Nonlinear Dyn (2012) 67:1337–1356

DOI 10.1007/s11071-011-0072-9

O R I G I N A L PA P E R

Observer-based model reference adaptive control

for unknown time-delay chaotic systems with input
nonlinearity
Hassan Adloo · Navid Noroozi ·
Paknosh Karimaghaee

Received: 7 November 2010 / Accepted: 4 May 2011 / Published online: 2 June 2011
© Springer Science+Business Media B.V. 2011

Abstract Existence of unknown time-delay in the Keywords Chaos control · Model reference adaptive
systems is a drastic restriction that it can menace the control · Observer-based system · Dead-zone and
stability criteria and even deteriorate the performance hysteresis nonlinearity
system. This undesired case would be more inten-
sified if that the uncertain input nonlinearity effects
are also considered. To handle the input nonlineari- 1 Introduction
ties effects (results in dead-zone and/or hysteresis phe-
nomena) and also unknown time-delay in the chaotic During past two decades, problem of chaos control has
systems, this paper presents an observer-based Model generated great deal interesting in the control com-
Reference Adaptive Control (MRAC) scheme for a munity. This comes from the fact that chaos phenom-
class of unknown time-delay chaotic systems with dis- ena in some practical systems such as gyroscopes and
turbances. This new method is a delay-independent plates generate the irregular oscillations which may of-
variable-structure control method which is integrated ten cause harm in the systems. However, having better
with an observer system. The main task of the pro- system performance against this catastrophic situation
posed approach is to accomplish a perfect tracking requires moving the state trajectory to a desired region
procedure such that unknown parameters are adapted through the control methods. From the Ott’s work [1]
via output estimation error. Furthermore, stability of in 1990 as a first effort concerning with the chaos con-
the closed-loop system is achieved by means of the trol, many researchers have developed control meth-
Lyapunov stability theory. Finally, the proposed meth- ods for various chaotic systems via different tech-
niques. Among them, robust control method [2], adap-
ods are applied to some famous chaotic systems to ver-
tive observer-based approach [3], sliding mode con-
ify the effectiveness of the proposed methods.
trol [4], CLF approach [5], and backstepping design
[6] can be mentioned. Nevertheless, most of the rele-
vant works in literature have been investigated on the
H. Adloo () · N. Noroozi · P. Karimaghaee
Department of Electrical and Computer Engineering,
chaotic systems without time delay. It is well known
Shiraz University, 71377-54633, Shiraz, Iran that in many natural systems, especially in the chaotic
e-mail: [email protected] systems, commonly there exists unknown time delay
N. Noroozi which is due to practical limitations in measurements
e-mail: [email protected] devices, transportation lag, computing time, and so on.
P. Karimaghaee However, time-delayed chaotic systems represent poor
e-mail: [email protected] system performance and because of their distributed
1338 H. Adloo et al.

characteristic, require to more complicated control de- systems when they exposed to the unknown input
sign. Tian and Gao [7] presented two model reference nonlinearities (dead-zone or hysteresis nonlinearity),
adaptive control methods for a class of first order time- a model reference adaptive variable-structure control
delay chaotic systems in two cases, known and un- is presented. Furthermore, an adaptive nonlinear ob-
known time delay. Nevertheless, in the previous men- server is designed to estimate the system states. Using
tioned works, except unknown time delay, any distur- Lyapunov stability theory, it is shown that the closed-
bance and/or uncertainties are not considered. Another loop system is stable and the states of the system track
work related to this case was presented by Hua and the desired trajectory. Finally, the effectiveness of the
Guan [8]. In that work, two adaptive and nonadaptive proposed methods is examined by presenting three nu-
control approaches, which are independent of the time merical examples.
delay, are investigated. However, in [8], it is assumed The rest of the paper is organized as follows. Sec-
that all states of the system is available while it is a re- tion 2 represents a lemma and some notations are
strictive assumption in practice. This problem can be reviewed. In Sect. 3, a general form of the time-
handled implicitly by using the observer systems. For delay chaotic systems is introduced and some assump-
example, a fuzzy observer-based adaptive control for tions are also presented. The main proposed adaptive
the chaotic-fuzzy systems was proposed by Ting [9]. method is proposed in Sect. 4. Section 5 contains the
The work has solved the problems of synchronization, simulation results. Section 6 provides the concluding
stabilization, and tracking control. To see the several remarks.
techniques concerning it, refer to [10–12].
Two another physical limitations which reduce sys-
tem performance and even may cause instability are
2 Preliminaries
dead-zone and hysteresis nonlinearities. Generally,
properties of these nonlinearities phenomena exist in
Throughout this paper, we will use some necessary no-
the control system components, for example, the hys-
tations and lemmas which are here presented. Assume
teresis nonlinearity can be observed in electronic re-
w is a vector such that
lay circuits and electromagnetic devices. Moreover,
the dead-zone nonlinearity effects generally appear in
w = [w1 · · · wi · · · wn ]T (1)
mechanical actuators, servo systems, and other physi-
cal dynamics. However, the proposed approaches in Let w∞ , w1 , and w represent ∞-norm,
the previous works are applied to the chaotic sys- 1-norm, and the Euclidean norm (2-norm) respec-
tems without the nonlinearities effects with unknown tively. Also, sgn(w) indicates a vector sign function
time delay simultaneously while it is appeared in some of w that is defined as follows:
chaotic systems. Motivated by the removing these ef-
 T
fects, for the nonchaotic systems, one can find sev- sgn(w) = sgn(w1 ) · · · sgn(wi ) · · · sgn(wn ) (2)
eral proposed techniques in literature (see [13–18]).
Beside, some earlier works concerning the chaos sys- where
tem control with only input nonlinearities can be men- ⎧
tioned in [19–24]. Adaptive control methods inte- ⎨ 1: wi > 0
grated with approximation procedure of nonlinearities sgn(wi ) = 0: wi = 0 for i = 1, 2, . . . , n. (3)
⎩
input function are effective to deal with the input non- −1: wi < 0
linearity effects in the chaotic systems is. For exam-
ple, [19] and [20] have proposed approximation meth- Lemma ([25, 29]) For any vector x and y in R n , we
ods to tackle unknown dead-zone nonlinearity using have
fuzzy-neural network and fuzzy logic systems. Other
x ≤ x1
approaches are also investigated in some authors such  T 
as the backstepping control method [21], fuzzy con- x y  ≤ x∞ y1 (4)
trol systems [22], sliding mode control [23], and the 1
adaptive variable-structure control method [24]. xTy ≤ x2 + λy2
4λ
In this paper, to perform a perfect tracking pro-
cedure for a class of unknown time-delay chaotic where λ is a positive real number.
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1339

3 Problem statement Assumption 4 (A4) There exist a matrix M and posi-

tive definite matrices P and Q satisfying the following
Consider a class of unknown time-delay chaotic sys- equalities
tems with disturbance that its dynamic equation is
P (A − MC) + (A − MC)T P = −Q
  (10)
ẋ(t) = Ax(t) + Bη x(t), x t − τ (t) B TP = C

