Available online at www.sciencedirect.

com

Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

www.elsevier.com/locate/cnsns

Modelling of hysteresis using Masing–Bouc-Wen’s

framework and search of conditions for the chaotic responses
Jan Awrejcewicz a, Larisa Dzyubak b,*
, Claude-Henri Lamarque c

a
Technical University of Łódź, Department of Automatics and Biomechanics, Stefanowskiego 1/15, 90-924 Łódź, Poland
b
Department of Applied Mathematics, National Technical University ‘‘Kharkov Polytechnic Institute’’,
21 Frunze St., 61002 Kharkov, Ukraine
c
Ecole Nationale des Travaux Public de l’Etat, DGCB/LGM, Rue Maurice Audin, F69518 Valux-en-Velin, Cedex, France

Received 17 March 2006; received in revised form 7 September 2006; accepted 8 September 2006
Available online 30 October 2006

Abstract

In the present work hysteresis is simulated by means of internal variables. It was shown that Masing’s imitating mech-
anism of the energy dissipation presented in the differential equations of Bouc-Wen’s structure allows to simulate hysteresis
from very different fields. The constructed analytical models of different types of hysteresis loops are simple, enable major
and minor loops reproducing and provide a high degree of correspondence with experimental data. The models of such
structure are convenient for the further investigation. Hysteretic systems under harmonic excitation described by models
of such structure may reveal chaotic behaviour. Using an effective algorithm based on analysis of the wandering trajecto-
ries [1–4,22,23], an evolution of chaotic behaviour regions of oscillators with hysteresis is presented in various parametric
planes. Substantial influence of a hysteretic dissipation value on the form and location of these regions, and also restraining
and generating effects of the hysteretic dissipation on a chaos occurrence are ascertained. Conditions for pinched hysteresis
are defined.
 2006 Elsevier B.V. All rights reserved.

PACS: 62.30.+d; 62.40.+i; 43.40.+s; 81.40.Jj

Keywords: Hysteresis; Modeling; Chaos; Parametric spaces

1. Introduction

It is known, hysteresis phenomenon is peculiar to very different processes in nature and hysteretic systems
are characterised above all as systems with memory and energy dissipation. If a hysteretic system is considered
as a ‘‘black box’’, the energy dissipation is become apparent in the output delay with respect to the input in the
input-output correlation. The problem of simulation and investigation of hysteresis occurs in many fields for

*
Corresponding author.
E-mail addresses: awrejcew@p.lodz.pl, k16@p.lodz.pl (L. Dzyubak).

1007-5704/$ - see front matter  2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.cnsns.2006.09.003
940 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

mechanical, engineering, physical, biological and even for sociological and economic systems. Therefore in
many cases an abstracted hysteresis is assumed as a basis of such investigation and ascertained properties,
principles, developed techniques are used in the various branches of science.
As a rule, the study of any phenomenon begins with experiments, observation and cumulation of data
about its properties (certainly any metaphysical ideas are not added in these frames). Then either basic prin-
ciples are formulated or the second stage is induced. It is a construction of adequate analytical models describ-
ing behaviour of the system. And the third stage is study of the analytical models with following experimental
testing. The same phenomenon may be considered within the framework of various models as, for instant, a
motion may be considered either within the classical or relativistic mechanics conception. Just as non-linear
dynamics supersedes linear dynamics so far as chaotic behaviour of a system is studied. Many questions arise
at once the presence of like chaos behaviour is ascertained. May be is it an errors of the experiment or numer-
ical methods caprices? Or may be is it a consequence of baseless complicating theoretical model? How the sys-
tem will behave itself in the reality? The present work intended to consider in a complex the modeling of
hysteresis, investigation of the analogous dynamic models and search of conditions for chaos occurring.
So, the present paper includes four sections. Section 2 is devoted to simulation of hysteretic systems. There
are a lot of different phenomenological approaches for hysteresis modelling according to which a system is
viewed as a black box and the system’s output (or response) is modelled with the use of analytical expressions
(e.g. exponential, hyperbolae, polynomials, arctangent, etc.) or differential equations supposing that the input
of the system is known. General mathematical models of hysteresis are presented and discussed in the mono-
graphs [5,6]. Recent results on analysis of hysteretic systems including modelling, experiments, dynamic
response and applications can be found in [7].
Recently a large number of publications [8–16] are devoted to hysteresis simulation because, it is known,
that hysteretic systems are complicated for investigation and various difficulties occur during existent models
applying. The question of multi-purpose and generally valid models describing the wide spectrum of hysteretic
phenomena is still open. The authors of the recent works see among basic defects of some existing models the
following: limitation of accuracy, high computational intensity, inability to memorise the material history
fully, applicability to narrow class of problems and others [13–16].
In the present work hysteresis is simulated by means of additional state variables (internal variables) using
Masing–Bouc-Wen’s framework [17–20]. Masing’s mechanical model is reduced to Bouc-Wen’s form of dif-
ferential equations. The models present fast numerical convergence and applications to different types of hys-
teresis loops confirmed that models with internal variables are well appropriate to simulate hysteresis from
very different fields. In particular the behaviour of magnetorheological/electrorheological (MR/ER) fluids
(which are known as smart materials) in a damper/absorber [21] is simulated. The developed model is effective
and contains principally less parameters than, for example, Bouc-Wen or Spencer models [16]. Other examples
are hysteresis in shape-memory alloys [14] (superelastic behaviour of an NiTi polycrystalline helix) and mod-
elling of stress–strain hysteresis in a steel rope including minor loops reproducing. In all cases the developed
models are computationally economical and provide a high degree of correspondence with experimental data.
The structure of the developed models is sufficiently convenient for subsequent analysis. It turned out dur-
ing simulation that for these hysteretic loops (corresponding to the mentioned experiments) a single internal
variable is necessary. It is known that if a phase space is bi-variate, solutions of autonomous systems cannot be
chaotic. However, under forced oscillations current lines may get entangled and chaotic motion becomes pos-
sible. Thereupon in Section 3 the abstracted Masing and Bouc-Wen hysteretic models under external excita-
tion are investigated using new methodology [1–4,22,23] based on analysis of the wandering trajectories. These
models are discontinuous and contain high non-linearities with memory dependent properties. Investigations
by analytical methods lead to a conclusion that in these systems chaos cannot occur (see, for example, detailed
review in the paper [24]). In the recent works [8,24,25] the authors have achieved through a bifurcations dia-
gram analysis that in such systems chaos can appear. They have found diapasons of an external excitation
frequency X, which correspond to chaotic motion. However, as it is pointed in [26], control parameter spaces
are investigated not sufficiently (at present they are not studied practically) and it is recommended a more
extensive numerical survey of the control parameter spaces. Standard procedures of Lyapunov exponents cal-
culations (even adapted for discontinuous systems Lyapunov exponents calculations procedure, which is
known as Müller’s procedure) are not widely adapted to hysteretic systems.
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 941

The algorithm of quantifying regular and chaotic dynamics, based on analysis of the wandering trajectories,
is more simple and faster from a computational point of view comparing with standard procedures and allows
sufficiently accurate to trace regular/irregular responses of the hysteretic systems. The evolution of chaotic
behaviour regions of oscillators with hysteresis is presented in various control parameter spaces. Substantial
influence of a hysteretic dissipation value on possibility of chaotic behaviour occurring in the systems with
hysteresis is shown. The restraining and generating effects of the hysteretic dissipation on a chaotic behaviour
occurring are demonstrated. Conditions for pinched hysteresis are also defined. It should be emphasised that
our approach can be also applied to control systems with impacts (for instance, reference [27]) and hysteresis,
as well as purely theoretical approach for quantification of non-smooth dynamic systems proposed recently
[28] can be extended to fit also hysteretic systems.

