2
10
−3
10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
−0.3 0.5
10 10
Kie
Ku
−0.4
10 0.3
10
−0.5
10
0.1
10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
Figure 14. The PDβ +I parameters (Ke , Kce , Kie , Ku ) versus β for G3 (s).
4.5.3. Example 3. Second-Order Transfer Function With Time Delay
In a third experiment, we consider a fuzzy PDβ +I controller which leads for lower error
to β = 0.9, applied to a process G5 (s) represented by a second-order transfer function with
a time delay (39).
e−s
G5 (s) = 2 (39)
4s + 4s + 1
Once more time, we consider for comparison the corresponding integer version (β = 1).
Figure 19 a) shows the unit step responses of both controllers.
to the values for the fuzzy integer controller: {Ke , Kce , Kie , Ku } ≡ {1.1616, 1.0076,
0.1993, 1.3628}, with J = 8.3228, and for the fuzzy fractional controller: {Ke , Kce , Kie ,
Ku }≡ {0.6600, 0.7851, 0.1171, 2.7525}, with J = 5.8728. These values lead us to remain
the previously conclusions drawn for G3 (s) and G4 (s), namely that the fuzzy fractional
order controller produced better results than the integer ones, since the transient response
(in particular the settling time, rise time and overshoot) and the error J are smaller. Figure
19b) shows the ITAE error as function of β. The graph shows that for this process we have
a lower error for β = 0.9.
parameters, namely ts , tr , tp and ov(%) versus β, for the closed-loop step response, when
minimizing the ITAE index. The variation of FLC parameters and the transient response
parameters reveal a smooth variation with β.
In conclusion, with the fuzzy fractional PDβ +I controller we get the best controller tun-
ing, superior to the performance revealed by the integer-order scheme. Moreover, we prove
the effectiveness of this control structure when used in systems with time delay. In fact, sys-
1
10
tr
t
0 0
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
2
10
1
10
ov (%)
p
t
0
10
0 −1
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
Figure 15. Parameters ts , tr , tp , ov(%) versus β of the step responses of the closed-loop
system with G3 (s) and with a fuzzy PDβ +I controller.
2
10
0.8
0.6
0.4
c (t)
0.2
J
−0.2
−0.4
−0.6
PIDβ
−0.8 PID
−1
0 5 10 15 20 25 30 0 0.2 0.4 0.6 0.8 1
time [s] β
(a) (b)
Figure 16. Step responses of the closed-loop system with fuzzy PD+I and PDβ +I (β = 0.8)
controllers (a) Error J versus β for G4 (s) (b).
tems with time delay are more difficult to be controlled with the classical methodologies,
however the proposed algorithm reveals that is very effective in the control of this type of
systems.
4.6. Influence of Order of Rational Approximations

Here we analyze the influence of the order of rational approximations used in the sim-
ulations for different values of m = n. In the experiments, the controller performance is
evaluated for several values of order m = n = {1, 2, ..., 10}. For this study we use the
0 1
10 10
−1
10 0
10
Kce
−2
Ke
10
−1
−3
10
10
−4 −2
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
0
10
0
10
Kie
−1
Ku
10
−2
10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
ts
tr
1 0
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
0
10
ov (%)
tp
1 −1
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
second-order transfer function with time delay G5 (s).

Figure 22 shows the unit step responses of PDβ +I controller for various orders of ap-
proximations m, and for the best case of 10 experiments, i.e. the simulation that leads to
the smaller J. We observe that the responses are very similar varying slightly with order m.
0.8
c (t)
1
10
J
0.6
0.4
0.2
PIDβ
PID
0
0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time [s] β
(a) (b)
Figure 19. Step responses of the closed-loop system with fuzzy PD+I and PDβ +I (β = 0.9)
controllers (a) Error J versus β for G5 (s) (b).
0
0 10
10
Ke
ce
K
−1
10
0.2 0.4 0.6 0.8 0 0.5 1
β β
Kie
Ku
−1 0
10 10
0 0.5 1 0 0.5 1
β β
For this process we have a lower error for β between 0.8 and 0.9 as function of m. These
facts can also be seen in Figure 23 that shows the ITAE error as function of β and m. From
the graph, it is clear that the best approximations are found for 2 6 m 6 5.
of order’s derivative β and order of approximation m, while Figure 25 shows the variation
of the transient response specifications, namely ts , tr , tp and ov(%) versus β and m, for
the closed-loop step response. It is observed that the FLC parameters and the transient
response specifications varies with β and m. For instance, the system overshoot is lower
for relatively high values of β and wide range of order’s 2 6 m 6 10, as seen in Figure 25.
2
10
ts
tr
10
0 0
10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
2
10
ov (%)
1
tp
10
0
10
0
10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
β β
1.2
0.8
c (t)
0.6
0.4
0.2
0
0 5 10 15 20
time [s]
Figure 22. Step responses of the closed-loop system with fuzzy PDβ +I controller versus
order m for G5 (s).
The same analysis can be done for other parameters.