+ Bu(t) + Bν(t) (5) Assumption 5 (A5) There exist a matrix K and posi-
y(t) = Cx(t) tive definite matrices R and S satisfying

S(A + BK) + (A + BK)T S = −R (11)

where x(t) ∈ R n is the state variable, u(t) ∈ R p and
y(t) ∈ R p are the input and the output of the sys- Moreover, in the proposed scheme, a linear reference
tem, respectively. η(·) is an unknown nonlinear vector- model represented in (12) have been employed which
valued function, τ (t) denotes the unknown varying gives the desired output trajectory.
time-delay, and v(t) represents the unknown distur-
bances. A, B, and C are known constant matrices with ẋm (t) = Am xm (t) + Bm r(t)
suitable dimensions. (12)
ym (t) = Cxm (t)

Remark 1 It can be easy shown that there are several where xm (t) ∈ R n is the reference state variable and
chaotic systems which can be categorized in the form r(t) ∈ R l and ym (t) ∈ R p are the reference input
of (5), such as the Chua circuit [9] and the Mackey– signal and the desired output trajectory, respectively.
Glass model [27]. Am and Bm are also constant matrices with proper di-
mensions. It is well known that Am is a Hurwitz ma-
Assume that system (5) satisfies some reasonable trix.
conditions and are stated as follows.
Assumption 6 (A6) There exist two matrices F ∈
R p×n and L ∈ R p×l such that
Assumption 1 (A1) There exist two unknown posi-
tive values, τ1 and τ2 such that Am = A + BF (13)

τ (t) ≤ τ1 (6) Bm = BL (14)

τ̇ (t) ≤ τ2 (7) Remark 2 All the above assumptions are reasonable

such that A1 presents the limitation of the time-delay
Assumption 2 (A2) Nonlinear vector-valued function function and its derivative that are natural conditions.
η(·) satisfies the following condition: However, most of the chaotic systems have fixed time

 




delay and they are satisfied in this assumption [7–9].

η x(t), x t − τ (t)
≤ α
x(t)
+ β
x t − τ (t)
A2 and A3 are commonly considered to guarantee ex-
istence of the solution of the system (5). A4 represents
(8)
a common strictly positive real (SPR) condition and
A5 shows the controllability of the system. Note that
where α and β are unknown positive real numbers.
two equalities represented in A6 are common match-
ing conditions.
Assumption 3 (A3) Consider the disturbance, v(t), is
bounded such that
 

4 Proposed adaptive control methods
supv(t) =
v(t)
∞ ≤ δ (9)
t
Consider the chaos system description (5) satisfying
where δ is an unknown positive real number. assumptions A1 to A6. In this section, for the sake of
1340 H. Adloo et al.

readability of the note, first, the main scheme of the ˙ = k2

B T P ê(t)

μ̂ (20)
1
controller is introduced when the input nonlinearities
are not considered; then by making some modifica- where k1 and k2 are the positive numbers. Now notice
tions in the controller, it is shown that the proposed the following theorem.
method is robust against the hysteresis and dead-zone
input nonlinearities effects. Theorem 1 Consider the time-delayed chaotic system
dynamic (5) satisfying assumptions A1–A6. By apply-
4.1 Design without input nonlinearity ing the adaptive controller (18) with respect to the ob-
server system (15) while the unknown parameters are
Notice that our main objective in this section is to adapted as (19) and (20), it is shown that the closed-
make the system output (5) capable to track a desired loop system dynamics are stable and the estimation er-
trajectory constructed by system (12). To perform this ror, ê(t), and the tracking error, e(t), converge to zero.
and noting to the fact that we do not have any access
to the states of the system, x(t), it is required to design Proof Noting (16) and according to (5) and (15), the
an observer system which gives the estimation of x(t). estimation error dynamics can be written as
For this purpose, introduce the following dynamic sys-  
˙ = Aê(t) + Bη x(t), x t − τ (t) + Bν(t)
ê(t)
tem as an observer model:
 θ̂ 
˙ = Ax̂(t) + Bu(t) + 1 θ̂ B y(t) − C x̂(t)
x̂(t) − B y(t) − C x̂(t)
2 2
1  μ̂ 
+ μ̂Bsgn y(t) − C x̂(t) (15) − Bsgn y(t) − C x̂(t) (21)
2 2
ŷ(t) = C x̂(t) On the other hand, from (12), (15), and (17), the track-
where x̂(t) ∈ ∈
R n , ŷ(t) Rp
are the state and output ing error dynamic is
estimation variable, respectively. θ̂ and μ̂ are the adap-
θ̂ 
tation parameters. ė(t) = Ax̂(t) + Bu(t) + B y(t) − C x̂(t)
2
Remark 3 Note that the dynamic equation of many μ̂ 
+ Bsgn y(t) − C x̂(t)
chaotic systems can be described by (15). For exam- 2
ple, Chen, Lorenz, Lu, and Rossler systems are the − Am xm − Bm r(t) (22)
important samples of them.
Replace control input (18) and then by applying
Define the estimation error and the tracking error, matching condition (13) and (14), we have
respectively, as follows:
ė(t) = (A + BK)e(t) + BF xm + BLr(t)
ê(t) = x(t) − x̂(t) (16) 1  
− B θ̂ y(t) − C x̂(t)
e(t) = x̂(t) − xm (t) (17) 2
 
+ μ̂ sgn y(t) − C x̂(t)
Now, consider that the control input, u(t), is chosen as
shown below: 1  
− BF xm − BLr(t) + B θ̂ y(t) − C x̂(t)
2
θ̂   
u(t) = Ke(t) + F xm + Lr(t) − y(t) − C x̂(t) + μ̂ sgn y(t) − C x̂(t) (23)
2
μ̂  After removing the extra terms in (23), the tracking
− sgn y(t) − C x̂(t) (18)
2 error dynamic is
where the adaptation parameters θ̂ and μ̂, are gener- ė(t) = (A + BK)e(t) (24)
ated by the following update laws:


2 Now to show the stability of the error dynamics
θ̂˙ = k1
B T P ê(t)
(19) (21) and (24) with respect to the updating laws (19)
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1341


2
and (20), let us introduce the following candidate − λ
ê t − τ (t)

Lyapunov–Krasovskii function

2 1
+
e t − τ (t)
+ θ̃˙ θ̃ + μ̃
1 ˙
 t μ̃ (28)
λ 

k1 k2
V (t) = êT P ê + eT Se +
ê(s)
2
1 − τ2 t−τ (t) According to A2 and noting to the errors in (16) and


2 1 2 1 2 (17), we have
+
e(s)
ds + θ̃ + μ̃ (25)
2k1 2k2  
2êT P Bη x(t), x t − τ (t)
where λ is a positive real constant and θ̃ = θ − θ̂, μ̃ =






