Вы находитесь на странице: 1из 23

Received: 5 November 2018 Revised: 22 March 2019 Accepted: 12 May 2019

DOI: 10.1002/acs.3014

RESEARCH ARTICLE

Prescribed performance adaptive fuzzy dynamic surface


control of nonaffine time-varying delayed systems with
unknown control directions and dead-zone input

Pouria Tooranjipour1 Ramin Vatankhah1 Mohammad Mehdi Arefi2

1
School of Mechanical Engineering,
Shiraz University, Shiraz, Iran Summary
2
School of Electrical and Computer In this paper, an adaptive prescribed performance control method is presented
Engineering, Shiraz University, Shiraz,
for a class of uncertain strict feedback nonaffine nonlinear systems with the
Iran
coupling effect of time-varying delays, dead-zone input, and unknown control
Correspondence directions. Owing to the universal approximation property, fuzzy logic systems
Ramin Vatankhah, School of Mechanical
are used to approximate the uncertain terms in the system. Since there is no
Engineering, Shiraz University, Shiraz
71936-16548, Iran. systematic approach to determine the required upper bounds of errors in con-
Email: rvatankhah@shirazu.ac.ir trol systems, the prior selection of control parameters to have a satisfactory
performance is somehow impossible. Therefore, the prescribed performance
technique as a solution is applied in this study to bring satisfactory performance
indices to the system such as overshoot and steady state performance within a
predetermined bound. Dynamic surface control strategy is also introduced to
the proposed control scheme to address the “explosion of complexity” behav-
ior existing in conventional backstepping methods. To ease the control design,
the mean-value theorem is utilized to transform the nonaffine system into the
affine one. Moreover, with the help of this theorem, the unknown dead-zone
nonlinearity is separated into the linear and nonlinear disturbance-like bounded
term. The proposed method relaxes a prior knowledge of control direction by
employing Nussbaum-type functions, and the effect of time-varying delays are
compensated by constructing the proper Lyapunov-Krasovskii functions. The
proposed controller guarantees that all the closed-loop signals are semiglob-
ally uniformly ultimately bounded and the error evolves within the decaying
prescribed bounds. In the end, in order to demonstrate the superiority of this
method, simulation examples are given.

K E Y WO R D S
adaptive fuzzy control, dead-zone nonlinearity, dynamic surface control, nonaffine systems,
prescribed performance, time-varying delays

1 I N T RO DU CT ION

In the past decades, there has been a significant achievement in controlling nonlinear systems, due to the advent of
backstepping technique, and feedback linearization method as two major controller design of nonlinear systems.1 The

1134 © 2019 John Wiley & Sons, Ltd. wileyonlinelibrary.com/journal/acs Int J Adapt Control Signal Process. 2019;33:1134–1156.
TOORANJIPOUR ET AL. 1135

backstepping strategy as a recursive and systematic technique is used to design a stable virtual controller, which is specif-
ically applied when mismatched conditions exist in the system.2 Moreover, the backstepping design is also utilized to
address the problem of unknown parameters via adaptive backstepping technique.3 Along with all of the benefits, the
integrator backstepping design inherently suffers from the problem of “explosion of complexity” caused by repeated dif-
ferentiation of virtual controllers, specifically when the order of the system is increased. In the work of Swaroop et al,4 the
dynamic surface control (DSC) strategy is proposed to handle the shortcoming of “explosion in terms” by introducing a
first-order filter passed through the virtual controller. Furthermore, the DSC method is extended to adaptive systems with
unknown parameters, which is linear respect to the nonlinear functions.5 Recently, in the work of Liu,6 a linear low pass
filter used in conventional DSC is substituted by nonlinear filters with a positive time-varying integral function, which
can guarantee the asymptotic tracking despite input hysteresis and external disturbances.
Due to structured and unstructured uncertainties without matching conditions exist in many practical problems,
the need for universal function approximators (UFA) such as fuzzy logic and neural network functions become more
required.7 Considerable attempts have been made to apply UFAs to capture the unknowns and uncertainties for the classes
of strict and pure feedback systems based on the adaptive DSC.8,9 The main advantage of fuzzy logic compared to radial
basis function neural networks is combining the expert knowledge to complex systems,10 and the main drawback that
exists in the fuzzy or neural network backstepping methods is updating numerous parameters within fuzzy rules or neu-
ral network nodes adjusted online through adaptive laws. In the work of Arefi et al,11 a novel technique is introduced in
order to update only one parameter.
The basic assumption, which is usually considered in the methods mentioned above, is that it is required to know
the gain sign of the control direction.12,13 If the prior knowledge about the sign of control direction is not known or it
changes with time, many approaches could be applied to handle this problem such as using Nussbaum gain techniques11,14
and utilizing a hysteresis-type function in adaptive fuzzy control.15 For the first time, the Nussbaum gain technique was
used for the class of first-order linear systems by introducing Nussbaum functions,16 and the concept of high-frequency
gain was going to emerge. In the work of Mårtensson,17 the Nussbaum-type technique is extended to adaptive control
of first-order nonlinear systems. Recently, this idea is extended to approximate the sign of nonlinear functions rather
than linear-in-parameters form, which is made by combining UFAs and adaptive laws.14,18 In many aforementioned arti-
cles with unknown gain signs,14,19 there have been some approaches to address the problem of immeasurable states. For
example, in the work of Arefi et al,14 they have proposed an observer-based neural network controller for a class of non-
affine strict feedback systems with the unknown sign of the control direction, and in the work of Chen and Zhang,18
a novel observer-based fuzzy controller is proposed to achieve the global stability, which is made by determining the
approximation domain.
Many controllers are proposed to deal with dead-zone effect, as the most important nonsmooth nonlinearity exist in
hydraulic servo valves, electric motors, etc, which can degrade the performance of the system and, indeed, make the
system unstable.20,21 The primary idea to address the control design of such systems with dead-zone nonlinearity is con-
structing an inverse dead-zone model.22 Another method is using special continuous indicator functions in which the
linear parameters of dead-zone are approximated online.23 However, in the aforementioned approaches, the minimum
and maximum values of dead-zone slopes are assumed to be known. Recently, due to using UFAs, the dead-zone param-
eters are completely approximated, and hence, the requirements on the dead-zone parameters are relaxed.24-26 In such
methods, the dead-zone nonlinearity is separated into the linear and nonlinear bounded term.20
The existence of time delay in the practical systems like dead-zone nonlinearity can bring a poor performance to
the systems and even can make the system unstable. The stability analysis and control design of the time-delay sys-
tems have attracted much attention in the past years.20,27,28 For instance, in the work of Khajeh Talkhoncheh,29 they
developed an observer-based adaptive neural controller for a class of strict-feedback nonlinear systems subjected to
unknown control directions, saturation, and input time delay, which is handled by using an integrator compensator
term in the controller design. Inspired by the aforementioned work,29 barrier Lyapunov functions (BLFs) are added to
the control scheme to constrain outputs.27 Constructing appropriate Lyapunov-Krasovskii functions is another approach
to cancel the unknown time-delay terms.20,28 Recently, using time-varying delays in control design of discrete complex
systems has received a great deal of attention.30-33 Recently, Shen et al,33 have proposed an asynchronous state feedback
controller for Markovian jump time-delay systems by using Lyapunov-Krasovskii functional, and Que et al31 have inves-
tigated a new globally exponentially synchronization criterion for complex dynamical networks by considering time
delay.
In recent years, a great progress has been made to control nonaffine systems. There are numerous systems such as bio-
chemical process, some aircraft dynamics, chemical reactors, etc, which are considered as nonaffine systems.14,34,35 Many
1136 TOORANJIPOUR ET AL.