15
10
J
5
0
0
0.2
2
0.4
4
0.6 6
β
0.8 8 m
1 10
Figure 23. Error J versus β and m for G5 (s).
1.5 2
1.5
1
Ke
Kce
1
0.5
0 0
0.5
2 2
0 4
4 0
0
0.2 6 0 6 m
0.4 m 0.2
0.6 8 0.4 8
β 0.6
0.8 0.8
1 10 β 10
1
0.4
4
0.3
3
Kie
0.2
2
Ku
0
0.1 1
0
2
0 2 0
0 0 4
4
0.2 0.2 6 m
0.4 6 m 0.4
0.6 8 0.6 8
β 0.8 β 0.8
1 10 1 10
Figure 24. The PDβ +I parameters (Ke , Kce , Kie , Ku ) versus β and m for G5 (s).
5. Conclusion
This chapter has presented several fuzzy control structures with fractional-order deriva-
tive and integral actions combined to form a PID-type controller. Comparison with conven-
tional fuzzy and PID controllers is performed to illustrate the effectiveness and robustness
of the proposed algorithms. Another distinguished feature of this study is the use of discrete
20
4.5
4
15
3.5
ts
tr
10 3
2.5
5 2
0 0 0
0 2
0.2 0.2
2 4
0.4 0.4
4
0.6 0.6 6
6 β m
β 0.8 m 0.8 8
8
1 10 1 10
10
15 8
ov(%)
10 4
tp
5 0 0
0 2 0
0.2 0
0.2
4 2
0.4 0.4
6 4
0.6 m 0.6 6
β 8 β m
0.8 0.8 8
1 10 1 10
Figure 25. Parameters ts , tr , tp , ov(%) versus β and m of the step responses of the closed-
loop system with G5 (s) and with a fuzzy PDβ +I controller.
fuzzy fractional-order controllers in the form of series and rational approximations which
reveals useful for a digital implementation of these algorithms.
In the first part of chapter, a methodology for tuning fuzzy fractional PID controllers
is introduced. This methodology is simple and effective and can be used to replace an
existent integer/fractional PID controller in order to enhance system performance. Also, it
was demonstrated that, with this approach, the designed controllers are equivalent to the
conventional fractional PD/PID controllers by using a linear input-output mapping of the
rule base of the fuzzy fractional controller. Moreover, by making the controller nonlinear,
the performance of the control system proves to be, in most systems, better than its linear
counterpart.
In the second part of chapter, were developed optimal fuzzy fractional PID controllers
in which the parameters were tuned through a PSO algorithm. A nonlinear fuzzy control
system with saturation in actuator is analyzed. In general, the control strategies presented,
give better results than those obtained with conventional integer control structures, showing,
once more, its effectiveness in the control of nonlinear systems. An analysis study regarding
the influence of order of discrete rational approximations is also provided.
Certainly, the incorporation of fuzzy reasoning into fractional-order controllers will
increase the applicability of these controllers.
References
[1] Kiam Heong Ang and Gregory Chong, PID control system analysis, design, and tech-
nology, IEEE Transactions on Control Systems Technology 13 (2005), 559–576.
[2] K. J. Astrom and T. Hagglund, PID controllers: Theory, design, and tuning, Instru-
ment Society of America, USA, 1995.
[3] Ramiro S. Barbosa, On linear fuzzy fractional PD and PD+I controllers, Proceed-
ings of FDA’10, 4th IFAC Workshop Fractional Differentiation and its Applications
(Badajoz, Spain), October 18-20, pp. 1–6.
[4] Ramiro S. Barbosa and Isabel S. Jesus, Application of fuzzy fractional PD+I con-
trollers tuned by a genetic algorithm, 39th Annual Conference of the IEEE Industrial
Electronics society-IEEE-IECON’13 (2013), 3333–3338.
[5] , A methodology for the design of fuzzy fractional PID controllers, 10th Inter-
national Conference on Informatics in Control, Automation and Robotics - ICINCO
2013 (2013), 276–281.
[6] Ramiro S. Barbosa, Isabel S. Jesus, and Manuel F. Silva, Fuzzy reasoning in
fractional-order PD controllers, Proceedings of AIC’10, 10th WSEAS International
Conference on Applied Informatics and Communications - New Aspects of Applied
Informatics, Biomedical Electronics Informatics and Informatics and Communica-
tions (Taipei, Taiwan), August 20-22, 2010, pp. 252–257.
[7] Ramiro S. Barbosa, J. A. Tenreiro Machado, and Isabel M. Ferreira, Tuning of PID
controllers based on bode’s ideal transfer function, Nonlinear Dynamics 38 (2004),
305–321.
[8] , Least-squares design of digital fractional-order operators, Proceedings of

the First IFAC Workshop on Fractional Differentiation and its Applications (FDA’04)
(Bordeaux, France), July 19-21, 2004, pp. 434–439.
[9] , Pole-zero approximations of digital fractional-order integrators and differen-

tiators using signal modeling techniques, 16th IFAC World Congress (Prague, Czech
Republic), July 4-8, 2005.
[10] Ramiro S. Barbosa, J. A. Tenreiro Machado, and Manuel F. Silva, Time domain design
of fractional differintegrators uing least-squares, Signal Processing 86 (2006), 2567–
2581.
[11] , Discretization of complex-order algorithms for control applications, Journal

of Vibration and Control 14 (2008), no. 9/10, 1349–1361.
[12] James Carvajal, Guanrong Chen, and Haluk Ogmen, Fuzzy PID controller: Design,
performance evaluation, and stability analysis, Journal of Information Science 123
(2000), 249–270.
[13] Y. Q. Chen and B. M. Vinagre, A new IIR-type digital fractional order differentiator,
Signal Processing 83 (2003), no. 11, 2359–2365.
[14] Y. Q. Chen, B. M. Vinagre, and I. Podlubny, Continued fraction expansion approaches