≤ 2
B T P êT
α
x(t)
+ β
x t − τ (t)

μ − μ̂ are the adaptation parameter errors such that



θ and μ denote the actual values. As you know, the = 2

B T P êT
α
ê(t) + x̂(t)

asymptotic stability occurs when the time derivative of

 

+ β
ê t − τ (t) + x̂ t − τ (t)

the Lyapunov–Krasovskii functional (25) be negative






definite, so ≤ 2
B T P êT
α
ê(t)
+ β
ê t − τ (t)






λ 
+ α
x̂(t)
+ β
x̂ t − τ (t)

V̇ (t) = 2êT P ê˙ + 2eT S ė +

ê(t)
2 +
e(t)
2
1 − τ2








2

2 ≤ 2
B T P êT
α
ê(t)
+ β
ê t − τ (t)

− λ
ê t − τ (t)
+
e t − τ (t)





+ α
e(t)
+ β
e t − τ (t)

1 ˙ 1 ˙ 




+ θ̃ θ̃ + μ̃ μ̃ (26) + α
xm (t)
+ β
xm t − τ (t)
(29)
k1 k2
Now replace (21) and (24) into (26), that is, Note that the desired state, xm , is bounded, therefore,
 it can be assumed that
 




V̇ (t) = 2ê P Aê(t) + Bη x(t), x t − τ (t)

T
α
xm (t)
+ β
xm t − τ (t)
< σ

θ̂  where σ is a constant positive number. So

+ Bν(t) − B y(t) − C x̂(t)
2  
2êT P Bη x(t), x t − τ (t)
μ̂ 





− Bsgn y(t) − C x̂(t)

2 ≤ 2
B T P êT
α
ê(t)
+ β
ê t − τ (t)






+ 2eT S(A + BK)e(t) + α
e(t)
+ β
e t − τ (t)
+ σ (30)
λ 




+
ê(t)
2 +
e(t)
2 Using the lemma, we can write
1 − τ2  


2

2 2êT P Bη x(t), x t − τ (t)
− λ
ê t − τ (t)
+
e t − τ (t)

4 2

2 

2
≤ α + β 2
B T P êT
+ λ
ê(t)

+ θ̃˙ θ̃ + μ̃
1 1 ˙
μ̃ (27) λ
k1 k2

2

2
+
ê t − τ (t)
+
e(t)

By applying A4, we can write



2
 +
e t − τ (t)
+ 2σ
B T P ê
(31)
V̇ (t) = êT P (A − MC) + (A − MC)T P ê
  Also, from the lemma, it can be easily shown that
+ 2êT P MC ê + 2êT P Bη x(t), x t − τ (t)

T

T

B P ê
≤
B P ê
(32)
+ 2êT P Bν(t) − θ̂ êT P B y(t) − C x̂(t) 1

− μ̂êT P Bsgn y(t) − C x̂(t) Thus,
 
+ 2eT S(A + BK)e 2P B êT η x(t), x t − τ (t)
λ 



4 2

2 

2
+
ê(t)
2 +
e(t)
2 ≤ α + β 2
B T P ê
+ λ
ê(t)

1 − τ2 λ
1342 H. Adloo et al.


2

2 
+
ê t − τ (t)
+
e(t)
4 1
T
2
+ −θ̂ + α 2 + β 2 +
B P ê


2

λ ϕ
+
e t − τ (t)
+ 2σ
B T P ê
1 (33) 

+ −μ̂ + 2σ + 2
v(t)
∞
B T P ê
1
Moreover, we have the following inequality: 1 ˙ 1 ˙
+ θ̃ θ̃ + μ̃ μ̃ (39)
1 k1 k2
2ê P MC ê ≤ ϕ êP MM P ê + êP BB T P ê
T T
(34)
ϕ So, substituting updating laws (19) and (20), we have
where ϕ is a positive value. In addition, it can be veri- 
λ
fied that V̇ (t) ≤ −ê Q − ϕP MM P − λI −
T T
ê
1−τ





êT P Bν(t) ≤
v(t)
∞
B T P ê
1 (35) λ
− eT R − λI − e
1−τ
Also, 
4 1
T
2
 + −θ̂ + α 2 + β 2 +
B P ê

μ̂êT P Bsgn y(t) − C x̂(t) λ ϕ





= μ̂êT P Bsgn B T P ê(t) = μ̂

B T P ê
1 (36) + −μ̂ + 2σ + 2
v(t)
∞
B T P ê
1


2

and − θ̃
B T P ê
− μ̃
B T P ê
1 (40)

θ̂ êT P B y(t) − C x̂(t) Thus,


2 
= θ̂ êT P BB T P ê = θ̂
B T P ê
(37) V̇ (t) ≤ −ê Q − ϕP MM T P − λI −
T λ
ê
1 − τ2
Now, apply expressions (33)–(37) to (28), therefore, 
λ
we have − eT R − λI − e
1 − τ2
1 
V̇ (t) ≤ −êT Qê + ϕ êP MM T P ê + êP BB T P ê 4 1

B T P ê
2
ϕ + −θ + α 2 + β 2 +
λ ϕ
4

2 

2 


T

+ α 2 + β 2
B T P êT
+ λ
ê(t)
+ −μ + 2σ + 2
v(t)

B P ê
(41)
λ ∞ 1


2

2
+ ê t − τ (t)
+
e(t)

Since the adaptations laws, (19) and (20), are increas-


2

ing, hence choosing sufficiently large positive values
+
e t − τ (t)
+ 2σ
B T P ê
1 for initial conditions θ̂0 and μ̂0 and also selecting






2 smallest values for ϕ and λ lead to V̇ (t) ≤ 0. There-
+ 2
v(t)
∞
B T P ê
1 − θ̂
B T P ê
fore, the closed-loop system is uniformly bounded and

− μ̂
B T P ê
1 + 2eT S(A + BK)e estimation error, ê, and tracking error, e, will be con-
verged to zero. 
λ 




+
ê(t)
2 +
e(t)
2
1 − τ2 4.2 Robustness to the input nonlinearities


2

2
− λ
ê t − τ (t)
+
e t − τ (t)
Due to some practical limitations in the control sys-
tems, it is commonly observed that the control input
+ θ̃˙ θ̃ + μ̃
1 1 ˙
μ̃ (38)
k1 k2 is conditioned by the unknown nonlinear and/or nons-
mooth functions. Among them, hysteresis and dead-
Then by doing some simplifications, we can write zone nonlinearities very commonly occurred, which
 limit the system performance and sometimes even
λ
V̇ (t) ≤ −ê Q − ϕP MM P − λI −
T T
ê causes the instability.
1 − τ2
 In this section, we will introduce the dead-zone
λ nonlinearity and a model of hysteresis function, re-
− eT R − λI − e
1 − τ2 spectively.
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1343

Theorem 2 For the chaotic system dynamic (43) sat-

isfying A1 to A6, if the observer system (15) and the
adaptive controller (44) with the updating laws (19)
and (20) are used then the closed-loop system is sta-
ble and the tracking error, e, and the estimation error,
ê, converge to the zero in spite of the unknown input
dead-zone nonlinearity.