approaches are introduced to transform the nonaffine systems to the affine one such as using Taylor series expansion35
and utilizing the mean-value theorem.
The performance constraints of control systems have received much attention for many industrial control systems in
terms of safety specifications or system damage. Several methods have been presented to control the performance of the
system. For instance, in the work of Bechlioulis and Rovithakis,36 by introducing performance functions to robust adaptive
control of feedback linearizable multiple-input–multiple-output (MIMO) systems and transforming the “constrained”
system into the equivalent “unconstrained” one, the tracking errors converge to an arbitrary small neighbor of the origin.
Moreover, this strategy can guarantee that the maximum overshoot of the system is constrained to a sufficiently small
prescribed constant. The BLF technique is another common approach to deal with the constrained systems,37 but a piece-
wise smooth BLF is needed to establish the stability of the closed-loop system, so utilizing BLF technique brings about
many obstacles, which are detailed in the work of Han and Lee.38 Recently, Chen,39 has used the advantages of two meth-
ods (BLF and prescribed performance method) to design an adaptive dynamic surface control for a class of unknown
pure-feedback nonlinear switched systems with output constraints.
Although a great progress has been made for adaptive backstepping control based on UFAs such as fuzzy logic and
neural network functions, there are still “open problems” in the control design of practical systems. Using UFAs in the
control design brings about some drawbacks. Traditionally, in the approximation-based control schemes, tracking errors
converge to a residual set, whose size depends on design parameters and upper bound of the unknown uncertainties.
However, there is no systematic approach to determine the required upper bounds, and thus, the prior selection of con-
trol parameters to have a satisfactory performance is somehow impossible.36 This condition can be worse, and the system
performances are severely being affected if unknown time-varying delays and dead-zone nonlinearity as the indispens-
able part of practical systems and actuators are considered, along with unknown disturbances and control directions. As
a result, the urge of using prescribed performance technique, which errors evolve strictly within the prescribed bounds,
is completely sensed. For example, in the work of Zhou et al,40 the authors proposed an adaptive fuzzy tracking con-
trol for a class of pure-feedback nonlinear systems with time-varying delay and unknown dead zone that the quality of
tracking signal such as the transient and steady state of the system has not been guaranteed. Although, in the work of
Zhang et al,41 the exact tracking control of nonlinear systems with time delays and dead-zone input is investigated, the
nonlinear system is in the canonical controllable form. Only a few papers have considered all of the aforementioned
conditions in the system with prescribed functions together. For instance, in the work of Na,42 an adaptive prescribed
performance control is proposed to address the output tracking of nonlinear SISO systems with a nonlinear dead zone
input. However, this control scheme42 suffers from “explosion of complexity” due to the repeatedly differentiating of
the virtual controller. This problem also exists in other works.11,20,28,34,35 Besides of using UFAs, Nussbaum functions,
as a technique to tackle with unknown control directions, are another reason for the poor transient response in the
system.43
Inspired by the preceding discussion, the main question raised with the schemes of DSC technique based on UFAs
with the coupling effect of unknown gain signs, time delays and dead-zone nonlinearity in the system, is how to achieve
the desired performance in which the error evolves strictly within the prescribed bounds. To the best of the authors'
knowledge, the question has not been addressed before, which motivated us for current study. The main contributions of
this paper are summarized as follows:

1. Compared with the work of Zhou et al,40 an adaptive fuzzy DSC technique with prescribed functions is introduced
for a class of strict feedback single-input–single-output (SISO) systems with unknown time-varying delays, input
dead-zone nonlinearity, and unknown control directions.
2. Several restrictive conditions and assumptions used in the literature are relaxed, and knowing about the character-
istic parameters of dead-zone nonlinearity or upper bound of disturbances and delayed functions are all avoided
in the design procedures.

In this paper, to handle the existence of time-varying delays in the system, Lyapunov-Krasovskii functions are used, and
the mean-value theorem is applied to separate the dead-zone nonlinearity into the linear and nonlinear disturbance-like
bounded term. Another usage of the mean-value theorem in this paper is transforming the nonaffine system into the
affine one. Fuzzy logic systems (FLS) are applied to approximate the unknown nonlinear functions. However, using
UFAs have some difficulties such as having burdensome computation when the number of parameters that need to be
updated by increasing adaptation laws. In this paper, to solve the computation burden that comes with updating numerous
parameters, the norm of the ideal weighting vector in FLS is considered as the adaptive estimation parameter instead of
TOORANJIPOUR ET AL. 1137

the elements of the weighting vector.11,44 Unlike in the work of Edalati et al,27 the Nussbaum technique aids us in this paper
to address the problem of completely unknown control directions. To address the problem of repeatedly differentiating of
the virtual controller, the DSC strategy is introduced by using the first-order filter passed through the virtual controller.
The rest of this paper is organized as follows. In Section 2, the problem definition and some preliminaries are defined.
Section 3 proposes an adaptive fuzzy controller based on the dynamic surface method and the prescribed performance
technique. Moreover, a detailed stability analysis is exploited in this section. Section 4 is devoted to validating the proposed
controller via numerical simulation examples. Finally, a conclusion is drawn in Section 5.