to discretizing fractional order derivatives-an expository review, Nonlinear Dynamics
38 (2004), 155–170.
[15] YangQuan Chen and Kevin L. Moore, Relay feedback tuning of robust PID controllers
with iso-damping property, IEEE Transactions on systems, Man, and Cybernetics-Part
B: Cybernetics 35 (1) (2005), 23–31.
[16] S. Das, I. Pan, and S. Das, Performance comparison of optimal fractional order hybrid
fuzzy PID controllers for handling oscillatory fractional order processes with dead
time, ISA Transactions 52 (2013), 550–566.
[17] S. Das, I. Pan, and A. Gupta, A novel fractional order fuzzy PID controller and its
optimal time domain tuning based on integral performance indices, Engineering Ap-
plications of Artificial Intelligence 25 (2012), 430–442.
[18] H. Dekavari, R. Ghaden, A. Ranjbar, and S. Momani, Fuzzy fractional order slid-
ing mode controller for nonlinear systems, Comunications in Nonlinear Sicence and
Numerical Simulation 15 (2010), no. 4, 963–978.
[19] Goldberg E., Genetic algorithms in search optimization and machine learning,
Addison-Wesley, 1989.
[20] R. C. Eberhart and J. Kennedy, A new optimizer using particle swarm theory, Sixth
International Symposium on Micro Machine and Human Science (1995), 39–43.
[21] R. C. Eberhart and Y. Shi, Particle swarm optimization: developments, applications

and resources, Congress on Evolutionary Computation (2001).
[22] I. Eker and Torun Yunis, Fuzzy logic control to be conventional methods, Energy
Coversion and Management 47 (2006), 377–394.
[23] S. Glichet and L. Foulloy, Fuzzy controllers: Synthesis and equivalences, IEEE Trans-
actions on Fuzzy Systems 3 (1995), no. 2, 140–148.
[24] Monson H. Hayes, Statistical digital signal processing and modeling, Wiley & Sons,
New York, 1996.
[25] R. Hilfer, Applications of fractional calculus in physics, World Scientific, Singapore,

2000.
[26] J. Jantzen, Foundations of fuzzy control, Wiley and Sons, Chichester, England, 2007.
[27] David F. Jenkins and Kevin M. Passino, An introduction to nonlinear analysis of fuzzy
control systems, Engineering and Technology 7 (1999), no. 1, 75–103.
[28] Isabel S. Jesus, Application of PSO algorithm for the study of fractional order electri-
cal potential, 18th IFAC World Congress - IFAC-WC’11 (2011), 10812–10817.
[29] Isabel S. Jesus and Ramiro S. Barbosa, Application of fuzzy fractional PD+I con-
trollers tuned by a genetic algorithm, 39th Annual Conference of the IEEE Industrial
Electronics society-IEEE-IECON’13 (2013), 3333–3338.
[30] , Tuning of fuzzy fractional PD+I controllers by genetic algorithm, 10th Inter-
national Conference on Informatics in Control, Automation and Robotics - ICINCO
2013 (2013), 282–287.
[31] Isabel S. Jesus and J. A. Tenreiro Machado, Application of fractional calculus in the
control of heat systems, Journal of Advanced Computational Intelligence and Intelli-
gent informatics 11(9) (2007), 1086–1091.
[32] , Implementation of fractional-order electromagnetic potential through a ge-

netic algorithm, Journal of Communications in Nonlinear Science and Numerical
Simulation 14 (2009), 1838–1843.
[33] Isabel S. Jesus, J. A. Tenreiro Machado, and Ramiro S. Barbosa, Fractional order
nonlinear control of heat system, 13rd workshop on Fractional Differentiation and its
Applications - IFAC-FDA’08 (Ankara, Turkey), 2008.
[34] Isabel S. Jesus, J. T. Machado, and Ramiro S. Barbosa, On the fractional order con-
trol of heat systems, Intelligent Engineering Systems and Computational Cybernetics,
Springer, Netherlands, 2008.
[35] J. Kennedy and R. C. Eberhart, Particle swarm optimization, IEEE international con-
ference on neural networks IV (1995), 1942–1948.
[36] C. C. Lee, Fuzzy logic in control systems: fuzzy logic controller-part I and II, IEEE
Transactions on System Man, and Cybernetics-Part B: Cybernetics 20 (1990), no. 2,
404–435.
[37] H.-H Li and H. B. Gatland, Conventional fuzzy control and its enhancement, IEEE
Transactions on System Man, and Cybernetics-Part B: Cybernetics 26 (1996), no. 5,
791–797.
[38] Mitchell M., An introduction to genetic algorithms, SMIT Press, 1998.
[39] J. A. Tenreiro Machado, Analysis and design of fractional-order digital control sys-
tems, SAMS Journal of Systems Analysis, Modelling and Simulation 27 (1997), 107–
122.
[40] G. K. I. Mann, B.-G Hu, and R. G. Cosine, Analysis of direct action fuzzy PID con-
troller structures, IEEE Transactions on Systems, Man, and Cybernetics-Part B: Cy-
bernetics 29 (1999), no. 3, 371–388.
[41] Z. Michalewicz, Genetic algorithms + data structures = evolution programs,