2. Hysteresis simulation by means of internal variables

Modelling of systems showing hysteresis based on the parametric approaches intends an use of various lin-
ear and non-linear elements which are able to simulate memory in the system as well as the energy dissipated in
each cycle. In a physical sense the energy losses can be described by including into the model, for example,
dashpots, friction elements, springs and other mechanical elements. Thus, the classical Masing model of hys-
teresis merges element of Hooke (elastic body model) and some number of St. Venant’s elements (plastic body
models) in parallel. The Biot model of hysteresis includes some number of Newton’s elements (Newton flow
models) instead of St. Venant’s elements. Not infrequently successive joints of mechanical (physical) elements
are also used as, for example, in the Spencer model which is an extension of the Bouc-Wen model. The param-
eters of each model can be defined during a parameter identification process with experimental data. The
validity of the model is confirmed by the experiment also.
In this section a hysteretic behaviour is simulated using Masing–Bouc-Wen’s framework [17–20]. This imi-
tation mechanism of the energy dissipation allows sufficiently accurate to model loops of a various shape
reflecting the actual behaviour of the hysteretic systems from very different fields. So, let us consider the joint
of N friction elements with the maximum friction forces F1, F2, . . . , FN and springs k0, k1, k2, . . . , kN in parallel
(the Masing model) as it is shown in Fig. 1. Here x is an input (input signal) and z is an output (response) of the
hysteretic system. The delaying in the arrival of the output with the respect to the input can be described with
the aid of the internal variables (forces) y1, y2, . . . , yN.
If absolute value of the internal variable jyij (i = 1, 2, . . . , N) is less then maximum friction force Fi or if both
this value is equal to Fi and the velocity of the input x_ and the force yi have different signs (including the case
when x_ ¼0), then the evolution of the force yi in time is governed by the follow equation:
y_ i ¼ k i x_ for jy i j < F i _ ðjy i j ¼ F i ^ sgnð_xy i Þ 6 0Þ i ¼ 1; 2; . . . ; N : ð1Þ
In all other cases this evolution is read as
y_ i ¼ 0: ð2Þ

k0

y1
F1
k1
z

yN
FN
kN

Fig. 1. Mechanical system with elastic–plastic properties for the hysteresis modelling (the Masing model).
942 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

The follow more general equation is an equivalent to the previous two equations (1) and (2)
1     
y_ i ¼ k i x_  1  sgn y 2i  F 2i  sgnð_xy i Þ 1 þ sgn y 2i  F 2i : ð3Þ
2
Using an approximation
 
 2   y i m
sgn y i  F i     1 for y 2i 6 F 2i ; m 2 Rþ ^ m P 1
2
Fi
(3) can be rewritten in the form of
  m 
1 y 
y_ i ¼ k i x_ 1  ð1 þ sgnðx_ y i ÞÞ i  : ð4Þ
2 Fi
And then the output (or response) of hysteretic system is given by
X
N
zðtÞ ¼ k 0 xðtÞ þ y i ðtÞ: ð5Þ
i¼1

A generalisation of Eqs. (4) and (5) in a case of friction elements and springs with properties of a more
complex character (in particular of non-linear character) yields
X
N
zðtÞ ¼ k 0 ðxÞxðtÞ þ y i ðtÞ;
i¼1
    ð6Þ
 y i m
y_ i ¼ 
Ai ðxÞ  ðbi þ ai sgnð_xÞsgnðy i ÞÞ  x_ ;
F ðxÞ i

where k0(x), Ai(x) P 0, Fi(x) > 0, input signal belongs to admissible codomain x 2 [xmin, xmax], ai, bi 2 R; or it
can be written in even greater general form of
z ¼ pðx; y 1 ; y 2 ; . . . ; y N Þ;
ð7Þ
y_ i ¼ qðx; x_ ; y i Þ i ¼ 1; 2; . . . ; N :
Here, in a general case p and q are non-linear functions of its arguments. They are chosen depending on the
properties of a hysteretic system and a loop form.
Note that accordingly to (6) the structure of the differential equations describing an evolution in time of the
internal variables is very similar to the structure of the differential equations of the Bouc-Wen hysteretic oscil-
lator model (14). There are no any universal ways to chose functions which figure in the mentioned Eq. (6) as
well as the same hysteretic behaviour can be constructed by means of various number of internal variables and
by means of various functions. A choice depends on an intuition and an experience of a researcher.
The parameters of functions p(x, y1, y2, . . . , yN) and qðx; x_ ; y i Þ (i = 1, 2, . . . , N) are determined via a procedure
minimising the criterion function
X   xðc1 ; . . . ; cj ; ti Þ; y 1 ða1 ; . . . ; ak ; ti Þ; . . . ;  2
Uðc1 ; . . . ; cj ; a1 ; . . . ; ak ; . . . ; b1 ; . . . ; bl Þ ¼ p  zi ; ð8Þ
i y N ðb1 ; . . . ; bl ; ti Þ
which characterises an error between the experimental curve and the calculated one. Here zi are responses of a
hysteretic system, which are known from an experiment and the values p(x(c1, . . . , cj, ti), y1(a1, . . . , ak, ti), . . . , yN
(b1, . . . , bl, ti)) are obtained as result of integration of the system which is described by means of model (7). To
minimise the criterion function (8) the method of gradient descent is used. The step-by-step descent to the min-
imum of the criterion function realises in the opposite direction to the criterion function gradient
oU oU oU oU oU oU
grad U ¼ ;...; ; ;...; ; ;...; :
oc1 ocj oa1 oak ob1 obl
During solution of the optimisation problem
Uðc1 ; . . . ; cj ; a1 ; . . . ; ak ; . . . ; b1 ; . . . ; bl Þ ! min
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 943

after a sufficiently good choice of the initial approximation c01 ; . . . ; c0j ; a01 ; . . . ; a0k ; . . . ; b01 ; . . . ; b0l the conver-
gence of an approximation is reached. If in the step-by-step descent
 