2 PROBLEM DEFINITIONS AND PRELIMINARIES

Consider the nonaffine strict-feedback SISO system with unknown dead-zone nonlinearity and time-varying delays as
follows:
.
⎧x1 = 𝑓1 (x1 (t)) + g1 (x1 (t)) x2 + h1 (x1 (t − 𝜏1 (t))) + d1 (t),

⎪⋮
⎪. ( ) ( )
⎪xi = 𝑓i xi (t) + gi xi (t) xi+1 + hi (x1 (t − 𝜏1 (t)), . … , xi (t − 𝜏i (t))) + di (t),
⎨ (1)
⎪⋮
⎪. ( )
⎪xn = 𝑓n xn (t), D (u(t)) + hn (x1 (t − 𝜏1 (t)), . … , xn (t − 𝜏n (t))) + dn (t),

⎩𝑦 = x1
where xi = [x1 , … , xi ]T ∈ Ri , u(t) ∈ R, y ∈ R, 𝜏 i (t), and di (t) ∈ R, i = 1,2, … ,n represent the system states, the
control input, the system output, the unknown time-varying delay of the states, and the unknown external distur-
bances, respectively. D(u(t))) is considered as dead-zone nonlinearity that will be defined later. 𝑓i (xi (t)), gi (xi (t)) and
hi (x1 (t − 𝜏 1 (t)), . … , xi (t − 𝜏 i (t))) , i = 1, … , n, are unknown smooth functions; an 𝑓n (xn (t), D(u(t))) is an unknown smooth
nonaffine function.
The main objectives of the proposed controller are
1. To ensure that the steady and transient responses of the system are evolved within the prescribed performance
functions;
2. To determine an adaptive control input u(t) so that the output of the system (1) eventually converges to the small
predefined neighborhood of the desired trajectory yd , while all the closed-loop signals are semiglobally uniformly
ultimately bounded (SGUUB).
Assumption 1. The unknown time-varying delays 𝜏 i (t), i = 1, … ,n should be bounded and satisfy 0 ≤ 𝜏 i (t) ≤
.
𝜏 0 < ∞, i = 1, … , n, and 𝜏 i ≤ 𝜏 ∗ < 1, where 𝜏 0 and 𝜏 ∗ are unknown positive constants.

Assumption 2. Assume all of the external disturbances satisfy |di | ≤ di , i = 1, . … , n, where di are unknown positive
constants.

Assumption 3. The unknown functions hi (xi (t − 𝜏i (t))), i = 1, … , n, satisfy the following inequality:

| ( )| ∑ i
( )
|hi xi (t − 𝜏i (t)) | ≤ 𝜎i,𝑗 x𝑗 (t − 𝜏𝑗 ) i = 1, … , n, (2)
| |
𝑗=1

where 𝜎 i, j (.) ≥ 0 are unknown bounded functions on any compact set Ci .

Assumption 4. The desired trajectory yd (t) is continuous and bounded on a compact set Cd . Therefore, there exists
a common positive constant d such that |yd (t)| ≤ d.
The dead-zone input, which has been shown in Figure 1 is described as23

⎧gr (u(t)) u > br ,



D(u(t)) = ⎨0 bl ≤ u ≤ br , (3)

⎩gl (u(t)) u < bl ,
1138 TOORANJIPOUR ET AL.

FIGURE 1 Nonsymmetric dead-zone representation

where br and bl are the unknown parameters of the dead-zone but their signs are known. The derivative of the
dead-zone functions gr (u(t)) and gl (u(t)) are smooth, which satisfy the following inequalities:
{
0 ≤ kl1 ≤ gl′ (u) ≤ kl2 u ∈ (−∞, bir ]
[ (4)
0 ≤ kr1 ≤ gr′ (u) ≤ kr2 u ∈ bil , ∞) ,
where kl1 , kl2 , kr1 , and kr2 are unknown positive constants. By using the mean-value theorem, there exist 𝜆l ∈ (−∞, br ]
and 𝜆r ∈ [bl , ∞) such that
{
gl (u) = gl (u) − gl (bl ) = gl′ (𝜆l )(u − bl ),
(5)
gr (u) = gr (u) − gr (br ) = gr′ (𝜆r )(u − br ).
Therefore, the dead-zone nonlinearity is separated into the linear and nonlinear parts as
D(u) = K T Ωu + 𝜑(u), (6)
where K(t) = [gr′ (𝜆r ), gl′ (𝜆l )]T and Ω = [Ωr , Ωl ]T and
{
1 u > bl
Ωr (t) =
0 u ≤ bl
{ (7)
1 u < br
Ωl (t) =
0 u ≥ br
and 𝜑(u) is obtained as
⎧−gr′ (𝜆r )br u ≥ br
⎪ [ ]
𝜑(u) = ⎨− gr′ (𝜆r ) + gl′ (𝜆l )bl u bl < u < br (8)
⎪ ′
⎩−gl (𝜆l )bl u ≤ bl .
Owing to the smooth and bounded derivative and gr′ 𝜑(u) is also bounded as |𝜑(u)| ≤ p∗ , where p∗ is a positive
gl′ ,
constant with the value of p∗ = (kl2 + kr2 ) max (br , bl ). This form will help us to cancel out the nonlinearity terms in
the stability analysis.

Lemma 1. Mean Value Theorem45 : If f(x,u) : Rn × R → R is differentiable at each point of an open set Rn × (a,b) and is
continuous at each endpoint, then there exists 𝛽 ∈ (a,b) such that

𝜕𝑓 ||
𝑓 (x, b) = 𝑓 (x, a) + (b − a). (9)
𝜕u ||u=𝛽
TOORANJIPOUR ET AL. 1139

In order to transform the nonaffine system into affine one, mean value theorem is applied into 𝑓 (xn (t), D(u(t))) as
follows:
𝜕𝑓 ||
𝑓 (x(t), D(u(t))) = 𝑓 (x(t), 0) + D(u(t)) (10)
𝜕D(u(t)) ||D(u(t))=𝛽
with 𝛽 ∈ (0, D(u)). Then, (10) can be written as follows:

𝑓 (x(t), D(u(t))) = 𝑓n (x) + gn (x) D(u), (11)


𝜕𝑓 |
where fn (x(t)) = f(x(t), 0) and gn (x(t)) = |
𝜕D(u(t)) |D(u(t))=𝛽
. By substituting (6) into (11), we have
( )
𝑓 (x(t), D(u(t))) = 𝑓n (x) + gn (x) K T Ωu + 𝜑(u) . (12)

Assumption 5. Functions gi (.), i = 1, 2, … , n, are nonsingular smooth functions, their signs are unknown, and there
exist positive constants g and gi such that 0 < g ≤ |gi (.)| ≤ gi .
i i

Definition 1 (See the work of Nussbaum46 ).