Springer-Verlag, 1996.
[42] K. S. Miller and B. Ross, An introduction to the fractional calculus and fractional
differential equations, Wiley & Sons, New York, 1993.
[43] M. Mizumoto, Realization of PID control by fuzzy control methods, Fuzzy Sets and
Systems 70 (1995), 171–182.
[44] K. B. Oldham and J. Spanier, The fractional calculus, Academic Press, New York,
1974.
[45] Manuel D. Ortigueira and António J. Serralheiro, New insights into pseudo-fractional
ARMA modelling, Proceedings of the Second IEEE International Conference on Com-
putational Cybernetics (ICCC 2004) (Vienna, Austria), August 30-September 1, 2004,
pp. 391–394.
[46] A. Oustaloup, La dérivation non entière: Théorie, synthèse et applications, Editions

Hermès, Paris, 1995.
[47] I. Pan and S. Das, Intelligent fractional order systems and control: An introduction,
Springer, Berlin, 2013.
[48] K. M. Passino and S. Yurkovich, Fuzzy control, Addison-Wesley, Menlo Park, Cali-
fornia, 1998.
[49] Eduardo S. Pires, Paulo B. Moura de Oliveira, J. A. Tenreiro Machado, and Isabel S.
Jesus, Fractional order dynamics in a particle swarm optimization algorithm, Sev-
enth International Conference on Intelligent Systems Design and Applications- IEEE
- ISDA’07 (2007), 703–708.
[50] I. Podlubny, Fractional differential equations, Academic Press, San Diego, 1999.
[51] , Fractional-order systems and PIλ Dµ -controllers, IEEE Transactions on Au-

tomatic Control 44 (1999), no. 1, 208–214.
[52] H. Shayeghi, A. Safari, and H. A. Shayanfar, Multimachine power system stabilizers

design using PSO algorithm, International Journal of Electrical Power and Energy
Systems Engineering 1 (2008), 703–708.
[53] M.-Y. Shieh and T.-H. S. Li, Design and implementation of integrated fuzzy logic
controller for a servomotor system, Mechatronics 8 (1998), 217–240.
[54] B. M. Vinagre, Y. Q. Chen, and I. Petras, Two direct tustin discretization methods for
fractional-order differentiator/integrator, Journal of the Franklin Institute 340 (2003),
349–362.
[55] Z. H. Zhan, J. Zhang, Y. Li, and H. S. H Chung, Adaptive particle swarm optimization,
IEEE Transactions on Systems, Man, and Cybernetics 39 (2009), 1362–1381.
INDEX
bisphenol, 200
blood pressure, 252
# bounds, 17
brain, 247
20th century, 96, 123 branching, 11
breathing, 256
Brownian motion, 230
A Bulgaria, viii
abstraction, 53
accelerometers, 69 C
acrylonitrile, 200
action potential, 247 calculus, viii, ix, 15, 20, 38, 39, 61, 65, 66, 88, 89,
actuation, 56 96, 120, 149, 150, 151, 162, 166, 167, 168, 169,
actuators, 150, 154, 161 171, 198, 199, 200, 201, 203, 204, 205, 207, 209,
adaptation, 20, 149, 153, 154, 156, 254 220, 221, 222, 223, 224, 227, 229, 243, 246, 247,
aggregation, 16, 19 255, 257, 259, 286, 287, 288
Algeria, 65 Caputo derivative, 151, 166
algorithm, 25, 34, 59, 67, 76, 79, 83, 85, 88, 149, cardiac output, 252
150, 151, 161, 254, 259, 260, 261, 268, 271, 273, Cauchy problem, 235, 236, 239, 241
274, 275, 279, 284, 285, 286, 287, 288 causality, 7, 9
amplitude, 42, 57, 68, 126, 127, 128, 132, 139, 140, CDC, 90, 256
142, 143, 182, 192, 216, 218, 262, 271 chaos, ix, 16, 19, 21, 23, 29, 30, 32, 34, 257
analgesic, 247 chemical, 158, 161, 246
anatomy, 256 Chicago, 226
anesthesiologist, 243, 252 China, viii
anomalous diffusion, 240 closure, 7, 16, 18, 55
assimilation, 244 combustion, 58, 61
asthmatic children, 147 community, viii, 2, 247
asymmetry, 203 compatibility, 19, 211
auditory evoked potentials, 257 compensation, 49
Austria, 288 competition, 16, 18, 274
automation, 66 complex non-integer orders, vii
automotive applications, 45, 61 complex numbers, 3
complexity, 46, 48, 69, 130
compliance, 181, 182, 183, 191
B composites, 225
compressibility, 39
bandwidth, 49, 50, 51, 52 compression, 165
base, 2, 263, 265, 266, 267, 270, 271, 274, 284 computation, 3, 76, 79, 262, 267
batteries, 2, 4, 16 computing, 53, 134
beams, 165 conduction, 94, 106, 119, 120, 147, 227, 228, 229,
behaviors, 50, 98, 101, 111, 112, 113, 114, 115 230, 232, 239, 240, 241
Belgium, viii, 243 conductivity, 94, 106, 124, 125
benefits, 38, 39, 40, 48, 52, 57
290 Index
configuration, 273 DSC, 163