oU
~ci ¼ ci  hci jgrad Uj ; i ¼ 1; j;
oci
 
oU
ai ¼ ai  hai
~ jgrad Uj ; i ¼ 1; k;
oai

 
~ oU
bi ¼ bi  hbi jgrad Uj ; i ¼ 1; l
obi
the value of the criterion function U have been increased or have not been changed, it is necessary to decrease
the values of the steps hc1 ; . . . ; hcj ; ha1 ; . . . ; hak ; . . . ; hb1 ; . . . ; hbl .
The models presents the sufficiently fast numerical convergence and applications to different types of hys-
teresis loops confirmed that models with internal variables are well appropriate to simulate hysteresis.
Consider three examples of such applications. As it was mentioned, hysteresis is widely found in nature.
Among others, hysteretic behaviour is peculiar to ‘‘smart’’ materials that more and more attract an attention
of commercial structures. Electrorheological/magnetorheological (ER/MR) fluids belong to number of such
materials. These fluids are used for construction of dampers, which act as interfaces between electronic control
systems and mechanical systems. The dynamic characteristics of ER/MR dampers are strongly non-linear that
is reflected in the numerous results of experimental studies of MRF dampers behaviour [21]. Physical accuracy
and simple for users models for dampers behaviour prediction are necessary for designing and producing of its
effective samples. Due to the non-linearities of hysteresis and jump-type phenomenon the difficulties arise dur-
ing developing of these models [16].
During analysis of seven known parametric models in [16] it have made an attempt to find a compromise
between a simplicity, ‘‘economy’’ of the model and its physical accuracy. However, it was concluded that the
simplest involution model with two parameters could not be applied to simulations of dampers behaviour. The
extension of the Bouc-Wen model proposed by Spencer enables the most accurately to predict an actual MR
damper behaviour, but this model contains nine parameters and is more bulky than the model that will be
suggested here.
We suggest here the developed model with single internal variable y1
zðtÞ ¼ y 1 ðtÞ;
ð9Þ
y_ 1 ¼ ðc1  ðc2 þ c3 sgnð_xÞsgnðy 1 ÞÞjy 1 jÞ_x;

which contains only three parameters ci ði ¼ 1; 3Þ and simulates the actual hysteretic MR damper behaviour in
the damping force vs. velocity plane. This analytical model is simple and provides a high degree of correspon-
dence with experimental data reported in [21]. Comparison of simulated loop with the experimental data is
presented in Fig. 2(a).
The choice of the parameters of the set (9) is based on geometrical similarity of the experimental loop [21]
and loops produced by the hysteretic Bouc-Wen’s model (14). So, in the differential set (6) the function A(x) is
chosen by way of a constant c1 and it is supposed that N = 1. An initial approximation is chosen c1 ¼ c01 ,
c2 ¼ c02 , c3 ¼ c03 . Further parameter identification of the model (9) with the experimental loop is realised (while
in rough outline). In case of need the model is complicated (for loops of other form). Programs for the numer-
ical simulation have been created in the programming system ‘‘Microsoft Visual C++ 6.0’’. Graphical presen-
tation of the results have been realised by means of the ‘‘Data Analysis and Technical Graphics’’ package
‘‘Microcal Origin 5.0’’. All differential sets are solved numerically by means of the fourth-order Runge–Kutta
method.
After the rough parameter identification a fitting of the parameters is begun. The criterion function
X
M
 2
Uðc1 ; c2 ; c3 Þ ¼ y i1  zi ! min
i¼1
944 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

12
1000

9
Damping Force [N]

500

Force (N)
6
0

3
-500

0
-1000
-0.06 -0.03 0.00 0.03 0.06 1.6 1.8 2.0 2.2
Velocity [m/s] Helixpitch (mm/turn)

15

10

0
z

-5

-10

-15
-12 -6 0 6 12
x

Fig. 2. Experimental (red/  ) and simulated (blue/—) (a) hysteresis loop (model (9)) for the magnetorheological damper filled with
MRF-132LD (applied current 0.15 A, frequency 5 Hz); (b) hysteresis in shape-memory alloys (model (10)): superelastic behaviour of an
NiTi polycrystalline helix; (c) stress–strain hysteresis (model (12)) with transient processes for the steel rope (stress (N) vs. strain (mm)).
(For interpretation of the references in colour in this figure legend, the reader is referred to the web version of this article.)

is minimised using the method of gradient descent. Here M is the number of experimental points, zi are re-
sponses of the hysteretic system, which are known from the experiment, y i1 ¼ y 1 ðti Þ are obtained as result
of integration of the model (9). n o
oU oU oU
The criterion function gradient grad U ¼ oc ; ;
1 oc2 oc3
is found numerically choosing steps hc1 ¼ Dc1 ,
hc2 ¼ Dc2 , hc3 ¼ Dc3 .
Adjusted values of the parameters c1, c2, c3 are found by means of the recurrence formulas
 
iþ1 i oU
c k ¼ c k  hc k jgrad Uj ; k ¼ 1; 3:
ock
The iteration process i = 0, 1, 2, . . . is continued until step-by-step descent to the minimum of the criterion
function U (c1, c2, c3) is reached.
Final values of the parameters used in the model (9) for identification of the experimental data are in
Table 1.
The admissible range of definition of the input signal (velocity, m/s) is x 2 [0.04 · 103, 0.04 · 103].

Table 1
Final values of the parameters used in the model (9) for identification of the experimental data
c1 c2 c3
70000 80.7208 3.002
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 945

Next example is simulation of hysteresis in shape-memory alloys. Both various stress–strain and strain–
temperature hysteresis cycles arise during transforming to different martensite crystalline structures by uniax-
ial tension or compression at various temperatures. The developed model (10) with the single internal variable
y1 and five parameters ci ði ¼ 1; 5Þ
zðtÞ ¼ c4 ðxðtÞ  xc Þ þ c5 þ y 1 ðtÞ;
ð10Þ
y_ 1 ¼ ðc1 tanh jx  xc j  ðc2 þ c3 sgnð_xÞsgnðy 1 ÞÞjy 1 jÞ_x
simulates the superelatic behaviour of an NiTi polycrystalline helix (Fig. 2(b)) at a constant temperature [14].
Hysteresis cycle for polycrystal (in contrast to single crystal) has increasing slope. It is a consequence of the
fact that uniaxial loading produces martensite plates multiply oriented. To produce a hysteretic loop of such
form the hyperbolic tangent is used in the model (10). The constant xc = 1.91 is a middle point between xmin
and xmax. Final values of the parameters used in the model (10) for identification of the experimental data
reported in [14] are in Table 2.
The admissible range of definition of the input signal (helixpitch, mm/turn) is x 2 [1.25, 2.25].
Further we consider the stress–strain hysteresis in a steel rope. During force upon steel rope some transient
processes are reflected in minor loops reproducing. In this case there are two (or more) ways to simulate these
hysteretic transients in the steel rope. The first one is to increase number of internal variables. But it leads to a
complication of the model and increase in number of parameters.
The model with five parameters, which contains single internal variable y1 and quite well describes the
major stress–strain hysteresis loop, is
zðtÞ ¼ c4 xðtÞ þ c5 þ y 1 ðtÞ;
ð11Þ
y_ 1 ¼ ðc1  ðc2 þ c3 sgnð_xÞsgnðy 1 ÞÞjy 1 jÞ_x:
It presents fast numerical convergence. However, when the transient processes are taken into account, the
model (11) does not enable to reproduce minor loops properly. They are not closed, though losses, which con-
nected with the transient processes, are presented. The quite accurate minor loops reproducing has became
possible after introducing in the model (11) of the additional value zc, which characterises the geometrical cen-
tre of each minor loop
zðtÞ ¼ c4 xðtÞ þ c5 þ y 1 ðtÞ;
ð12Þ
y_ 1 ¼ ðc1  ðc2 þ c3 sgnð_xÞsgnðy 1  zc ÞÞjy 1  zc jÞ_x:
The model (12) enables to simulate minor loops (Fig. 2(c)) and presents fast numerical convergence. The val-
ues of the parameters for the model (12), which are obtained during minimisation of the criterion function (8),
are in Table 3.
The admissible range of definition of the input signal (strain, mm) is x 2 [12, 12]. Probably the models ((9),
(10), (12)) will describe the actual behaviour of the hysteretic systems adequately and in the cases when the
input signals exceed the ranges of definition, but it should be verified by experiments.