Proof The proof is given in Appendix A. 

4.2.2 Hysteresis nonlinearity

Fig. 1 Dead-zone nonlinearity The hysteresis nonlinearities are another undesirable

phenomenon for which various models have been pre-
sented. Because the current output variable of the hys-
4.2.1 Dead-zone nonlinearity
teresis models depends on the rate of the input vari-
able, its input-output relation is inherently dynamical.
Generally, the static model of the dead-zone nonlinear- A well-known model of the hysteresis phenomenon is
ity is described by the following input-output relation: the Backlash-like hysteresis model. The general de-
⎧ scription of the backlash-like hysteresis can be repre-
⎨ mi zi − ni1 for zi > ni1
sented by the following equations:
D i (zi ) = 0 for ni2 ≤ zi ≤ ni1 (42)
⎩ 
mi zi − ni2 for zi < ni2
wi (t) = H i zi (t), żi (t)
⎧
where ni1 , ni2 are the widths of the dead zone and mi ⎪
⎪ ci (zi (t) − B), if żi (t) > 0 and
is the slope of function. The dead-zone function (42) ⎪
⎪ w(t) = ci (zi (t) − Bi )
⎨
can be depicted as shown in Fig. 1. Note that, if z is = ci (zi (t) − B), if żi (t) < 0 and (45)
given as a vector such as z = [zi ]p×1 , then we have ⎪
⎪
⎪
⎪ w(t) = ci (zi (t) + Bi )
⎩
D(z) = [D i (zi )]p×1 . wi (t− ) otherwise
Now, let us consider that the input control system in
(5) is exposed to the unknown dead-zone nonlinearity, where zi and wi are the input and output variable, re-
such that spectively, for i = 1, 2, . . . , p and ci and Bi are the
positive constants which represent the slope of the
 
ẋ(t) = Ax(t) + Bη x(t), x t − d(t) lines and the backlash distance, respectively. A suit-
 able approximation of the backlash-like model (45)
+ BD z(t) + Bν(t) (43)
has been proposed by Su et al. [26] which is an dy-
y(t) = Cx(t) namic equation description as shown below
 
where z(t) ∈ R p is the control input variable. The dwi  dzi 

= αi  (ci zi − wi ) + B1i dzi (46)
main task of overall closed-loop system is again to dt dt  dt
track the desired output trajectory constructed by
model reference (12). According to this, let the con- where αi , ci, and B1i are unknown positive constants
trol input, z(t), is given as with ci > B1i . A typical illustration of the backlash-
like hysteresis nonlinearity is shown in Fig. 2, where
z(t) = M̂ −1 u(t) (44) zi (t) = 2.5 sin(2.3t), αi = 1, ci = 1, B1i = 0.9, and
wi (0) = 0.
where M̂ = diag(m̂1 , m̂2 , . . . , m̂p ) is a positive defi- Here, consider, the control input of the system (5)
nite diagonal matrix which is designed later. And u(t) is conditioned by the backlash-like hysteresis model
was also introduced in (18). (46) such that
1344 H. Adloo et al.

that paper, only dead-zone nonlinearity is considered

without any time-delay consideration.

Remark 5 Note that, in order to remove the dis-

continuous’ effect due to the sign function, sgn(w),
in the proposed methods, experimentally, instead of
sign function, one can use the saturation function,
sat(w/ε), which is a good approximation of sign func-
tion, if ε is sufficiently assigned small.

Fig. 2 Backlash like hysteresis nonlinearity

5 Simulation results

  To show the capability and the applicability of the pro-

ẋ(t) = Ax(t) + Bη x(t), x t − d(t)
 posed methods concerning with the forcing the un-
+ BH z(t), ż(t) + Bν(t) (47) known time-delayed chaotic system to track the de-
y(t) = Cx(t) sired trajectory while removing the input nonlinearity
effects and disturbances, three numerical examples are
where z(t) ∈ R p is the control input and H (z(t), ż(t)) presented. In the first example, a MIMO 3D chaotic
= [H i (zi (t), żi (t))]p×1 . To perform the perfect track- system has been presented and two others are SISO
ing procedure, the following controller is assigned: first-order delayed-chaotic systems.

z(t) = Ĉ −1 u(t) (48)

5.1 Chua circuit system
where Ĉ = diag(ĉ1 , ĉ2 , . . . , ĉp ) is a positive definite
diagonal matrix and u(t) is (18). As the first example, a 3D Chua circuit system is con-
sidered with the following dynamic equation [9]:
Theorem 3 Consider the time-delayed chaotic system 
(47) with assumptions A1–A6. By applying the adap- ẋ1 (t) = σ1 −x1 + x2 − f (x1 ) − cx1 (t − τ )
tive controller (48) and utilizing the observer system 
+ (1 + sin 3t) x1 (t − τ )
(15) and updating laws (19) and (20), it can be shown
that the closed-loop system is stable and, therefore, the + N (z1 ) + d(t)
estimation error and the tracking error will converge (49)
ẋ2 (t) = x1 − x2 + x3 + cx2 (t − τ ) + N (z2 )
to zero in spite of the unknown input hysteresis nonlin- 
earity. ẋ3 (t) = −σ2 x2 + c 2x1 (t − τ )

+ x3 (t − τ ) + N (z3 )
Proof The proof is given in the Appendix B. 
where σ1 = 10, σ2 = 15, f (x1 ) = bx1 + 0.5(a −
Remark 4 As it is shown, the proposed approaches
b)(|x1 + 1| − |x1 − 1|), a = −1.28, b = −0.69, and
solved the problem of tracking procedure of the un-
c = 0.1. Also, d(t) denotes the disturbance that is
known time-delay chaotic systems which are exposed
to the nonlinearity effects and disturbances. However,
d(t) = (0.8 sin 2t)x1 + (1 + 0.3 sin 2t)x2
there are a few works around the chaos control prob-
lem for the chaotic systems restricted by the nonlin- + (−1 + 0.5 cos t)x3 (50)
earity effects. Noroozi et al. [23] proposed a variable-
structure control approach for a large class of chaotic Also, N (zi ) denotes the input nonlinearity. In this sim-
system while the time-delay is not considered. Also, ulation, the unknown time delay τ = 1 and the state
Roopaei et al. [24] proposed a new adaptive chaos con- values of the Chua circuit system for t ∈ [−1, 0] is
trol approach using sliding mode control method. In X = [x1 x2 x3 ]T = [−1 −1 −1]T while the ini-
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1345