conscious awareness, 247 duality, 11, 129
conservation, 227 dynamic systems, 60, 62, 267
constant load, 168 dynamical properties, 23, 26, 269
construction, 7, 14, 53, 208, 230 dynamical systems, 23, 24, 88, 149, 162, 274
constructivism, 6
consumption, 37, 38, 57
contour, 192, 206 E
contradiction, 4, 5
convergence, 48, 81, 150, 272 elastic deformation, 53
copper, 123, 124, 125, 140, 141, 142, 143, 144, 145, electrochemistry, 20, 93
146 electrodes, 16
correlations, 1, 7, 9, 10, 16, 17 electrolyte, 2
cost, 37, 39, 49, 268 electromagnetic, 287
creep, 168, 170, 186, 191, 194, 196, 199, 221, 224, electromagnetism, 93
225 electrons, 3
cryptography, 35 energy, 1, 2, 3, 4, 5, 6, 7, 9, 10, 11, 12, 15, 16, 17,
18, 19, 20, 42, 81, 88, 95, 227, 228, 240, 243
energy consumption, 20
D engineering, viii, ix, 4, 12, 18, 69, 150, 154, 161,
162, 168, 215, 224, 243, 274
damping, vii, 69, 140, 167, 194, 199, 210, 216, 217, England, 62, 286
220, 222, 224, 225, 286 entropy, 3, 6, 7, 9, 11, 15, 17, 18
data analysis, 274 environment, 2, 7, 10, 11, 13, 17, 18, 49, 53, 243,
data structure, 287 247, 274
decay, 153, 228 environmental policy, 37, 38
decomposition, 15, 38, 40, 46, 56, 57, 58 equality, 213, 214, 215
deconstruction, 6, 7, 12 equilibrium, 7, 9, 10, 14, 23, 24, 25, 27, 28, 37, 40,
deformation, 177, 181, 183, 217, 221 43, 44, 45, 47, 55, 81, 84, 86, 194, 210, 216, 232,
degenerate, 11, 208 240
depth, 106, 243, 244, 246, 252 equipment, 49
derivatives, viii, 25, 26, 44, 59, 62, 66, 123, 150, Euclidean space, 11
151, 152, 158, 165, 166, 167, 169, 188, 191, 192, evolution, 81, 210, 268, 287
201, 211, 220, 223, 224, 225, 233, 262, 263, 285 execution, 274
diesel engines, 58 exponential functions, 6
differential equations, 1, 8, 19, 23, 34, 35, 61, 66,
151, 161, 221, 287, 288
diffusion, viii, ix, 4, 8, 11, 15, 24, 93, 94, 106, 108, F
119, 124, 227, 228, 229, 230, 231, 232, 233, 235,
236, 237, 238, 239, 240, 241, 243, 246, 247, 256, families, 263, 267, 268, 270
260 FDA, viii, 59, 60, 285, 287
diffusion process, 15 fidelity, 130
diffusion time, 108 filters, 263
diffusivity, 95, 106, 124, 228 financial support, 88, 255
direct action, 287 fish, 260, 274
discontinuity, 154, 158, 208, 210 fitness, 268, 275
discretization, 272, 273, 288 flatness, 67
disequilibrium, 10 flexibility, 39, 50, 66, 161, 260, 269
disordered systems, 247 fluid, 24, 42, 53, 158, 167
dispersion, 200, 221, 240, 241 force, 4, 42, 81, 181, 182, 192, 215, 216, 217, 219,
displacement, 42, 69, 210, 215, 232, 247 230
disposition, 256 Ford, 34
distribution, vii, 2, 3, 5, 15, 16, 17, 18, 40, 53, 134, formula, vii, 41, 42, 51, 53, 67, 151, 166, 167, 171,
194, 199, 216, 218, 249, 252 172, 173, 175, 176, 179, 189, 192, 204, 206, 213,
distribution function, 194, 199, 216 215, 217, 218, 219, 234, 252, 272
divergence, 11, 18 foundations, 1, 165, 260
dosing, 243, 256 fractal dimension, 4, 5, 6, 10
drug release, 244, 247, 256 fractal space, 5, 6, 10, 11
drugs, 243, 248, 255, 256 fractal structure, 2, 16, 17
Index 291
fractality, 1, 4, 5, 6, 10, 13, 18 injections, 56

fractional differentiation, 20, 66, 123, 148, 225 integration, 39, 46, 48, 49, 53, 66, 67, 68, 72, 73, 94,
fractional order differential equations, 35, 151 98, 101, 111, 112, 115, 116, 120, 148, 150, 151,
fractional order systems, 24, 25, 30, 46, 66, 67, 89, 192, 213, 261, 273
93, 121, 123, 124, 162, 288 intelligence, 260, 274, 275
France, vii, viii, 21, 34, 59, 60, 61, 89, 90, 93, 123, intensive care unit, 256
147, 285 interface, ix, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 15, 49, 93,
freedom, 15, 39, 48, 50 94, 96, 106, 119, 120, 124, 146
friction, 42, 53, 81, 167, 181, 182, 221, 222 interrogations, 4
fuel consumption, 37, 38 invariants, 4
fuzzy set theory, 260 inversion, 6, 16, 19, 192, 204, 206, 218, 220, 233
ions, 3
Iran, 149
G Ireland, 58
iron, 123, 124, 125, 140, 141, 142, 143, 144, 145,
general anesthesia, ix, 248, 256 146
geometrical parameters, 42 Italy, viii, 37, 60, 146
geometry, 1, 2, 3, 4, 6, 7, 10, 11, 15, 20, 56, 217 iteration, 132, 275
glue, 125
gravity, 19, 81, 263
groundwater, 230 J
Japan, 60, 89
H Jordan, 192, 193
half-integer order controller, vii