Table 2
Final values of the parameters used in the model (10) for identification of the experimental data
c1 c2 c3 c4 c5
82.8007 0.926997 3.33899 6.17671 4.88777

Table 3
Final values of the parameters used in the model (12) for identification of the experimental data
c1 c2 c3 c4 c5
2.22254 0.0010226 0.338787 0.387749 1.45286
946 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

One can see, the structure of all constructed models is the same. In the next section we are considering
abstracted hysteretic models of Masing’s and Bouc-Wen’s type under harmonic excitation. The main aim
of the investigation will be to find regions of irregular motion in various parametric planes and conditions
for pinched hysteresis.

3. Evolution of chaotic behaviour regions in control parameter planes of the Masing and Bouc-Wen hysteretic
oscillators. conditions for pinched hysteresis

Stability of motion depends on all parameters of the considered hysteretic models including initial condi-
tions. We succeeded sufficiently accurate to trace irregular responses of the non-autonomous Masing and
Bouc-Wen hysteretic oscillators. The quantifying is realised using the approach based on analysis of the wan-
dering trajectories [2–4,22,23]. The main idea of this method is to consider nearby trajectories of the motion
and to trace cases of admissible/inadmissible divergence of the trajectories in sense of regularity of the motion.
It have been obtained the full picture of the evolution of chaotic behaviour regions and regions for pinched
hysteresis in various control parameter planes as an amplitude of external excitation vs. frequency (X, F)
and an amplitude of external excitation vs. damper coefficient (l, F) with increasing of hysteretic dissipation.
The control parameter planes as an amplitude vs. hysteretic dissipation value (m, F) and (d, F) for the fixed fre-
quency and damping coefficient are investigated too. This investigation allowed observing of a wealthy variety
of hysteretic oscillators behaviour and also to ascertain two important phenomena as a restraining and gen-
erating effects of the hysteretic dissipation on a chaotic behaviour occurring.
In the non-autonomous case an external periodic excitation with an amplitude F and frequency X acts on
the abstracted mass m which oscillates along an inertial base. This oscillator possesses hysteretic properties
and it is supposed that there is a linear viscous damper with a coefficient 2l.
The following set of differential equations governs a motion of the Masing oscillator:
x_ ¼ y;
y_ ¼ 2ly  ð1  mÞgðxÞ  mz þ F cos Xt; ð13Þ
z  zi
z_ ¼ g0 y:
2
In the above m 2 [0, 1]; gðxÞ ¼ ð1dÞx 1 þ dx; R = (1  m)g(x) + mz is total restoring force with non-linear elastic
ð1þjxjn Þn
part (1  m)g(x) and with hysteretic part mz. The case m = 1 corresponds to maximum hysteretic dissipation and
m = 0 corresponds to elastic behaviour of the oscillator. The parameter d characterises a ratio between the
post- and pre-yielding stiffness. The parameter n governs the smoothness of the transitions from the elastic
to the plastic range. A couples ±(xi, zi) represent the velocity reversal points at x_ ¼ 0. Accordingly to Masing’s
rule which is disseminated on the case of steady-state motion of the hysteretic oscillator, the loading/unload-
ing branches of a hysteresis loop are similar geometrically. So, if f(x, z) = 0 is the equation of a virgin loading
curve, then the equations f((x ± xi)/2, (z ± zi)/2) = 0 describe loading/unloading branches of the hysteresis
loop. In the case of non-steady state motion of the Masing’s oscillator it is supposed, that the equation of
any hysteretic response curve can be obtained by applying the original Masing rule to the virgin loading curve
using the latest point of velocity reversal.
A motion of Bouc-Wen oscillator is governed by the following set of differential equations:
x_ ¼ y;
y_ ¼ 2ly  dx  ð1  dÞz þ F cos Xt; ð14Þ
z_ ¼ ½k z  ðc þ bsgnðyÞsgnðzÞÞjzjn y;
where R = dx + (1  d)z is total restoring force; the parameters (kz, b, n) 2 R+ and c 2 R govern the shape of
the hysteresis loop. The parameters d and n have the same sense as in the case of the Masing model.
The systems (13) and (14) under considerations are non-smooth dynamical systems. The right-hand sides of
these systems have discontinuities at y ¼ x_ ¼ 0 and y = 0, z = 0 correspondingly. It should note, that N. Lev-
inson [29,30] was one of the first, who observed and theoretically proved the existence of chaos using a non-
smooth ordinary differential equation. However, the presence of chaos in hysteretic systems had been ascer-
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 947