An even function N(𝜍) : R → R is called Nussbaum function if it has the following properties:
z
1
lim sup N(𝜁)d𝜁 = +∞ (13)
z→∞ z∫
0

z
1
lim inf N(𝜁)d𝜁 = −∞. (14)
z→∞ z∫
0

Some common Nussbaum functions are exp(𝜍 )cos (𝜍 ), e𝜍 cos(𝜋 𝜍2 ) and exp(𝜍 2 )cos( 𝜋𝜍
2 2 2

2
) . In this article, exp(𝜍 2 ) cos (𝜍 2 )
is employed.

Lemma 2 (See the work of Tong et al47 ).


If 𝜍(t) and V(t) > 0 are smooth functions within intervals [0, tf ], gn (x(𝜏)) is a nonzero time-varying parameter, which takes
value in the unknown closed intervals I : [l− , l+ ], c0 is a suitable constant, and the following inequality:
t

n
[ ] .
V(t) ≤ c0 + e −c1 t
gi N(𝜍i ) + 1 𝜍 i ec1 𝜏 d𝜏 0 ≤ t < t𝑓 (15)
i=1

0
∑n t . c𝜏
is true for positive c1 , then it can be concluded that V(t), 𝜍(t), and i=1 ∫0 [gi N(𝜍i )+1] 𝜍 i e d𝜏 must be bounded on 0 ≤ t < tf
1

for a special Nussbaum-type function N(𝜍) = exp (𝜍 2 ) cos (𝜍 2 ).

Lemma 3 (See the work of Ge and Tee48 ).


If the compact set Ωczi , 1 ≤ i ≤ n, is defined as Ωczi = {zi | |zi | < 0.8814 ki }, where ki is an arbitrary positive constant, then
z
for zi ∉ Ωczi , the inequality (1 − 2 tanh2 ( ki )) ≤ 0 is satisfied.
i

2.1 Function approximation using fuzzy structures


In this section, FLSs are used to approximate and identify unknown nonlinear function f (X), X ∈ Rn . In this article,
Mamdani-type inference system with singleton fuzzifier and the center average defuzzifier are used, and therefore, the
rules are constructed as follows:
If x1 is A11 and x2 is A12 and … and xn is A1n , then C1 ,
⋮ (16)
If x1 is AN1 and x2 is AN2 and … and xn is ANn , then CN ,
where Aij and Ci , i = 1, … , N, j = 1, … , n, are the linguistic variables of if-part and then-part characterized by membership
functions 𝜇Aij and 𝜇Ci , respectively. The antecedent grade of rule i is calculated by T-norm operator as follows:

𝜇i (x1 , . … , xn ) = 𝜇Ai1 (x1 )𝜇Ai2 (x2 )…𝜇Ain (xn ). (17)


1140 TOORANJIPOUR ET AL.

By using center average defuzzifier, the crisp output of the FLS is obtained as10

N
𝜇i 𝑦 i
i=1
𝑦= = 𝜃 T S (X) , (18)

N
𝜇i
i=1
[ ]
𝜇1 𝜇
where N is the number of rules and 𝑦i is the point, where 𝜇Ci has the maximum value S(X) = ∑N , .…, ∑N N and
i=1 𝜇i i=1 𝜇i
𝜃 = [𝑦1 , .…, 𝑦N ]T . Now, we can approximate 𝑓 (X) defined on a compact set Ω as

𝑓 (X) = 𝜃 T S(X) + 𝛿 𝛿 ≤ 𝜀, (19)
where 𝛿 is a fuzzy approximation error, 𝜀 is an upper bound 𝛿, and 𝜃 * is an optimal parameter, which is defined as
{ }
| |
𝜃 ∗ = arg min sup |𝑦 − 𝑓 (X)| , (20)
| |
where 𝜃̂ is the estimation of 𝜃 * and 𝜃̃ = 𝜃 ∗ − 𝜃̂ is the parameter error. Equation (19) is valid just on the compact set Ω. For
the sake of simplicity, the new parameter is defined as
𝜓 ∗ = ‖𝜃 ∗ ‖2 , (21)
where ||.|| denotes to the Euclidean norm of a vector.
Remark 1. By this way, instead of updating the vector 𝜃, we can update just scalar 𝜓 so that, in the stability analysis,
̂ and 𝜓̃ instead of 𝜃 ∗ , 𝜃̂ , and 𝜃,
we use 𝜓 ∗ , 𝜓, ̃ respectively.

2.2 Prescribed performance analysis


According to the work of Bechlioulis and Rovithakis,36 a prescribed transient and steady state of the system can be achieved
by defining performance functions h(t) if they satisfy the following properties.
1. h(t) is a smooth, positive, and decreasing function,
2. limt→∞ h(t) = h∞ > 0.
For example, h(t) = (h0 − h∞ )e−nt + h∞ satisfies aforementioned properties, where h0 and n are positive constants. The
aim is that the error e(t) = y − yd evolves strictly within the prescribed functions as follows42 :
−𝛿min h(t) < e(t) < 𝛿max h(t), (22)
where 𝛿 min > 0 and 𝛿 max > 0 are chosen as constant parameters.
Remark 2. The constant term h∞ denotes to maximum steady state error, which can be prescribed and adjusted
arbitrarily reflecting the resolution size of the measurement systems. Performance functions can also determine the
maximum overshoot of the system by adjusting 𝛿 max , and from (22), it is apparent that 𝛿 max = 0 or 𝛿 min = 0 mean
we have no overshoot or undershoot in the system. Furthermore, the transient response in a finite time can also be
predefined by n.
In order to represent (22) in an equality form without constraints, the smooth and strictly increasing function 𝜙(s1 )
of the transformed error s1 is introduced as
e(t) = h(t) 𝜙(s1 ), (23)
where 𝜙 is a function that has specific properties detailed in the work of Bechlioulis and Rovithakis.36
By getting proper 𝜙i , which possesses aforementioned properties, inequality (22) is satisfied by (23). The following
error transformation is employed here as19 :
𝛿max es1 − 𝛿min e−s1
𝜙(s1 ) = . (24)
es1 + e−s1
The inverse transformation of (24) is
( )
e(t) 1 𝜆(t) + 𝛿min
s1 = 𝜙 −1
= ln , (25)
h 2 𝛿max − 𝜆(t)
TOORANJIPOUR ET AL. 1141

e(t)
where 𝜆(t) = . As a result, the transformed error dynamics is obtained as42
h(t)