half-integer orders, vii K
heart rate, 252
heat capacity, 95 kinetics, 241, 243, 247, 256
height, 221, 249 Korea, 62
heredity, 168, 172, 208, 210, 212, 217, 219
historical overview, 169
history, viii, 24, 257 L
homogeneity, 5
human body, 245 laws, 2, 13, 167, 220, 228, 247
Hungary, 89 lead, 11, 18, 50, 54, 57, 136, 148, 227, 247, 276,
Hunter, 217, 221 277, 278
hybrid, 286 leakage, 11
hyperbolic systems, 18 lean body mass, 249
hypnosis, 244 Lebanon, 93, 120, 123, 146
light, 248
linear law, 221
I linear model, 38, 40, 44, 46, 53, 83, 85, 131
linear systems, 88, 162
ideal, 16, 39, 58, 65, 67, 68, 77, 78, 87, 88, 167, 262, liquids, 20
285 lithium, 20
identification, 47, 62, 93, 94, 119, 120, 147, 148, locus, 196, 206, 208, 218
252, 253, 254 LPG, 58
identity, 11, 180 LTD, 222
impulses, 247 Luo, 59, 60, 162
impulsive, 221 Lyapunov function, 155, 156
independence, 3, 10
India, 23, 256
individuals, 268, 274 M
induction, 120, 124, 147, 250
industry, 6, 38, 39, 57, 58, 61, 90, 261 machine learning, 286
inequality, 17, 25, 158, 213 macromolecules, 168
inertia, 42, 81, 167, 215, 275 magnet, 162
inhibitor, 239
292 Index
magnitude, 49, 50, 68, 102, 179, 189, 192, 194, 196, nonlocality, 24
202, 206, 208, 212, 213 null, 219, 262
management, 275 numerical applications, vii
mapping, 10, 12, 263, 265, 267, 274, 284 numerical computations, 235
mass, vii, 41, 42, 81, 84, 85, 86, 87, 192, 199, 215,
223, 227, 228, 230, 245, 249
mass-damping, vii O
materials, 123, 124, 125, 126, 128, 146, 167, 168,
177, 200, 217, 225, 262 one dimension, 3
mathematics, 6, 20 operating range, 47, 53, 55
matrix, 24, 25, 27, 31, 70, 76, 79, 257 operations, 55
matter, 2, 3, 4, 7, 44, 227, 228, 230, 231, 232, 239 opportunities, 6, 19
measurement(s), 3, 5, 6, 9, 11, 15, 17, 83, 94, 106, optimization, 2, 6, 93, 119, 147, 162, 225, 255, 259,
131, 132, 146 260, 268, 269, 270, 274, 275, 286, 287, 288
media, 3, 5, 6, 7, 20, 117, 119, 165, 166, 168, 208, orthogonality, 216
210, 212, 219, 228
median, 139
medicine, 243, 256, 257 P
membership, 263, 267, 268, 269, 270, 274, 275
memory, 24, 45, 167, 212, 220, 222, 227, 228, 231, pain, 244, 245, 246, 247, 248
240, 262, 267, 275 parallel, 223
metals, 241 parallelism, 268
methodology, 38, 40, 46, 47, 49, 57, 88, 93, 259, parameter estimation, 150
260, 261, 270, 283, 285 partition, 6, 17
microwaves, 2 path integrals, 9
Ministry of Education, 219 percolation, 2, 247
mixing, 15, 16, 17 pharmacokinetics, 256, 257
mobile robots, 80 pharmacology, 256
modelling, viii, 24, 257, 275, 288 phase diagram, 49, 50, 183
models, vii, 9, 24, 39, 43, 44, 66, 91, 119, 131, 150, physical characteristics, 97, 108
167, 168, 169, 171, 188, 189, 191, 199, 200, 220, physical environment, 2
224, 225, 243, 244, 245, 246, 247, 248, 249, 251, physical phenomena, 12
252, 255, 256, 257 physics, viii, ix, 4, 5, 6, 18, 19, 93, 162, 247, 286
modulus, 171, 177, 178, 190, 192, 193, 194, 196, physiology, 257
215, 216, 217, 219 plane waves, 225
moisture, 120 plants, 39, 49, 150, 158, 161, 260, 261, 267
momentum, 10, 41 platinum, 125
monomers, 16 Poincaré, 9, 10, 11, 12
Montana, 147 Poisson ratio, 232
Moscow, 222, 223, 224, 241 Poland, 227
motion control, 59, 60 polar, 234, 237
motivation, 39, 69 polarization, 124
multiplication, 6, 16, 165, 173 polycarbonate, 200
multiplier, 233 polymer, 200, 221, 247
mutation, 268, 269, 270 polymer systems, 221
polymeric materials, 168
polymers, 16, 168, 200, 221, 225, 247
N polymorphism, 53
population, 252, 269, 270, 276
natural gas, 37, 38, 59 population size, 269, 270
natural selection, 268 porosity, 247
nerve, 247 porous media, 24, 230
Netherlands, 61, 287 Portugal, viii, 59, 259
neural network(s), 274, 287 POWER, 1
neurotransmitters, 247 process control, 93, 94, 119
non-integer differentiation, vii, viii project, viii, 33, 58, 88
non-integer models, vii propagation, 43, 44, 165, 168, 220, 221, 223
nonlinear dynamics, 239, 243, 260 prototypes, 38, 40
nonlinear systems, 153, 161, 162, 284, 286 pumps, 158, 159
Index 293
self-organization, 19
Q semicircle, 6, 7
sensitivity, 37, 45, 131, 135, 136, 137, 138, 250
quantum mechanics, 19 sensors, 58, 69, 244
shape, 57, 68
shear, 178, 181, 183
R shear deformation, 181
shock, 165, 208, 210, 211, 212, 215, 219, 220, 223
radius, 215
shock waves, 165, 208, 210, 219
random walk, 239, 240, 247
showing, ix, 167, 260, 284
reactions, 246, 251, 252
signals, 31, 32, 33, 35, 42, 44, 128, 247, 252
reasoning, 259, 284, 285
signs, 176, 210
recall, viii, 4, 7, 13, 69, 252
simulation, 38, 40, 45, 53, 54, 55, 57, 60, 66, 94,
recovery, 243
101, 118, 120, 128, 130, 139, 141, 147, 149, 150,
regulations, 37, 38
151, 159, 162, 252, 259, 267, 274, 276, 280
rejection, 38
Singapore, 58, 162, 257, 286
relaxation, 7, 168, 171, 181, 183, 186, 188, 190, 191,
social learning, 275
194, 196, 197, 198, 199, 200, 201, 203, 207, 208,
society, 19, 285, 286
209, 215, 218, 219, 221, 222, 224, 225, 240
software, 40
relaxation process, 168, 208, 225
solid state, 167
relaxation processes, 168, 208, 225
solution, 7, 23, 34, 45, 67, 75, 83, 96, 107, 168, 176,
relaxation times, 183, 191, 208
192, 193, 194, 196, 197, 201, 202, 203, 204, 205,
relevance, 2, 10
210, 211, 216, 218, 220, 235, 236, 237, 239, 247,
renormalization, 16, 18
254, 275
reserves, 2
space-time, 4, 5, 10, 19, 241
residues, 193
Spain, viii, 90, 285
resistance, 10, 125
species, 2, 3, 274
resonator, 90
specifications, 38, 40, 45, 48, 49, 51, 57, 131, 132,
resources, 286
135, 141, 145, 261, 281
respiratory rate, 252
speech, 19
response, 1, 9, 18, 30, 31, 38, 43, 48, 49, 50, 54, 56,
spinal cord, 247
60, 67, 69, 78, 87, 94, 98, 101, 102, 103, 111, 112,
St. Petersburg, 224
114, 115, 126, 128, 130, 134, 139, 140, 141, 142,
stability, vii, 23, 24, 27, 38, 40, 45, 46, 47, 48, 49,
143, 144, 159, 168, 215, 217, 224, 240, 249, 250,
50, 55, 56, 57, 58, 59, 66, 67, 69, 75, 76, 81, 84,
256, 260, 271, 272, 273, 275, 276, 277, 278, 281
87, 123, 124, 132, 133, 135, 137, 138, 140, 145,
responsiveness, 257
146, 149, 155, 156, 285
restrictions, 51
stability-precision, vii
retardation, 168, 171, 181, 186, 188, 191, 194, 196,
stabilization, 59, 65, 66, 69, 80, 83, 84, 85, 87, 89,
200, 203, 204, 205, 218, 224
90, 159
robotics, 260
stabilizers, 288
roots, 11, 25, 46, 47, 69, 75, 96, 174, 176, 180, 189,
stable states, 16
192, 193, 194, 195, 196, 197, 198, 199, 200, 201,
state, ix, 7, 10, 14, 15, 23, 40, 42, 43, 49, 56, 65, 66,
202, 203, 204, 206, 207, 208, 209, 216
67, 69, 70, 72, 73, 75, 76, 77, 79, 80, 82, 83, 84,
rotations, 14
85, 86, 87, 88, 89, 90, 119, 121, 132, 135, 137,
rules, ix, 1, 16, 18, 45, 48, 50, 54, 61, 259, 260, 265,
138, 140, 145, 149, 152, 158, 159, 162, 165, 166,
267, 268, 274
167, 232, 244, 245, 248, 266, 270, 271, 276
Russia, 1, 165, 224
states, 10, 14, 16, 95, 106, 152
statistics, 5, 20
S steel, 40, 217
stimulus, 247
safety, 37, 38 storage, 2, 6, 12
saturation, 55, 56, 81, 89, 135, 260, 265, 273, 284 strategy use, 81
scaling, 1, 3, 4, 5, 7, 11, 17, 20, 245, 247, 252, 265, stress, 168, 171, 181, 190, 210, 220, 232, 236, 237,
267 238, 245
scaling law, 1, 247 structure, 2, 3, 4, 11, 12, 16, 17, 18, 20, 27, 94, 168,
search space, 268 216, 223, 228, 230, 256, 274, 278
second generation, 123, 125, 132, 146 style, 267
secure communication, 29 Sun, 59, 62
superconductivity, 19
294 Index
suppression, 69, 150 Turkey, viii, 287