tained only recently [24,25]. The point set of discontinuities need of a special attention during a numerical
approximation. It is possible [31,32] firstly, to find the correct solution naturally, using, for example, the
Runge–Kutta algorithm taking sufficiently large amounts of integration points (in this case for an optimisa-
tion of the calculations process the method of the adaptive step size can be used). Secondly, it is possible to
construct a, so-called, ‘‘band’’ around the discontinuous hyper-surface. Within the band the vector field is
such that the flow is pushed to the middle of the band, thus to the discontinuous hyper-surface. And, thirdly,
as an alternative, one can replace the discontinuous vector field by a smoothed vector field. It is known that in
a smooth linear dynamical system chaos cannot occur. N. Levinson had shown that in a linear dynamical sys-
tem with two parallel discontinuous lines chaotic motions would occur. So, an application of a smoothing
algorithm may distort the picture of chaotic behaviour analysis. Therefore in this work we have preferred a
natural scheme of the numerical approximation, which will be described below in detail.
At first let describe briefly the approach based on analysis of the wandering trajectories in view of the state
vector of the systems (13), (14) for x 2 R3. A chaotic behaviour of non-linear deterministic systems supposes a
wandering of trajectories of motion around the various equilibrium states. They are characterised by unpre-
dictability and sensitive dependence on the initial conditions. By analyzing trajectories of motion of these sys-
tems, it is possible to find the chaotic vibrations regions in control parameters space.
The continuous dependence property on the initial conditions x(0) = x(t0) of a solutions of the set (13) or
(14) will be used: For every initial condition x(0), x ~ð0Þ 2 R3 , for every number T > 0, no matter how large, and
for every preassigned arbitrary small  e >ð0Þ 0 it is possible to indicate a positive number d > 0 such that if the
distance q between x(0) and x ~ð0Þ , q xð0Þ ; x
~ < d, and jtj 6 T, the following inequality is satisfied:
~ðtÞÞ < e:
qðxðtÞ; x
That is if the initial points are chosen close enough, than during the preassigned arbitrary large time interval
T 6 t 6 T the distance between simultaneous positions of moving points will be less given positive number e.
For the sake of tracing chaotic and regular dynamics, it is supposed that with the increase of time all tra-
jectories remain in the closed bounded domain of a phase space, i.e.
9C i 2 R : maxt jxi ðtÞj 6 C i ; i ¼ 1; 2; 3:
To analyze trajectories of the sets (13) and (14), the characteristic  vibration amplitudes Ai of components of
the motion are introduced Ai ¼ 12 maxt1 6t6T xi ðtÞ  mint1 6t6T xi ðtÞ. Here and below index number i run over
three values corresponding to three generalised coordinates x, y, z. [t1, T]  [t0, T] and [t0, T] is the time inter-
val, in which the trajectory is considered. The interval [t0, t1] is the time interval, in which all transient pro-
cesses are damped.
For the sake of our investigation it seems the most convenient to use the embedding theorem and to con-
sider an three-dimensional parallelepiped instead of a hyper-sphere with the center at point x. Two neighbour-
ing initial points x(0) = x(t0) and x ~ðt0 Þ (x = (x, y, z)T or x = (x1, x2, x3)T) are chosen in the three-
~ð0Þ ¼ x
ð0Þ  ð0Þ ð0Þ 
dimensional parallelepiped P dx ;dy ;dz ðx Þ such that xi  ~xi  < di , where di > 0 is small in comparison with
Ai. In the case of regular motion it is expected that the ei > 0 used in inequality jxi ðtÞ  ~xi ðtÞj < ei is also small
in comparison with Ai. The wandering orbits attempt to fill up some bounded domain of the phase space. At
instant t0 the neighbouring trajectories diverge exponentially afterwards. Hence, for some instant t1 the abso-
lute values of differences jxi ðtÞ  ~xi ðtÞj can take any values in closed interval [0, 2Ai]. An auxiliary parameter a
is introduced, 0 < a < 1. aAi is referred to as divergence measures of observable trajectories in the directions of
generalised coordinates and with the aid of parameter a one have been chosen, which is inadmissible for the
case of ‘‘regularity’’ of the motion. The domains, where a chaotic behaviour of considered systems is possible,
can be found using following condition:
9t 2 ½t1 ; T  : jxðt Þ  ~xðt Þj > aAx : ð15Þ
If this inequality is satisfied in some nodal point of the sampled control parameter space, then such motion is
relative to chaotic one (including transient and alternating chaos). The manifold of all such nodal points of the
investigated control parameter space set up domains of chaotic behaviour of the considered systems.
Now let consider the numerical realisation of this algorithm. All programs have been written, compiled and
executed in the programming system ‘‘Microsoft Visual C++ 6.0’’. Graphical presentation of the results have
 948
J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958
Table 4
Numerical simulation parameters description for the Masing hysteretic oscillator (13)
Uniform sampling of control parameter planes
Parametric plane Number of nodal points on the parametric plane Parameters of the set (13) that are fixed
(m, F) 100 · 100; X = 0.16, l = 0.0005, d = 0.05, n = 10.0;
0 6 m 6 1, 0.01 6 F 6 1.61 mi = iDm (i = 0, 1, . . . , 100), Fj = jDF(j = 0, 1, . . . , 100); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0
(Fig. 3(a)) Dm = 102, DF = 1.6 · 102
(X, F) 100 · 100; l = 0.0005, d = 0.05, n = 10.0;
0.01 6 X 6 0.66, 0.01 6 F 6 1.51 Xi = iDX (i = 0, 1, . . . , 100), Fj = jDF (j = 0, 1, . . . , 100); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0;
(Fig. 5) DX = 6.5 · 103, DF = 1.5 · 102 m = 0 (a), m = 0.5 (b), m = 0.8 (c)
(l, F) 100 · 100; X = 0.12, d = 0.05, n = 10.0;
0.01 6 l 6 0.28, 0.01 6 F 6 1.91 li = iDl (i = 0, 1, . . . , 100), Fj = jDF (j = 0, 1, . . . , 100); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0;
(Fig. 6) Dl = 2.7 · 103, DF = 1.9 · 102 m = 0 (a), m = 0.5 (b), m = 0.8 (c)

Integration of the governing (13) by the fourth-order Runge–Kutta method

Time period for the Time period, where transient processes are Number of nodal points, Nearby initial conditions Value of a in the
simulation damped integration step inequality (15)
T ¼ 200p
X [t0, t1] = [0, T/2] N = 20 · 103, ~xð0Þ ¼ xð0Þ 5  103 Ax ; a = 1/3
h ¼ Xp  102 ~y ð0Þ ¼ yð0Þ 5  103 Ay ;
~zð0Þ ¼ zð0Þ 5  103 Az
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958
Table 5
Numerical simulation parameters description for the Bouc-Wen hysteretic oscillator (14)
Uniform sampling of control parameter planes
Parametric plane Number of nodal points on the parametric plane Parameters of the set (13) that are fixed
(d, F) 150 · 75; X = 0.2, l = 0, kz = 0.5, c = 0.3, b = 0.005, n = 1.0;
0 6 d 6 0.15, 0 6 F 6 2 di = iDd (i = 0, 1, . . . , 150), Fj = jDF (j = 0, 1, . . . , 75); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0
(Fig. 3(b)) Dd = 103, DF  2.7 · 102
(X, F) 120 · 120; kz = 0.5, c = 0.3, b = 0.005, n = 1.0, l = 0;
0.0 6 X 6 0.36, 0.01 6 F 6 2.05 Xi = iDX (i = 0, 1, . . . , 120), Fj = jDF (j = 0, 1, . . . , 120); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0;
(Fig. 8) DX = 3 · 103, DF  1.7 · 102 d = 0.03 (a), d = 0.01 (b), d = 0.001 (c)
(l, F) 120 · 160; X = 0.2, kz = 0.5, c = 0.3, b = 0.005, n = 1.0;
0.001 6 l 6 0.03, 0.01 6 F 6 1.91 li = iDl (i = 0, 1, . . . , 120), Fj = jDF (j = 0, 1 . . . , 160); x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0;
(Fig. 9) Dl = 2.5 · 104, DF  1.19 · 102 d = 0.03 (a), d = 0.01 (b), d = 0.001 (c)