( . )
. . . h(t)e(t)
s 1 = 𝜈1 x 1 − 𝑦d − , (26)
h(t)
[ ]
1 1 1 h0 (𝛿min +𝛿max )
where 𝜈1 = 𝜆+𝛿min
− 𝜆−𝛿max
, fulfilling 𝜈1 ≥ = 𝜈M ≥ 0.
2h 2(e(0)+h0 𝛿min )(h0 𝛿max −e(0))

3 CO NTRO LLER DESIGN AND STABILITY ANALYSIS

In this section, first, an adaptive fuzzy control scheme is designed by combining the DSC technique with prescribed
performance functions, and after that, by getting a proper Lyapunov candidate in each step, the stability of the whole
system is proved.

3.1 Controller design


The design procedure is detailed as follows.
Step 1: By using (26) and (1), we have
( . )
. . h1
s 1 = 𝜈1 𝑓1 + g1 x2 + h1 + d1 − 𝑦d − e(t) . (27)
h1

The first virtual controller 𝛼 1 is designed as

s1 𝜈1
𝛼1 = N(𝜍1 )𝜓̂ 1 S1T (Z1 )S1 (Z1 ), (28)
2a21

where S1 (Z1 )is a vector of fuzzy basis function defined as (18) and Z1 will be defined later, a1 is a positive design constant,
𝜓̂ 1 is the estimate of 𝜓 1 * defined as (21), and N(𝜍 1 ) is an even Nussbaum function defined as Definition 1. 𝜍 1 and 𝜓 1 are
adaptive parameters adjusted through the following adaptive laws:

. s21 𝜈12
𝜍1 = 𝜓̂ 1 S1T (Z1 )S1 (Z1 ), (29)
2a21

. 𝛾1 2 2 T
𝜓̂ 1 = s 𝜈 S (Z1 )S1 (Z1 ) − 𝜂1 𝜓̂ 1 , (30)
2a21 1 1 1

where 𝛾 1 and 𝜂 1 are positive design constants. By applying the DSC method, a new state variable q2 is introduced, and by
letting 𝛼 1 pass through a first-order filter with time constant 𝜁 2 , it can be obtained
.
𝜁2 q2 + q2 = 𝛼1 , q2 (0) = 𝛼1 (0). (31)

The output error of the first-order filter X2 is defined as

X2 = q2 − 𝛼1 . (32)

The second error surface s2 according to the DSC method is considered as

s2 = x2 − q2 . (33)
1142 TOORANJIPOUR ET AL.

By using (31)-(33), the time derivatives of q2 and X2 become


. 𝛼1 − q2 X
q2 = = − 2, (34)
𝜁2 𝜁2
. X ( )
. .
X 2 = q2 − 𝛼 1 = − 2 + H 1 x1 , h1 , 𝑦d , 𝜓̂ 1 , 𝜍1 , (35)
𝜁2
where H 1 (.) is a continuous function.
Step i (i = 2, … , n): In these steps, similar to the first step, virtual controllers should pass through the first-order filters
as follows:
.
𝜁i+1 qi+1 + qi+1 = 𝛼i , qi+1 (0) = 𝛼i (0), i = 2, … , n − 1, (36)
where qi+1 is a filtered virtual controller and 𝜁 i+1 is a positive time constant variable. The output error of the first order
filter Xi+1 and error surface si are also defined as follows:
si+1 = xi+1 − qi+1 i = 2, … , n − 1, (37)

Xi+1 = qi+1 − 𝛼i i = 2, … , n − 1. (38)


Design the virtual controller 𝛼 i as follows:
si
𝛼i = N(𝜍i )𝜓̂ i SiT (Zi )Si (Zi ), i = 2, … , n − 1, (39)
2a2i
where Si (Zi ) is a vector of fuzzy basis function; ai and 𝜓̂ i are a positive design constant and estimation of 𝜓 i * , respectively.
By using (36)-(38) and (1), similar to the first step, the time derivative of si and Xi+1 are
( )
. X
si = 𝑓i + gi xi+1 + hi + di + i , i = 2, … , n − 1, (40)
𝜁i
. Xi+1 ( )
. .
X i+1 = qi+1 − 𝛼 i = − + H i xi , hi , 𝜍i , 𝜓̂ i , i = 2, … , n − 1, (41)
𝜁i+1
where H i (.) is a continuous function. Adaptive parameters 𝜁 i and 𝜓̂ i are tuned as
. s2i
𝜍i = 𝜓̂ i SiT (Zi )Si (Zi ), i = 2, … , n − 1, (42)
2a2i
. 𝛾i 2 T
𝜓̂ i = s S (Zi )Si (Zi ) − 𝜂i 𝜓̂ i , i = 2, … , n − 1, (43)
2a2i i i
where 𝛾 i and 𝜂 i are positive design constants.
Step n: The time derivative of sn according to (1) is obtained as
( )
. ( ) X
sn = 𝑓n + gn K T Ω𝑢 + 𝜑 (u) + hn + dn + n , (44)
𝜁n
and u is designed as
sn
u= N (𝜍n ) 𝜓̂ n Sn T (Zn ) Sn (Zn ) . (45)
2a2n
The adaptation laws as the previous steps are
sn 2
.
𝜍n =𝜓̂ n Sn T (Zn ) Sn (Zn ) , (46)
2a2n
. 𝛾n
𝜓̂ n = 2 sn 2 Sn T (Zn ) Sn (Zn ) − 𝜂n 𝜓̂ n , (47)
2an
where 𝛾 n and 𝜂 n are positive design constants.
Remark 3. It is worth pointing out that the adaptation laws, virtual controller, and the control input u do not require
parameters mentioned in the Assumptions 1 to 5 such as di , 𝜏0 , 𝜏 ∗ , g , gi , and 𝜎 i, j i = 1, … ,n, so they are just employed
i
for stability analysis.