Fig. 3 States of the Chua circuit (solid line), observer system (dash line) and reference model system (dotted line) in case 1

tial condition of the observer system is chosen as 2, m2 = 1, and m2 = 0.5. In the numerical simulation,
[−0.5 − 0.5 − 0.5]T . Moreover, we used following matrix in the controller (44):
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
−8.1 10 0 1 0 0 1.5 0 0
F =⎣ 1 1 1⎦, L = ⎣0 1 0⎦ M̂ = ⎣ 0 0.8 0 ⎦
0 −15 3 0 0 1 0 0 0.35

and the desired trajectories are selected as The simulation results have been shown in Figs. 3–5.
⎧ Figure 3 depicts the state variables of the Chua cir-
⎪ πt −0.5t
⎪
⎪ r1 (t) = e cuit system, observer, and reference model systems.
⎪
⎪ 2 
⎪
⎨ In Figs. 4, and 5, the control input and the adaptation
1 πt
r2 (t) = cos parameters through the time are demonstrated, respec-
⎪
⎪ 2 4
⎪
⎪ tively.
⎪
⎪ 1
⎩ r3 (t) = cos(2πt)
2
Case 2: (Backlash-like hysteresis) In this case, the
Case 1: (Dead-zone) In this case, the adaptation backlash-like hysteresis model (46) has been consid-
gains are k1 = 5 and k2 = 20. Also, the dead-zone ered with αi = 1, and Bi1 = 0.5 for i = 1, 2, and 3.
parameters for all inputs, z1 (t), z2 (t), and z3 (t), are Moreover, c1 = 0.1, c2 = 0.2, and c3 = 0.5. In the sim-
ni1 = 1, ni2 = −1 for i = 1, 2, and 3. Moreover, m1 = ulation, we assigned k1 = 5 and k2 = 15. The initial
1346 H. Adloo et al.

Fig. 4 Inputs of the Chua circuit system, in case 1

The simulation results of this numerical example have

been shown in Figs. 6–8.

5.2 Mackey–Glass model

The Mackey–Glass model [27] is a first-order time-

delayed chaotic system which represents some physi-
ological systems behavior. The general dynamic equa-
tion of the Mackey–Glass model with input nonlinear-
ity is described as

x(t − τ )
ẋ(t) = −x(t) + 2
Fig. 5 Estimation parameters, θ (solid line), μ (dash line), in 1 + |x(t − τ )|w
case 1 
+ N z(t) + Bv(t), w > 1 (51)

conditions are selected to be same as the previous ex- Here, we consider w = 10 and τ = 0.8. Also, the dis-
ample. In addition, turbance is v(t) = 0.5 cos(t) and F = 1 and L = 1.
⎡ ⎤ The reference input is also r(t) = sin(πt/2). The state
0.05 0 0 response for the Mackey–Glass model is considered as
Ĉ = ⎣ 0 0.1 0 ⎦ x(t) = 1 − 0.2 sin(t) for t ∈ [−0.8, 0]. In addition, the
0 0 0.3 initial condition of the observer system is x̂0 = −0.5.
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1347

Fig. 6 States of the Chua circuit system (solid line), observer system (dash line) and reference model system (dotted line) in case 2

Case 3: (Dead-zone nonlinearity) Dead-zone non- Case 5: (Dead-zone) The dead-zone nonlinearity’s
linearity’s parameters are n1 = n2 = 1 and m = 1.5. parameters are n1 = n2 = 1 and m = 2. Moreover,
The simulation results of this example for k1 = 1, k2 = k1 = 2.5 and k2 = 10. The simulation results for m̂ =
1.5, and m̂ = 0.7 are illustrated in Fig. 9. 1.2 are illustrated in Fig. 11.

Case 4: (Backlash-like hysteresis) In this example, Case 6: (Backlash-like hysteresis) The backlash-like
the parameters of the backlash model are α = 0.3, hysteresis’ parameters are α = 0.3, B1 = 1, and c = 1.
B1 = 0.5, and c = 2. The simulation results for k1 = By selecting k1 = 6 and k2 = 0.6 and ĉ = 0.05, the
1, k2 = 1.5, and ĉ = 1 are shown in Fig. 10. simulation results are obtained as shown in Fig. 12.

Ikeda model The Ikeda model is another first-order

chaotic system with time delay that its dynamic equa- 6 Conclusion
tion is [28]:
  The main results of this article was concerning the
ẋ(t) = −3x(t) + 24 sin x(t − τ ) + N z(t) + Bv(t) chaos MRAC of a nonlinear unknown time-delay sys-
tem with the input nonlinearities. Using the observer-
The time delay is τ = 0.4 and v(t) = 0.1 cos(t). It is
also assumed that the reference signal is r(t) = sin(3t) based adaptive control approach, it was shown that the
and F = −2 and L = 1. The state response is x(t) = output system tracks a desired trajectory. Finally, the
1 − 0.1 sin(t) for t ∈ [−0.8, 0]. effectiveness of the proposed method was evaluated
via simulation results.
1348 H. Adloo et al.

Fig. 7 Inputs of the Chua circuit system, in case 2

 
˙ = Aê(t) + Bη x(t), x t − τ (t)
ê(t)
 
+ B D z(t) − u(t)
θ̂ 
+ Bν(t) − B y(t) − C x̂(t)
2
μ̂ 
− Bsgn y(t) − C x̂(t) (A.1)
2
Note that the model of the dead-zone nonlinearity (42)
can be reformulated as
 
Fig. 8 Estimation parameters, θ (solid line), μ (dash line), in D z(t) = M̄z(t) + ζ z(t) (A.2)
case 2
where M̄ = diag(m1 , m2 , . . . , mp ) and ζ (z(t)) =
[ζi (zi (t))]1×p where ζi (zi (t)) for i = 1, . . . , p are
Appendix A: Robustness to dead-zone evaluated as follows:
nonlinearity ⎧m n
⎪ i 1i for zi ≥ n1i ,
 ⎨ −mi zi for 0 ≤ zi < n1i ,
ζi zi (t) = (A.3)
Regarding system (43) and observer system (15), the ⎩ −mi zi for −n2i < zi < 0,
⎪
estimation error dynamic is mi n2i for zi ≤ −n2i
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1349

Fig. 9 a State of Mackey model (solid line), observer system (dash line) and reference model (dotted line), b Control input, c Estima-
tion parameters, θ (solid line), μ (dash line), in case 3

It is clear that ζi (zi (t)) for i = 1, . . . , p are bounded, μ̂ 

− Bsgn y(t) − C x̂(t) (A.5)
and satisfy |ζi (zi (t))| ≤ ρi where ρi are the pos- 2
itive numbers. Therefore, it can be assumed that Then, by applying the control input (44), we have
ζ (z) ≤ E. It is also assumed that mi , ni1 , and ni2
 
for i = 1, . . . , p are unknown and only it is known ˙ = Aê(t) + Bη x(t), x t − τ (t)
ê(t)
that there are two positive definite matrices, M1 and  
M2 , such that + B M̄ M̂ −1 − I u(t) + B ν(t) + ζ (t)
θ̂ 
M1 < M̄ < M2 (A.4) − B y(t) − C x̂(t)
2
By applying these, we can write μ̂ 
− Bsgn y(t) − C x̂(t) (A.6)
2
 
˙ = Aê(t) + Bη x(t), x t − τ (t)
ê(t)
 Now, define M̃ = M̄ − M̂, so we have
+ B M̄z(t) − u(t)  
˙ = Aê(t) + Bη x(t), x t − τ (t)
ê(t)
 θ̂  
+ B ν(t) + ζ (t) − B y(t) − C x̂(t) + B M̃ M̂ −1 u(t) + B ν(t) + ζ (t)
2
1350 H. Adloo et al.