Integration of the governing Eq. (14) by the fourth-order Runge–Kutta method

Time period for Time period, where transient Number of nodal points, Nearby initial conditions Value of a in the
the simulation processes are damped integration step inequality (15)
T ¼ 200p
X [t0, t1] = [0, T/2] N = 8 · 103, ~xð0Þ ¼ xð0Þ 5  103 Ax ; a = 1/3
p
h ¼ 40X ~y ð0Þ ¼ yð0Þ 5  103 Ay ;
~zð0Þ ¼ zð0Þ 5  103 Az

949
950 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

1.6

1.2

Amplitude F
0.8

hysteretic dissipation
0.4
increment

0.0 0.2 0.4 0.6 0.8 1.0

Hysteretic dissipation
coefficient ν

2.0

1.6
δ cr
Amplitude F

1.2

0.8

0.4
hysteretic dissipation
increment
0.00 0.03 0.06 0.09 0.12 0.15
δ

Fig. 3. Influence of hysteretic dissipation on chaos occurring in the case of the (a) Masing hysteretic oscillator (13) (X = 0.16, l = 0.0005,
d = 0.05, n = 10.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0); (b) Bouc-Wen hysteretic oscillator (14) (X = 0.2, l = 0, kz = 0.5, c = 0.3, b = 0.005,
n = 1.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0).

εr

εr

Fig. 4. Hysteresis loops with various dissipation properties.

 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 951

been realised by means of the ‘‘Data Analysis and Technical Graphics’’ package ‘‘Microcal Origin 5.0’’. The
full information about parameters of the numerical approximation is in Tables 4 and 5. Let begin with the
investigation of the control parameter plane (m, F) ‘‘an amplitude of external excitation’’ vs. ‘‘hysteretic dissi-
pation value’’ for the Masing oscillator (13). The part of this plane (0 6 m 6 1, 0.01 6 F 6 1.61) have been sam-
pled by means of a uniform rectangular grid. For this aim two families of straight lines have drawn through
dividing points of the axes:
m ¼ mi ¼ iDm ði ¼ 0; 1; . . . ; LÞ;
F ¼ F j ¼ jDF ðj ¼ 0; 1; . . . ; MÞ:

Here Dm = 102, DF = 1.6 · 102, L = 100, M = 100. Then in the nodal points (mi, Fj) of the constructed grid
the governing Eq. (13) are twice solved numerically with two nearby initial conditions by the fourth-order
Runge–Kutta method, which has an error of order h5 (h is an integration step size). Initial conditions of
the nearby trajectories are distinguished by 0.5% with ratio to characteristic vibration amplitudes, e.g. the
starting points of these trajectories are in the three-dimensional parallelepiped ðjxðt0 Þ  ~xðt0 Þj < 0:005Ax ,
j_xðt0 Þ  ~x_ ðt0 Þj < 0:005Ax_ , jzðt0 Þ  ~zðt0 Þj < 0:005Az Þ:
~xð0Þ ¼ xð0Þ 5  103 Ax ;
~y ð0Þ ¼ yð0Þ 5  103 Ay ;
~zð0Þ ¼ zð0Þ 5  103 Az :

Fig. 5. Evolution of the chaotic regions (olive) for the Masing hysteresis model (13) in the (X, F) plane with increasing of the hysteretic
dissipation value (a) m = 0; (b) m = 0.5; (c) m = 0.8. Regions of pinched hysteresis are dark yellow (er/e < 1%) and yellow (1% < er/e < 5%).
The parameters l = 0.0005, d = 0.05, n = 10.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0 are fixed for all cases. (For interpretation of the references
in colour in this figure legend, the reader is referred to the web version of this article.)
952 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

Ai (index number i run over coordinates x, y, z) are calculated for all nodal points (mi, Fj) of the constructed
grid simultaneously with the integration of the governing equations. The time period for the simulation T is of
200p
X
non-dimensional time units. During computations, one half of the time period T corresponds to the time
p
interval [t0, t1], where transient processes are damped. The integration step size is 100X . After the integration of
the governing equations (13) the inequality (15) is verified. The parameter a is chosen to be equal to 1/3. In
that way, the manifold of the nodal points (mi, Fj), for which the inequality (15) is satisfied, set up regions of
chaos.
The Masing oscillator (13) is non-linear both in the case of a pure elastic behaviour without hysteretic dis-
sipation (m = 0) and in the case of motion with hysteretic dissipation (m > 0). And therefore at m = 0 chaos may
exists too. Confirming this statement, Fig. 3(a) represents the regions where chaotic behaviour of the Masing
oscillator (13) is possible in the (m, F) plane. The region is constricted with the hysteretic dissipation increment.
It is a natural result because hysteresis implies dissipation, and any dissipation as a rule suppresses (reduces) a
chaotic state of the system.
Additionally in all parametric planes the conditions for pinched hysteresis are defined. Pinched hysteresis
possesses reduced dissipation properties and is consequence of a residual phenomena diminution or ‘‘delay’’
reduction of the output relative to the input in a hysteretic system. In this connection an analysis of pinching
phenomena is of great importance for various practical applications. It have been ascertained that for the non-
zero fixed value of hysteretic dissipation (which in changing from minimum to maximum is able to produce
behaviour of the system with/without delay) and for the fixed parameters which govern a shape of hysteretic
loop it is possible to choose ‘‘appropriate’’ hysteretic process with wishful residual phenomena with the aid of
an amplitude and frequency of an external periodic excitation.

Fig. 6. Evolution of the chaotic regions (blue) for the Masing hysteresis model (13) in the (l, F) plane with increasing of the hysteretic
dissipation value (a) m = 0; (b) m = 0.5; (c) m = 0.8. Regions of pinched hysteresis are dark yellow (er/e < 1%) and yellow (1% < er/e < 5%).
The parameters X = 0.12, d = 0.05, n = 10.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0 are fixed for all cases. (For interpretation of the references in
colour in this figure legend, the reader is referred to the web version of this article.)
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 953

The relation er/e have been chosen as value that characterises a pinch in a hysteretic loop. As shown in
Fig. 4 e is the distance between the projection of the velocity reversal points ±(xi, zi) to the input-axis, er is
the distance characterising the residual phenomena. It is clear that no sense to discuss dissipation properties
in the case of chaotic behaviour of hysteretic oscillators.
Figs. 5 and 6 display the evolution of chaotic behaviour domains with increasing of hysteretic dissipation
value in the ‘‘amplitude vs. frequency of external periodic excitation’’ and ‘‘amplitude vs. damping coefficient’’
parametric planes. Numerical simulation parameters description is in Table 4. The (X, F) and (l, F) planes had
been uniformly sampled by 100 · 100 nodal points in the rectangles (0.01 6 X 6 0.66; 0.01 6 F 6 1.51) and
(0.01 6 l 6 0.28; 0.01 6 F 6 1.91), respectively. The time period for the simulation T is of 200p X
non-dimen-
sional time units. During computations, one half of the time period T corresponds to the time interval