Remark 4. One of the main superiority of the proposed controller over the existing ones in which Nussbaum technique
with UFAs are used34,42 is the number of adaptation laws in each step. In this paper, two adaptation laws are used in
each step instead of three, which can effectively reduce the computation burden of the updating parameters.
TOORANJIPOUR ET AL. 1143

3.2 Stability analysis


Theorem 1. Under Assumptions 1 to 5, consider the closed-loop system consisting of the nonlinear system (1), and the
control law (45) with the adaptive laws (29)-(30), (42)-(43), and (46)-(47) are applied. Then, the following properties
are held:
1. All the closed-loop signals are SGUUB. Since the fuzzy approximator is guaranteed on the compact set Ω, the stability
results are semiglobal. Furthermore, the transient performance and the steady response of the system is under control
of the prescribed functions defined as (22) at all times.
2. The transformed error s1 will finally converge to a small predefined residual set around zero in the mean square sense.

Remark 5. Due to the selection of a sufficiently large compact set Ω, there will be no concern about the semiglobal
results.

Proof. Step 1: Consider the following Lyapunov candidate as


1 2 1
V1 = s + 𝜓̃ 1 𝜓̃ 1 + Vt,1 (t) , (48)
2 1 2𝛾1
where 𝜓̃ 1 = 𝜓 ∗ 1 − 𝜓̂ 1 is an adaptive parameter defined as (21) and 𝛾 1 is a positive design constant, respectively. Vt,1 as
a Lyapunov-Krasovskii function is introduced to tackle with time-delay terms defined as
t
e−r(t−𝜏0 )
Vt,1 (t) = ers 𝜎 2 1,1 (x1 (s)) ds, (49)
1 − 𝜏∗ ∫
t−𝜏1 (t)

where r is a positive constant. The time derivative of Vt,1 (t) is


[ ]
. e−r(t−𝜏0 ) rt 2 . r (t−𝜏1 (t)) 2
V t,1 (t) = e 𝜎 1,1 (x 1 ) − (1 − 𝜏 1 (t)) e 𝜎 1,1 (x 1 (t − 𝜏 1 (t))) − r Vt,1 , (50)
1 − 𝜏∗
and taking the time derivation of (48) using (50) and (27) yields
[ . ]
. . h 1 .
V 1 = s1 𝜈1 𝑓1 (x1 ) + g1 (x1 ) x2 + d1 − 𝑦d + h1 (x1 (t − 𝜏1 (t))) − 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1
s1 +
h 𝛾1
(51)
[ ]
e−r(t−𝜏0 ) rt 2 .
+ e 𝜎 1,1 (x1 ) − (1 − 𝜏 1 (t)) er(t−𝜏1 (t)) 𝜎 2 1,1 (x1 (t − 𝜏1 (t))) .
1 − 𝜏∗
By using Assumption 1, the following inequality is obtained:
[ . ]
. . h 1 .
V 1 ≤ s1 𝜈1 𝑓1 (x1 ) + g1 (x1 ) x2 + d1 − 𝑦d + h1 (x1 (t − 𝜏1 (t))) − s1 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1
h 𝛾1 (52)
er𝜏0 2
+ 𝜎 1,1 (x1 (t)) − 𝜎 2 1,1 (x1 (t − 𝜏1 (t))) .
1 − 𝜏∗
In order to cancel out the time-delay term 𝜎 2 1,1 (x1 (t − 𝜏 1 (t))) and h1 (x1 (t − 𝜏 1 (t))), Assumption 3 and Young's inequality
are utilized as
[ . ]
. . h 1 .
V 1 ≤ s1 𝜈1 𝑓1 (x1 ) + d1 − 𝑦d − s1 + s1 𝜈1 g1 (x1 ) x2 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1
h 𝛾 1

er𝜏0 2 1
+ 𝜎 1,1 (x1 (t)) − 𝜎 2 1,1 (x1 (t − 𝜏1 (t))) + h2 1 (x1 (t − 𝜏1 (t))) + s21 𝜈12 ,
1 − 𝜏∗ 4
[ . ] (53)
. h 1 .
≤ s1 𝜈1 𝑓1 (x1 ) + d1 − 𝑦d − s1 + s1 𝜈1 g1 (x1 ) x2 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1
h 𝛾 1

er𝜏0 2 1
+ 𝜎 1,1 (x1 (t)) + s21 𝜈12 .
1−𝜏 ∗ 4
1144 TOORANJIPOUR ET AL.

By using Assumption 2, (32) and (33) yield to


[ . ]
. . h 1 .
V 1 ≤ s1 𝜈1 𝑓1 (x1 ) − 𝑦d − s1 + s1 𝜈1 g1 (s2 + X2 + 𝛼1 ) + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1
h 𝛾1
(54)
er𝜏0
1 1s21 𝜈12 2
+ 𝜎 2 1,1 (x1 (t)) + s21 𝜈12 + + 𝜌1 2 d1 ,
1 − 𝜏∗ 4 2𝜌1 2 2
where 𝜌1 and c1 are positive constants. By adding and subtracting c1 s1 2 𝜈 1 2 term to the previous inequality and
introducing tanh(.) into the above equation can reach to
( ) [ .
. 1 1 . h 1
V 1 ≤ −s21 𝜈12 c1 − − 2
+ s1 𝜈1 𝑓1 (x1 ) + c1 s1 𝜈1 − 𝑦d − s1 + s1 𝜈1 g1 2
4 2𝜌1 h 2
( ) ]
er𝜏0 2 2 s1 1 . (55)
+ 𝜎 1,1 (x1 (t)) tanh2 + s1 𝜈1 g1 𝛼1 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1 + s2 2
1 − 𝜏∗ s1 𝜈1 k1 𝛾1
[ ( )] r𝜏
1 2 s1 e 0 2
+ X2 2 + 𝜌1 2 d1 + 1 − 2tanh2 𝜎 1,1 (x1 (t)) ,
2 k1 1 − 𝜏∗
where k1 > 0 is a design constant. Then, fuzzy logic is utilized to estimate nonlinear part of the above equations as
follows:
( )
. 1 1 1 .
V 1 ≤ −s1 𝜈1 c1 − −
2 2
+ s1 𝜈1 𝑓 1 + s1 𝜈1 g1 𝛼1 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1 + s2 2
4 2𝜌1 2 𝛾1
[ ( )] r𝜏 (56)
1 2 2 s1 e 0 2
+ X2 + 𝜌1 d1 + 1 − 2tanh
2 2
𝜎 1,1 (x1 (t)) ,
2 k1 1 − 𝜏∗
. ( )
. er𝜏0 2 s1
where 𝑓 1 = 𝑓1 (x1 ) + c1 s1 𝜈1 − 𝑦d − h s1 + 12 s1 𝜈1 g1 2 + 1−𝜏 ∗
𝜎 1,1 (x1 (t)) s 2𝜈 tanh2 k1
is approximated by FLS as
h 1 1