Fig. 10 a State of Mackey model (solid line), observer system (dash line) and reference model (dotted line), b Control input, c Esti-
mation parameters, θ (solid line), μ (dash line), in case 4


θ̂  + M̃ M̂ −1 Ke(t)
− B y(t) − C x̂(t)
2  
+ 2êT (t)P B M̃ M̂ −1 F xm (t) + Lr(t)
μ̂  
− Bsgn y(t) − C x̂(t) (A.7) + ν(t) + ζ (t)
2

On the other hand, noting (15), (18), and (12), the − θ̂ êT (t)P B M̃ M̂ −1 y(t) − C x̂(t)
tracking error dynamic is 
− μ̂êT (t)P B M̃ M̂ −1 sgn y(t) − C x̂(t)
ė(t) = (A + BK)e(t) (A.8) + 2eT (t)S(A + BK)e(t)
λ 




Here, assume that the Lyapunov function (25) is also +
ê(t)
2 +
e(t)
2
considered. Thus, according to (18), (A.3), and (A.4), 1 − τ2


2

2
and with some simplification, the derivation of the − λ
ê t − τ (t)
+
e t − τ (t)

Lyapunov function is
1 ˙ 1 ˙
+ θ̃ θ̃ + μ̃ μ̃ (A.9)
V̇ (t) = 2êT (t)P Aê(t) k1 k2
  
+ 2êT (t)P B η x(t), x t − τ (t) Noting A4, we have
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1351

Fig. 11 a State of Ikeda model (solid line), observer system (dash line) and reference model (dotted line), b Control input, c Estimation
parameters, θ (solid line), μ (dash line), in case 5

 

2

2
V̇ (t) = êT (t) P (A − MC) + (A − MC)T P ê(t) − λ
ê t − τ (t)
+
e t − τ (t)

+ 2êT (t)P MC ê(t) 1 ˙ 1 ˙

+ θ̃ θ̃ + μ̃ μ̃ (A.10)
   k1 k2
+ 2êT (t)P B η x(t), x t − τ (t)

+ M̃ M̂ −1 Ke(t) Now, regarding (30), we can write the following ex-
 pressions:
+ 2êT (t)P B M̃ M̂ −1 F xm (t) + Lr(t)

+ 2êT (t)P B ν(t) + ζ (t)    
 2êT (t)P B η x(t), x t − τ (t) + M̃ M̂ −1 Ke(t)
− θ̂ êT (t)P B M̃ M̂ −1 y(t) − C x̂(t)






 ≤ 2
B T P êT
α
ê(t)
+ β
ê t − τ (t)

− μ̂êT (t)P B M̃ M̂ −1 sgn y(t) − C x̂(t) 



+ (α + ε1 )
e(t)
+ β
e t − τ (t)
+ σ
+ 2eT (t)S(A + BK)e(t)






≤ 2
B T P êT
ᾱ
ê(t)
+ β
ê t − τ (t)

λ 




+
ê(t)
2 +
e(t)
2 




1 − τ2 + ᾱ
e(t)
+ β
e t − τ (t)
+ σ
1352 H. Adloo et al.

Fig. 12 a State of Ikeda model (solid line), observer system (dash line) and reference model (dotted line), b Control input, c Estimation
parameters, θ (solid line), μ (dash line), in case 6




where M̃ M̂ −1 K ≤ ε1 and ᾱ = α + ε1 . So, from the ≤

M̃ M̂ −1 F xm (t) + Lr(t)
∞
lemma, it is shown that

+
ν(t)
∞ +
ζ (t)
∞
B T P ê
1
   

2êT (t)P B η x(t), x t − τ (t) + M̃ M̂ −1 Ke(t) ≤ (κ + δ + E)

B T P ê
(A.12)
1
4

2 

2
≤ ᾱ 2 + β 2
B T P ê
+ λ
ê(t)

λ where it is assumed M̃ M̂ −1 (F xm (t) + Lr(t))∞ ≤ κ.


2

2 In addition,
+
ê t − τ (t)
+
e(t)


2


+
e t − τ (t)
+ 2σ
B T P ê
1 (A.11) θ̂ êT P B y(t) − C x̂(t)


2
= θ̂ êT P BB T P ê = θ̂
B T P ê
(A.13)
Also, from A3, it can be shown that
   and
êT (t)P B M̃ M̂ −1 F xm (t) + Lr(t) + ν(t) + ζ (t)

 
≤
M̃ M̂ −1 F xm (t) + Lr(t) + ν(t) μ̂êT P Bsgn y(t) − C x̂(t)





+ ζ (t)
∞
B T P ê
1 = μ̂êT P Bsgn B T P ê(t) = μ̂
B T P ê
1 (A.14)
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1353

Moreover, and with respect to the fact that adaptations laws, (19)
and (20), are increasing, therefore, choosing positive
1
2êT P MC ê ≤ ϕ êP MM T P ê + êP BB T P ê (A.15) values for initial conditions θ̂0 and μ̂0 leads to θ̂ (t) > 0
ϕ and μ̂(t) > 0. So, it is not difficult to show that
Now, by applying expressions and inequalities (A.10)– 
−θ̂ êT (t)P B M̃ M̂ −1 y(t) − C x̂(t)
(A.14), we have

1 −μ̂êT (t)P B M̃ M̂ −1 sgn y(t) − C x̂(t) < 0 (A.19)
V̇ (t) ≤ −êT Qê + ϕ êP MM T P ê + êP BB T P ê
ϕ Finally, by choosing sufficiently large positive values
4

2 

2 for θ̂0 and μ̂0 , and smallest values for ϕ and λ, we
+ ᾱ 2 + β 2
B T P ê
+ λ
ê(t)

λ have V̇ (t) ≤ 0. Therefore, the closed-loop system is


2

2 uniformly bounded and estimation error, ê, and track-
+
ê t − τ (t)
+
e(t)
ing error, e, will be converged to zero.