3 3

2 2

1 1

0 0
z
v

-1 -1

-2 -2

-3 -3
-15 -10 -5 0 5 10 15 -15 -10 -5 0 5 10 15
x x

6 2

3 1

0 0
z
v

-3 -1

-6 -2
-8 -4 0 4 8 -8 -4 0 4 8
x x

0.4 1.0

0.2 0.5

0.0 0.0
v

-0.2 -0.5

-0.4 -1.0
-0.9 -0.6 -0.3 0.0 0.3 0.6 0.9 -0.9 -0.6 -0.3 0.0 0.3 0.6 0.9
x x

Fig. 7. Phase portraits and hysteresis loops of the Masing hysteretic oscillator (13) in the cases of (a, b) chaotic (X = 0.12, F = 1.27,
l = 0.057, m = 0.5); (c, d) periodic (X = 0.6, F = 0.9, l = 0.0005, m = 0.5) responses; (e, f) phase portrait and a loop of pinched hysteresis
(X = 0.3, F = 0.6, l = 0.0005, m = 0.5). In all cases the parameters d = 0.05, n = 10.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0 are fixed.
954 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

p
[t0, t1], where transient processes are damped. The integration step size is 100X . Initial conditions of the closed
trajectories are distinguished by 0.5% with ratio to characteristic vibration amplitudes. The parameter a is cho-
sen to be equal to 1/3. All domains are multiply connected and have a complex structure. There are some num-
ber of scattered points, streaks and ‘‘islets’’ here. Such structure is characteristic for domains, where chaotic
vibrations are possible.
One can observe the effect of restraining of the chaotic regions with the increasing of the hysteretic dissipa-
tion value in the (X, F) plane (Fig. 5(a)–(c)). The ‘‘quickness’’ of the restraining decreases when m increases. So,
in the case of maximum hysteretic dissipation value m = 1 the chaotic regions in the (X, F) plane are distin-
guished from the regions (c), Fig. 5, non-principally.
In the (l, F) plane (Fig. 6(a)–(c)) the form and location of the chaotic domains are changed depending on
the hysteretic dissipation.
Fig. 7 characterises the obtained domains and demonstrates various character of motion of the Masing
oscillator as chaos and hysteresis loss (a, b), periodic response (c, d) and pinched hysteresis (e, f).
Other situation occurs for the Bouc-Wen oscillator (14) which naturally is linear (when d = 1). So, the hys-
teretic dissipation adding leads to chaotic responses occurring in this system. This unnatural statement (nev-
ertheless dissipation implies more restraining than generating) is confirmed by Fig. 3(b), which represent the
regions where chaotic behaviour of the Bouc-Wen oscillator is possible in the (F, d) plane (numerical simula-
tion parameters description is in Table 5). Note, that chaotic responses of the Bouc-Wen oscillator are not
observed right up till d = 0.4 when the influence of the non-linear terms becomes critical. It demonstrates gen-
erating effect of the hysteretic dissipation on chaos occurring in the hysteretic system, which appears after
some critical value dcr. Figs. 8 and 9 present the evolution of chaotic behaviour regions with increasing of hys-

2.0 2.0

1.5 1.5
Amplitude F

Amplitude F

1.0 1.0

0.5 0.5

0.0 0.0
0.1 0.2 0.3 0.1 0.2 0.3
Frequency Ω Frequency Ω

2.0

1.5
Amplitude F

1.0

0.5

0.0
0.1 0.2 0.3
Frequency Ω

Fig. 8. Evolution of the chaotic regions (pink) for the Bouc-Wen oscillator (14) with hysteresis in the (X, F) plane with increasing of the
hysteretic dissipation value (a) d = 0.03; (b) d = 0.01; (c) d = 0.001 at kz = 0.5, c = 0.3, b = 0.005, n = 1.0, x(0)=0.1, x_ ð0Þ ¼ 0:1, z(0) = 0
and l = 0. Regions of pinched hysteresis are dark yellow (er/e < 1%) and yellow (1% < er/e < 5%). (For interpretation of the references in
colour in this figure legend, the reader is referred to the web version of this article.)
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 955

1.8 1.8

1.6 1.6

Amplitude F
Amplitude F

1.4 1.4

1.2 1.2

0.00 0.01 0.02 0.03 0.00 0.01 0.02 0.03

Damping coefficient μ Damping coefficient μ

1.8

1.6
Amplitude F

1.4

1.2

0.00 0.01 0.02 0.03

Damping coefficient μ

Fig. 9. Evolution of the chaotic regions (dark blue) for the Bouc-Wen oscillator (14) with hysteresis in the (l, F) plane with increasing of
the hysteretic dissipation value (a) d = 0.03; (b) d = 0.01; (c) d = 0.001 at kz = 0.5, c = 0.3, b = 0.005, n = 1.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1,
z(0) = 0 and X = 0.2. Regions of pinched hysteresis are dark yellow (er/e < 1%) and yellow (1% < er/e < 5%). (For interpretation of the
references in colour in this figure legend, the reader is referred to the web version of this article.)

teretic dissipation value in the (X, F) and (l, F) planes correspondingly. After dcr both the form and location of
the chaotic behaviour regions are changed with increasing of the hysteretic dissipation. In the case of maxi-
mum hysteretic dissipation value (when d = 0) the chaotic behaviour regions are the same as in the case
(c), Fig. 8, practically.
Fig. 10 characterises the obtained regions of irregular motion and depicts various responses of the Bouc-
Wen oscillator as chaos (a, b), periodic response (c, d) and pinched hysteresis (e, f). All figures agree well with
the obtained regions of regular/irregular behaviour of hysteretic oscillators and with the regions of pinched
hysteresis.

4. Conclusions

In the present work the hysteretic loops of a various form are simulated by means of additional state vari-
ables (internal variables) using Masing–Bouc-Wen’s framework. This imitation mechanism of the energy dis-
sipation allows sufficiently accurate to model loops of a various shape reflecting behaviour of the hysteretic
systems from very different fields. In particular the behaviour of magnetorheological/electrorheological
(MR/ER) fluids in a damper/absorber is simulated as well as hysteresis in shape-memory alloys (superelastic
behaviour of a NiTi polycrystalline helix) and stress–strain hysteresis with transient processes in a steel rope.
The developed models are effective, enable to produce minor loops, present fast numerical convergence and
provide a high degree of correspondence with experimental data.
The structure of the Masing–Bouc-Wen equations form is sufficiently convenient for the future work with
the models in the investigation process. In the developed in this work hysteretic models chaos can be found
956 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

Fig. 10. Phase portraits and hysteresis loops of the Bouc-Wen hysteretic oscillator (14) in the cases of (a, b) chaotic (X = 0.2, F = 1.38,
l = 0.0022, d = 0.01); (c, d) periodic (X = 0.3, F = 1, l = 0.0, d = 0.001) responses; (e, f) phase portrait and a loop of pinched hysteresis
(X = 0.2, F = 1, l = 0.0, d = 0.03). In all cases the parameters kz = 0.5, c = 0.3, b = 0.005, n = 1.0, x(0) = 0.1, x_ ð0Þ ¼ 0:1, z(0) = 0 are
fixed.