swarm intelligence, 274
Switzerland, 60
symmetry, 4, 6, 23 U
synchronization, 23, 29, 30, 31, 32, 33, 34, 35
synchronize, 30 U.S. Army Corps of Engineers, 121
synthesis, vii, ix, 38, 52, 57, 60, 62, 94, 130 United States (USA), 10, 35, 62, 88, 171, 223, 284
system analysis, 62, 284 USSR, 150
universality, 21
universe, 265, 267
T USSR, 150
Taiwan, 285
tanks, ix, 158, 159, 161, 163 V
target, 217, 221, 250, 252, 257
techniques, 30, 48, 52, 154, 255, 260, 268, 274, 285 validation, 53, 94
technologies, 58 valve, 37, 38, 40, 42, 44, 54, 55, 56
technology, 66, 146, 284 variables, 3, 42, 45, 46, 53, 54, 69, 82, 152, 212, 221,
temperature, 41, 53, 94, 96, 97, 99, 101, 103, 104, 234, 239, 265
106, 107, 108, 109, 110, 111, 114, 116, 117, 118, variations, 37, 38, 39, 41, 44, 45, 46, 48, 49, 50, 53,
123, 124, 125, 139, 140, 141, 142, 143, 144, 145, 55, 56, 57, 58, 68, 123, 125, 132, 146, 159, 211
146, 227, 228, 229, 230 vector, 65, 70, 76, 79, 82, 152, 210, 228, 229, 230,
testing, 55, 59, 121, 148 232
thermodynamic equilibrium, 7 vehicles, 37, 38
thermodynamics, 9, 167 velocity, 4, 69, 83, 162, 210, 215, 217, 227, 230, 275
time increment, 262 vibration, 69, 89, 90, 150, 167, 197, 199
time series, 28, 29 viscoelastic properties, 199
torus, 9, 13 viscosity, 19, 181, 210, 220
tracks, 146
trajectory, 152, 159, 222
transduction, 247 W
transformation, 7, 26, 38, 41, 70, 73, 76, 77, 79, 192,
204, 212 water, ix, 158, 159, 161
transmission, 80, 247 wave propagation, 168
transport, 2, 3, 4, 5, 9, 15, 230, 239, 240, 241, 246 wear, 150, 154
transport processes, 230, 239 working conditions, 39, 40, 43, 45, 46, 56
transportation, 3, 10, 18