𝑓 1 (Z1 ) = 𝜃1T S1 (Z1 ) + 𝛿1 (Z1 ) |𝛿1 (Z1 ) | ≤ 𝜀1 (57)
.
.
and 𝜀1 is a positive constant. By substituting (57) into (56), using Young's inequality, where Z1 = [s1 , x1 , 𝑦d , h, h]T ∈ R5 ,
we have
( )
. 1 1 ( ∗ ) 1 .
V 1 ≤ −s1 𝜈1 c1 − −
2 2
+ s1 𝜈1 𝜃1T S1 (Z1 ) + 𝛿1 (Z1 ) + s1 𝜈1 g1 𝛼1 + 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1 + s2 2
4 2𝜌1 2 𝛾1
[ ( )] r𝜏
1 2 2 s1 e 0 2
+ X2 + 𝜌1 d1 + 1 − 2tanh
2 2
𝜎 1,1 (x1 (t))
2 k1 1 − 𝜏∗
( ) (58)
3 1 s21 𝜈12 ∗ T 1 2 1 2 1 .
≤ −s21 𝜈12 c1 − − + 𝜓 S
1 1 (Z 1 ) S 1 (Z 1 ) + a 1 + 𝜀1 + s 𝜈 g
1 1 1 1 𝛼 + 𝜓
̃ 𝜓
̃
1 1
4 2𝜌1 2 2a1 2 2 2 𝛾1
[ ( )] r𝜏
1 2 s1 e 0 2
− 𝑟𝑉 t,1 + s2 2 + X2 2 + 𝜌1 2 d1 + 1 − 2tanh2 𝜎 1,1 (x1 (t)) ,
2 k1 1 − 𝜏∗

where as it has been mentioned in (21) 𝜓 1 * = ||𝜃 1 * ||2 . By substituting the adaption laws (29)-(30) and the virtual
control input 𝛼 1 (28) in (58), we have
( )
. 3 1 𝜂1 .
V 1 ≤ −s21 𝜈12 c1 − − − 𝜓̃ 1 𝜓̃ 1 − 𝑟𝑉 t,1 + (g1 N1 (𝜍1 ) + 1) 𝜍 1
4 2𝜌1 2 2𝛾1
[ ( )] r𝜏 (59)
s1 e 0 2
2 2
+ s2 + X2 + 1 − 2tanh 2
𝜎 1,1 (x1 (t)) + Δ1 ,
k1 1 − 𝜏∗
𝜂 2
where Δ1 = 12 a1 2 + 12 𝜀1 2 + 2𝛾1 𝜓1 ∗ 2 + 12 𝜌1 2 d1 is a positive constant and let c1 > 3
4
+ 1
2𝜌1 2
.
1
Step k (k = 2, … , n − 1): Consider the following Lyapunov candidate as
1 1 1
Vk = Vk−1 + sk 2 + 𝜓̃ k 𝜓̃ k + Xk 2 + Vt,k , (60)
2 2𝛾k 2
TOORANJIPOUR ET AL. 1145

where 𝛾 k > 0, 𝜓̃ k = 𝜓 ∗ k − 𝜓̂ k and Vt,k (t) is defined as


t
e−r(t−𝜏0 ) ∑
k
Vt,k (t) = ers 𝜎 2 k,i (xi (s)) ds. (61)
1 − 𝜏 ∗ i=1 ∫
t−𝜏i (t)

Taking the time derivative of (61) yields


[ k ]
. e−r(t−𝜏0 ) ∑ rt 2 . r (t−𝜏i (t)) 2
V t,k (t) = e 𝜎 k,i (xi ) − (1 − 𝜏 i (t)) e 𝜎 k,i (xi (t − 𝜏i (t))) − 𝑟𝑉 t,k
1 − 𝜏 ∗ i=1
(62)
e r𝜏0 ∑
k

k
≤ −𝑟𝑉 t,k + 𝜎 2 k,i (xi (t)) − 𝜎 2 k,i (xi (t − 𝜏i (t))) .
1 − 𝜏∗ i=1 i=1

By using Young's inequality and combining the time derivative of (60) with (37)-(38), we get
[ ]
. . ( ) ( ) X 1 . . .
V k = V k−1 + sk 𝑓k xk + gk xk xk+1 + hk + dk + k + 𝜓̃ k 𝜓̃ k + Xk X k + V t,k
𝜁k 𝛾k
[ ]
. ( ) ( ) X
≤ V k−1 + sk 𝑓k xk + gk xk (sk+1 + Xk+1 + 𝛼k ) + hk + dk + k (63)
𝜁k
er𝜏0 ∑ 2 ∑
k k
1 . .
+ 𝜓̃ k 𝜓̃ k + Xk X k − 𝑟𝑉 t,k + 𝜎 k,i (xi (t)) − 𝜎 2 k,i (xi (t − 𝜏i (t))) .
𝛾k 1 − 𝜏 ∗ i=1 i=1