only in the dynamic non-autonomous case. Admittedly chaos exists also in the autonomous case in more com-
plex hysteretic systems with larger number of internal variables. Highly non-linear non-autonomous hysteretic
models of Masing’s and Bouc-Wen’s type with discontinuous right-hand sides are investigated using effective
approach based on analysis of the wandering trajectories. This algorithm of quantifying regular and chaotic
dynamics is more simple and faster from a computational point of view comparing with standard procedures
and allows sufficiently accurate to trace regular/irregular responses of the hysteretic systems. The evolution of
chaotic behaviour regions of oscillators with hysteresis is presented in various control parameter spaces: in the
damping coefficient – amplitude, in the frequency – amplitude of external periodic excitation planes and in the
hysteretic dissipation value – amplitude planes. Substantial influence of a hysteretic dissipation value on pos-
 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958 957

sibility of chaotic behaviour occurring in the systems with hysteresis is shown. The restraining and generating
effects of the hysteretic dissipation on a chaotic behaviour occurring are ascertained. The regions of pinched
hysteresis with various dissipation properties are presented.
Simulation of hysteresis with two or more internal variables, an analysis of possibility to produce chaotic
motion by these models and an address to real experiments for testing are of great interest of future
investigations.

Acknowledgements

This work has been supported by the J. Mianowski Foundation of Polish Science Support, as well as the
Polish Ministry of Science and Higher Education (grant No. 4 T07A 031 28) for the years 2005–2008.

References

[1] Awrejcewicz J, Mosdorf R. Numerical analysis of some problems of chaotic dynamics, WNT, Warsaw, 2003 [in Polish].
[2] Awrejcewicz J, Dzyubak L. Stick-slip chaotic oscillations in a quasi-autonomous mechanical system. Int J Nonlinear Sci Numer Simul
2003;4(2):155–60.
[3] Awrejcewicz J, Dzyubak L. Regular and chaotic behavior exhibited by coupled oscillators with friction, Facta Universitatis. Ser Mech
Autom Control Robot 2003;3(14):921–30.
[4] Awrejcewicz J, Dzyubak L, Grebogi C. A direct numerical method for quantifying regular and chaotic orbits. Chaos, Solitons Fract
2004;19:503–7.
[5] Mayergoyz ID. Mathematical models of hysteresis. Berlin: Springer; 1991.
[6] Visintin A. Differential models of hysteresis. Berlin: Springer; 1994.
[7] Vestroni F, Noori M. Hysteresis in mechanical systems – modelling and dynamic response. Int J Non-Linear Mech 2002;37:1261–459.
[8] Bernardini D, Rega G. Thermomechanical modelling, nonlinear dynamics and chaos in shape memory alloys oscillators. Math
Comput Model Dyn Syst 2005;11(3):291–314.
[9] Kádár G, Szabó G. Hysteresis modelling. J Magn Magn Mater 2000;215–216:592–6.
[10] Kolsch H, Ottl D. Simulation des mechanischen Verhaltens von Bauteilen mit statischer Hysterese. Forsch Ingenieurwes 1993;4:66–71
[in Germany].
[11] Koltermann PI, Righi LA, et al. A modified Jiles method for hysteresis computation including minor loops. Physica B
2000;275:233–7.
[12] Ktena A, Fotiadis DI, et al. A Preisach model identification procedure and simulation of hysteresis in ferromagnets and shape-
memory alloys. Physica B 2001;306:84–90.
[13] Makaveev D, Dupre L, et al. Dynamic hysteresis modelling using feed-forward neural networks. J Magn Magn Mater 2003;254–
255:256–8.
[14] Ortin J, Delaey L. Hysteresis in shape-memory alloys. Int J Non-Linear Mech 2002;37:1275–81.
[15] Ossart F, Hubert O, Billardon R. A new internal variables scalar model respecting the wiping-out property. J Magn Magn Mater
2003;254–255:170–2.
[16] Sapinski B, Filus J. Analysis of parametric models of MR linear damper. J Theor Appl Mech 2003;41(2):215–40.
[17] Masing G. Zur Heynschen Theorie der Verfestigung der Metalle durch verborgen elastische Spannungen. Wiss Veroffentl aus dem
Siemens-Konzern 1923;3(1):231–9 [in German].
[18] Masing G Eigenspannungen und Vertfestigung beim Messing. In: Proceedings of the second international congress of applied
mechanics, Zurich, Switzerland. 1926. p. 332–5 [in German].
[19] Bouc R. Forced vibrations of mechanical systems with hysteresis. In Proceedings of the fourth international conference on nonlinear
oscillations, Prague, Czechoslovakia, 1967.
[20] Wen YK. Method for random vibration of hysteretic systems. ASCE J Eng Mech 1976;120:2299–325.
[21] Sapinski B. Dynamic characteristics of an experimental MR fluid. Eng Trans 2003;51(4):399–418.
[22] Awrejcewicz J, Dzyubak L. Influence of hysteretic dissipation on chaotic responses. J Sound Vibr 2005;284:513–9.
[23] Awrejcewicz J, Dzyubak L. Quantifying smooth and non-smooth regular and chaotic dynamics. Int J Bifur Chaos
2005;15(6):2041–55.
[24] Lacarbonara W, Vestroni F. Nonclassical responses of oscillators with hysteresis. Nonlinear Dyn 2004.
[25] Li HG, Zhang JW, Wen BC. Chaotic behaviors of a bilinear hysteretic oscillator. Mech Res Comm 2002;29:283–9.
[26] Capecchi D, Masiani R. Reduced phase space analysis for hysteretic oscillators of Masing type. Chaos, Soliton Fract
1996;7:1583–600.
[27] Jin C, Fan L, Qiu Y. The vibration control of a flexible linkage mechanism with impact. Comm Nonlinear Sci Numer Simul
2004;9:459–69.
[28] Luo A. A theory for non-smooth dynamic systems on the connectable domains. Comm Nonlinear Sci Numer Simul 2005;10:1–55.
[29] Levinson N. A second order differential equation with singular solutions. Ann Math 1949;50:127–53.
[30] Coddington EA, Levinson N. Theory of ordinary differential equations. New York: McGraw-Hill; 1955.
958 J. Awrejcewicz et al. / Communications in Nonlinear Science and Numerical Simulation 13 (2008) 939–958

[31] Leine RI. van de Vrande BL, vanCampen DH. Bifurcations in nonlinear discontinuous systems. Report WFW 99.010, Vakgroep
Fundamentele Werktuigkunde, Eindhoven, 1999.
[32] Awrejcewicz J, Lamarque C-H. Bifurcations and chaos in nonsmooth mechanical systems. New Jersey, London, Singapore, Hong
Kong: World Scientific; 2003.