2

+
e t − τ (t)
+ 2σ
B T P ê
1

+ 2(κ + δ + E)
B T P ê
1 Appendix B: Robustness to hysteresis nonlinearity

− θ̂ êT (t)P B M̃ M̂ −1 y(t) − C x̂(t) In this case, noting (47) and (15), we can write the
 estimation error dynamics as shown below:
− μ̂êT (t)P B M̃ M̂ −1 sgn y(t) − C x̂(t)
   


2

ê˙ = Aê + Bη x(t), x t − τ (t) + B H z(t) − u(t)
− θ̂
B T P ê
− μ̂
B T P ê
1
θ̂ 
+ 2e (t)S(A + BK)e(t)
T + Bν − B y(t) − C x̂(t)
2
λ 




+
ê(t)
2 +
e(t)
2 μ̂ 
1 − τ2 − Bsgn y(t) − C x̂(t) (B.1)
2


2
− λ
ê t − τ (t)
It is worthy to note that the analytic solution of the


2 1 dynamic equation (49) is
+
e t − τ (t)
+ θ̃˙ θ̃ + μ̃
1 ˙
μ̃ (A.16)
k1 k2 H (z, ż) = C̄z + ξ(z, ż) (B.2)
Then by substituting adaptations laws (19) and (20), it
where H (z(t), ż(t)) = [H i (zi (t), żi (t))]p×1 such that
can be easily shown that
H i (zi , żi ) = ci zi (t) + ξi (zi , żi ) and C̄ =

λ diag (c1 , c2 , . . . , cp ) with
V̇ (t) ≤ −êT (t) Q − ϕP MM T P − λI −
1 − τ2
 ξi (zi , żi )
λ
× ê(t) − eT (t) R − λI − e(t) = (w0i − ci z0i )e−αi (zi −z0i )sgn(żi )
1 − τ2  zi
 −αi zi sgn(żi )
4 2 1
T
2 +e (B1i − ci )eαi λsgn(żi ) dλ (B.3)
+ −θ + ᾱ + β + 2
B P ê
z0i
λ ϕ


where w0i and z0i denote wi (0) and zi (0). It can be
+ −μ + 2σ + 2(κ + δ + E)
B T P ê
1
easily verified that the expression (B.2) is bounded

− θ̂ êT (t)P B M̃ M̂ −1 y(t) − C x̂(t) such that [26]
  
− μ̂êT (t)P B M̃ M̂ −1 sgn y(t) − C x̂(t) ξi (zi , żi ) ≤ ci − B1i (B.4)
αi
(A.17)
Thus, let us suppose that ξ(z, ż) ≤ Ω where Ω is a
In this expression, clearly if M̂ is chosen, such that positive constant. Moreover, assume that there are two
0 < M̂ ≤ M1 , then we have positive definite matrices, C1 and C2 such that

M̃ > 0 (A.18) C1 < C̄ < C2 (B.5)

1354 H. Adloo et al.







Noting to these, the error estimation dynamic (B.1) is ≤ 2
B T P êT
α
ê(t)
+ β
ê t − τ (t)

   




ê˙ = Aê + Bη x(t), x t − τ (t) + B C̃ Ĉ −1 − I u + (α + ε2 )
e(t)
+ β
e t − τ (t)
+ σ







θ̂  ≤ 2
B T P êT
ᾱ
ê(t)
+ β
ê t − τ (t)

+ Bν − B y(t) − C x̂(t) 




2 + ᾱ
e(t)
+ β
e t − τ (t)
+ σ (B.10)
μ̂ 
− Bsgn y(t) − C x̂(t) (B.6) Thus,
2
   
Define C̃ = C̄ − Ĉ, then êT (t)P B η x(t), x t − τ (t) + C̃ Ĉ −1 Ke(t)
  2 2

2
˙ = Aê(t) + Bη x(t), x t − τ (t) + B C̃ Ĉ −1 u(t)
ê(t) ≤ ᾱ + β 2
B T P ê

λ
 θ̂  

2

2
+ B ν(t) + ξ(t) − B y(t) − C x̂(t) + λ
ê(t)
+
ê t − τ (t)

2


2

2
μ̂  +
e(t)
+
e t − τ (t)

− Bsgn y(t) − C x̂(t) (B.7)

2
+ σ
B T P ê
1 (B.11)
Also, from (15), (12), and (18) it is verified that
where ᾱ = α + ε2 and C̃ Ĉ −1 K ≤ ε2 . Noting A2 it
ė(t) = (A + BK)e(t) (B.8)
is shown that
Let us consider again Lyapunov function (25). Now,   
according to A4, (B.7), and (B.8), we can write êT (t)P B C̃ Ĉ −1 F xm (t) + Lr(t) + ν(t) + ζ (t)


V̇ (t) = êT (t) P (A − MC) + (A − MC)T P ê(t) ≤ (ε + δ + Ω)

B T P ê
1 (B.12)

+ 2êT (t)P MC ê(t) where C̃ Ĉ −1 (F xm (t) + Lr(t))∞ ≤ ε. Now, by ap-

   plying expressions and inequalities (B.11) and (B.12),
+ 2êT (t)P B η x(t), x t − τ (t)
 we have
+ M̃ M̂ −1 Ke(t)
1
 V̇ (t) ≤ −êT Qê + ϕ êP MM T P ê + êP BB T P ê
+ 2êT (t)P B C̃ Ĉ −1 F xm (t) + Lr(t) ϕ
4

2 

2
+ ν(t) + ζ (t) + ᾱ 2 + β 2
B T P ê
+ λ
ê(t)

 λ
− θ̂ êT (t)P B C̃ Ĉ −1 y(t) − C x̂(t)

2

2
 +
ê t − τ (t)
+
e(t)

− μ̂êT (t)P B C̃ Ĉ −1 sgn y(t) − C x̂(t)


2

 +
e t − τ (t)
+ 2σ
B T P ê
1
− θ̂ êT (t)P B y(t) − C x̂(t)

 + 2(ε + δ + Ω)
B T P ê
1
− μ̂êT (t)P Bsgn y(t) − C x̂(t) 
− θ̂ êT (t)P B C̃ Ĉ −1 y(t) − C x̂(t)
+ 2eT (t)S(A + BK)e(t) 
λ 



− μ̂êT (t)P B C̃ Ĉ −1 sgn y(t) − C x̂(t)
+
ê(t)
2 +
e(t)
2

2

1 − τ2 − θ̂
B T P ê
− μ̂
B T P ê



2

2 1
− λ
ê t − τ (t)
+
e t − τ (t)
+ 2e (t)S(A + BK)e(t)
T

1 ˙ 1 ˙ λ 




+ θ̃ θ̃ + μ̃ μ̃ (B.9) +
ê(t)
2 +
e(t)
2
k1 k2 1 − τ2


2

2
Similar to that of the previous proof, it can be shown − λ
ê t − τ (t)
+
e t − τ (t)

that
    1 ˙ 1 ˙
+ θ̃ θ̃ + μ̃ μ̃ (B.13)
êT (t)P B η x(t), x t − τ (t) + C̃ Ĉ −1 Ke(t) k1 k2
Observer-based model reference adaptive control for unknown time-delay chaotic systems 1355

Then by substituting adaptations laws (19) and (20), 6. Harb, A., Harb, B.: Chaos control of a third-order phase
we have locked loop using backstepping nonlinear controller. Chaos
Solitons Fractals 20, 719–723 (2004)

λ 7. Tian, Y.C., Gao, F.: Adaptive control of chaotic continuous-
V̇ (t) ≤ −ê (t) Q − ϕP MM P − λI −
T T
time systems with delay. Physica D 117, 1–12 (1998)
1 − τ2
8. Hua, C., Guan, X.: Robust control of time-delay chaotic

λ systems. Phys. Lett. A 314, 72–80 (2003)
× ê(t) − eT (t) R − λI − e(t) 9. Ting, C.S.: An observer-based approach to controlling
1 − τ2 time-delay chaotic systems via Takagi–Sugeno fuzzy

4 1
T
2 model. Inf. Sci. 177, 4314–4328 (2007)
+ −θ + ᾱ 2 + β 2 +
B P ê
10. Bowong, S., Tewa, J.J.: Unknown inputs’ adaptive observer
λ ϕ for a class of chaotic systems with uncertainties. Math.