Fractional Calculus - Applications

Uploaded by

Copyright:

Available Formats

Fractional Calculus - Applications

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

Fractional Calculus - Applications

Uploaded by

Copyright:

Available Formats

MATHEMATICS RESEARCH DEVELOPMENTS

Additional books in this series can be found on Nova’s website

Additional e-books in this series can be found on Nova’s website

ROY ABI ZEID DAOU

NOTICE TO THE READER

LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA

Published by Nova Science Publishers, Inc. † New York

Foreword Alain Oustaloup vii

Chapter 2 Dynamics of Fractional Order Chaotic Systems 23

Chapter 3 Pressure Control of CNG Engines by Non-integer Order 37

Chapter 4 Linear Integer Order System Control by Fractional PI-State 65

Chapter 5 From the Formal Concept Analysis to the Numerical Simulation 93

Chapter 6 Temperature Control of a Diffusive Medium Using the CRONE 123

Chapter 7 Adaptive Second-Order Fractional Sliding Mode Control 149

Chapter 8 Features of Fractional Operators Involving Fractional Derivatives 165

Chapter 9 Theory of Diffusive Stresses Based on the Fractional 227

Chapter 10 Modelling Drug Effect Using Fractional Calculus 243

Chapter 11 Fuzzy Fractional PID Controllers: Analysis, Synthesis 259

THE HEURISTIC POWER OF THE NON INTEGER

2. A Nice Equation for an Heuristic Power

()d i  const , (1)

We maintain that, including in a majority of experimental cases, it exists a direct link

3. SWOT Method, Non Integer Diff-Integral and Co-Dimension

cannot be reduced to the diffusive formulation

() 2 i 1/ d  const . (9)

or, in Fourier space, the geometric co-density of energy

is obviously the Hausdorff-Mandelbrot dimensional content of the co-fractal space. Let us

4. The Generalization of the Exponential Concept

This is a transformation obtained by a geometrical reversing of a straight line with respect

Then, we can confirm that

To give a content to this remarks it is possible to refer to the impedance related to

5. Diffusion Under Field

is then precisely the impedance. As shown by J. P. Badiali in the frame of statistical

6. Riemann Zeta Function and Non-Integer Differentiation

This operation is a canonical surjection, that is to say an epimorphism. A part of

7. Auto Organization and Emergence

R1 / 2  R cos / 2  R  R cos   R (27)

Figure 7. Schematic representation of the problematic of „entropy‟ production on self-similar structure,

(Le Méhauté, 1974) Le Méhauté A, Application à la hiérarchie départementale française d‟un

DYNAMICS OF F RACTIONAL O RDER

Keywords: Chaos, fractional order, stability

|arg(Eig(J))| > απ/2. (2.3)

Theorem 2.3. [18] Consider the polynomial

Define the descriminant for equation (2.7) as

D(P ) = 18a1 a2 a3 + (a1 a2 )2 − 4a3 (a1 )3 − 4(a2 )3 − 27(a3 )2 . (2.8)

|arg(λ)| > απ/2 (2.9)

2. If D(P ) < 0, a1 ≥ 0, a2 ≥ 0, a3 > 0, α < 2/3 then (2.9) is satisfied.

2.2. Numerical Method for Solving Fractional Differential Equations

Dα y(t) = f (t, y(t)) , 0 ≤ t ≤ T, (2.10)

The IVP (2.10)–(2.11) is equivalent to the Volterra integral equation

The preliminary approximation yhP (tn+1 ) is called predictor and is given by

maxj=0,1,··· ,N |y(tj ) − yh (tj )| = O(hp ), (2.16)

where p = min (2, 1 + α).

where a, b, c and e are constants. We propose a modification in system (3.17) involving

1. Symmetry. The system is symmetric about z-axis because the transformation

The system is dissipative if c < a + b.

3. Non-generalized Lorenz system. The generalized Lorenz system (GLS) discussed

where X = [x, y, z]T , λ3 ∈ ℜ and A is a 2 × 2 matrix with eigenvalues λ1 , λ2 ∈ ℜ

is having different structure than the system (3), it is not a GLS.