+ −μ + 2σ + 2(ε + δ + Ω)
B T P ê
1 Comput. Model. 48, 1826–1839 (2008)
11. Solak, E., Morgül, Ö., Ersoy, U.: Observer-based control of

− θ̂ êT (t)P B C̃ Ĉ −1 y(t) − C x̂(t) a class of chaotic systems. Phys. Lett. A 279, 47–55 (2001)
12. Hua, C., Guan, X., Li, X., Shi, P.: Adaptive observer-based

− μ̂êT (t)P B C̃ Ĉ −1 sgn y(t) − C x̂(t) (B.14) control for a class of chaotic systems. Chaos Solitons Frac-
tals 22, 103–110 (2004)
13. Boulkroune, A., Saad, M.M., Farza, M.: Adaptive fuzzy
Consider that Ĉ is chosen to be satisfied 0 < Ĉ ≤ C1 , controller for multivariable nonlinear state time-varying de-
then we have lay systems subject to input nonlinearities. Fuzzy Sets Syst.
164(1), 45–65 (2011)
C̃ > 0 (B.15) 14. Hu, Q., Ma, G., Xie, L.: Robust and adaptive variable-
structure output feedback control of uncertain systems with
input nonlinearity. Automatica 44, 552–559 (2008)
and with respect to the fact that adaptations laws, (19) 15. Tarbouriech, S., Prieur, C., Queinnec, I.: Stability analy-
and (20), are increasing, therefore, choosing positive sis for linear systems with input backlash through sufficient
values for initial conditions θ̂0 and μ̂0 leads to θ̂ (t) > 0 LMI conditions. Automatica 46(11), 1911–1915 (2010)
and μ̂(t) > 0, so, it is not difficult to show that 16. Barreiro, A., Baños, A.: Input–output stability of systems
with backlash. Automatica 42, 1017–1024 (2006)
 17. Liu, Y.J., Wang, W., Tong, S.C., Liu, Y.S.: Robust adap-
−θ̂ êT (t)P B C̃ Ĉ −1 y(t) − C x̂(t) tive tracking control for nonlinear systems based on bounds
 of fuzzy approximation parameters. IEEE Trans. Syst. Man
−μ̂êT (t)P B C̃ Ĉ −1 sgn y(t) − C x̂(t) < 0 (B.16) Cybern. 40, 170–184 (2010)
18. Liu, Y.J., Wang, W.: Adaptive fuzzy control for a class of
Finally, by choosing sufficiently large positive values uncertain nonaffine nonlinear systems. Inf. Sci. 177, 3901–
for θ̂0 and μ̂0 , and smallest values for ϕ and λ, we 3917 (2007)
have V̇ (t) ≤ 0. Therefore, the closed-loop system is 19. Liu, Y.J., Zhou, N.: Observer-Based adaptive fuzzy-neural
control for a class of uncertain nonlinear systems with un-
uniformly bounded and estimation error, ê, and track- known dead-zone input. ISA Trans. 49, 462–469 (2010)
ing error, e, will be converged to zero. 20. Liu, Y.J., Zheng, Y.Q.: Adaptive robust fuzzy control for
a class of uncertain chaotic systems. Nonlinear Dyn. 57,
431–439 (2009)
21. Zhou, N., Liu, Y.J., Tong, S.C.: Adaptive fuzzy output feed-
References back control of uncertain nonlinear systems with nonsym-
metric dead-zone input. Nonlinear Dyn. 63(4), 771–778
1. Ott, E., Grebogi, C., Yorke, J.A.: Controlling chaos. Phys. (2010)
Rev. Lett. 164, 1196–1199 (1990) 22. Liu, Y.J., Tong, S.C., Wang, W., Li, Y.M.: Observer-based
2. Lee, K.W., Singh, S.N.: Robust control of chaos in Chua’s direct adaptive fuzzy control of uncertain nonlinear systems
circuit based on internal model principle. Chaos Solitons and its applications. Int. J. Control. Autom. Syst. 7, 681–
Fractals 31, 1095–1107 (2007) 690 (2009)
3. Wang, Y., Chai, T., Zhang, Y.: State observer-based adap- 23. Noroozi, N., Roopaei, M., Karimaghaee, P., Safavi, A.A.:
tive fuzzy output-feedback control for a class of uncertain Simple adaptive variable-structure control for unknown
nonlinear systems. Inf. Sci. 180, 5029–5040 (2010) chaotic systems. Commun. Nonlinear Sci. Numer. Simul.
4. Nazzal, J.M., Natsheh, A.N.: Chaos control using sliding- 15, 707–727 (2010)
mode theory. Chaos Solitons Fractals 33, 695–702 (2007) 24. Roopaei, M., Sahraei, B.R., Lin, T.C.: Adaptive sliding
5. Wang, H., Han, Z.Z., Zhang, W., Xie, Q.Y.: Synchroniza- mode control in a novel class of chaotic systems. Commun.
tion of unified chaotic systems with uncertain parameters Nonlinear Sci. Numer. Simul. 15, 4158–4170 (2010)
based on the CLF. Nonlinear Anal., Real World Appl. 10, 25. Khalil, H.K.: Nonlinear Systems. Prentice Hall, Englewood
715–722 (2009) Cliffs (2003)
1356 H. Adloo et al.

26. Su, C.Y., Stepanenko, Y., Svoboda, J., Leung, T.P.: Robust 28. Lu, H., He, Z.: Chaotic behaviors in first order autonomous
adaptive control of a class of nonlinear systems with un- continuous time systems with delay. IEEE Trans. CAS I 43,
known backlash-like hysteresis. IEEE Trans. Autom. Con- 700–702 (1996)
trol 45, 2427–2432 (2000) 29. Jiang, Z.P., Praly, L.: Design of robust adaptive controllers
for nonlinear systems with dynamic uncertainties. Auto-
27. Mackey, M.C., Glass, L.: Oscillation and chaos in physio- matica 32, 1211–1215 (1998)
logical control systems. Science 197, 287–289 (1977)