When Assumptions 2-3 are held and by using (41), we get


[ ]
. . ( ) ( ) Xk
V k ≤ V k−1 − ck sk + sk 𝑓k xk + gk xk (sk+1 + Xk+1 + 𝛼k ) +
2
+ ck sk
𝜁k
( ( ))
sk 2 1 2 2 1 2 1 . Xk
+ + 𝜌k dk + sk + 𝜓̃ k 𝜓̃ k + Xk − + H k xk , hk , 𝜍k , 𝜓̂ k − 𝑟𝑉 t,k (64)
2𝜌k 2 2 4 𝛾k 𝜁k
er𝜏0 ∑ 2
k
+ 𝜎 k,i (xi (t)) .
1 − 𝜏 ∗ i=1

Rewrite the following equation as:


( ) [ ( ) ]
. . 1 1 ( ) sk g k
2 x
k
V k ≤ V k−1 − sk 2 ck − − + sk 𝑓k xk + + ck sk
2 2𝜌k 2 2
( )
( ) 1 1 1 .
+ s k+1 + X k+1 + sk gk xk 𝛼k + X k
2 2 2
− + Xk H k + 𝜓̃ k 𝜓̃ k (65)
𝜁k 2 𝜁 k 𝛾 k

er𝜏0 ∑ 2
k
1
+ 𝜌k 2 dk − 𝑟𝑉 t,k + 𝜎 k,i (xi (t)) .
2 1 − 𝜏 ∗ i=1

Similar to the step 1, tanh(.) is introduced to (65) as follows:


( )
. . 1 1 ( )
V k ≤ V k−1 − sk ck − −
2
2
+ sk 𝑓 k + s2 k+1 + X 2 k+1 + sk gk xk 𝛼k
2 2𝜌k
( )
1 1 1 . 1
+ X 2k − + Xk H k + 𝜓̃ k 𝜓̃ k + 𝜌k 2 dk − 𝑟𝑉 t,k (66)
𝜁k 2 𝜁k 𝛾k 2
[ ( )] r𝜏
e 0 ∑ 2
k
sk
+ 1 − 2tanh2 𝜎 k,i (xi (t)) .
kk 1 − 𝜏 ∗ i=1
[ ( ) sk gk 2 (xk )
( )∑ ]
er𝜏0 2 sk k
Fuzzy logic is used to approximate 𝑓 k = 𝑓k xk + 2
+ ck sk + 1−𝜏 ∗ sk
tanh2 kk i=1 𝜎 2
k,i (x i (t)) as (19)


𝑓 k (Zk ) = 𝜃kT Sk (Zk ) + 𝛿k (Zk ) |𝛿k (Zk ) | ≤ 𝜀k , (67)
1146 TOORANJIPOUR ET AL.

where Zk = [xk , sk ]T ∈ Rk+1 and 𝜀k is a positive constant. By applying Young's inequality, we have
( )
. . 1 sk 2 1 1
V k ≤ V k−1 − sk ck − 1 −
2
2
+ 2
𝜓k ∗ Sk T (Zk ) Sk (Zk ) + ak 2 + 𝜀k 2 + s2 k+1
2𝜌k 2ak 2 2
( )
( ) 1 1 1 . 1
+ X 2 k+1 + sk gk xk 𝛼k + X 2 k − + Xk H k + 𝜓̃ k 𝜓̃ k + 𝜌k 2 dk − 𝑟𝑉 t,k (68)
𝜁k 2 𝜁k 𝛾k 2
[ ( )] r𝜏
e 0 ∑ 2
k
sk
+ 1 − 2tanh2 𝜎 k,i (xi (t)) ,
kk 1 − 𝜏 ∗ i=1

where ak is a positive design constant. By substituting the adaptation laws as (42)-(43) and the virtual controller 𝛼 k
(39), we can get
( )
. . 1 . 𝜂k ∗ 2 1 2 1 2
V k ≤ V k−1 − sk ck − 1 −
2
+ (gk N (𝜍k ) + 1) 𝜍 k + 𝜓 + ak + 𝜀k + s2 k+1
2𝜌k 2 2𝛾k k 2 2
( )
1 1 𝜂 k 1
+ X 2 k+1 + X 2 k − + Xk H k − 𝜓̃ k 𝜓̃ k + 𝜌k 2 dk − 𝑟𝑉 t,k (69)
𝜁k 2 𝜁k 2𝛾k 2
[ ( )] r𝜏
e 0 ∑ 2
k
sk
+ 1 − 2tanh2 𝜎 k,i (xi (t)) .
kk 1 − 𝜏 ∗ i=1

Rewrite (69) as
( ) ( )
. 3 1 ∑
k
1 ∑k
. ∑k
𝜂i
Vk ≤ −s21 𝜈12 c1 − − − s2i
ci − 1 − + (gi N (𝜍i ) + 1) 𝜍 i − 𝜓̃ i 𝜓̃ i
4 2𝜌1 2i=2
2𝜌 i
2
i=1 i=1
2𝛾 i
( )
∑k
( 2 ) ∑ k
1 1 ∑ k
∑ k
∑k
+ X i+1 + s2 i+1 + X 2i − + X H
i i − r V t,i + Δi (70)
i=2 i=2 𝜁i 2 𝜁i i=2 i=1 i=1
k [ ( )] r𝜏
∑ s𝑗 e 0 ∑ 2
k
+ 1 − 2tanh2 𝜎 k,i (xi (t)) ,
𝑗=1
k𝑗 1 − 𝜏 ∗ i=1
𝜂
where Δk = 2𝛾k 𝜓k 2 + 12 ak 2 + 12 𝜀k 2 + 12 𝜌k 2 dk ≥ 0 and let ci > 1 + 2𝜌1 2 , i = 1, … , k.

k i
Step n: In this final step, eventually, the actual control input u will be obtained. The final Lyapunov function is
chosen as
1 1 1
Vn = Vn−1 + sn 2 + 𝜓̃ n 𝜓̃ n + Xn 2 + Vt,n , (71)
2 2𝛾n 2
where 𝛾 n > 0 is a design constant and Vt,n is defined as
t
e−r(t−𝜏0 ) ∑
n
Vt,n (t) = ers 𝜎 2 n,i (xi (s)) ds. (72)
1 − 𝜏 ∗ i=1 ∫
t−𝜏i (t)

Similar to the derivations in the step k along with (6) and (44), we have
[ ]
. . ( T ) Xn
V n ≤ V n−1 + sn 𝑓n (x) + gn (x) K Ω𝑢 + 𝜑 (u) + hn (t − 𝜏n (t)) + dn (t) +
𝜁n
(73)
1 . . r𝜏
e 0 ∑n
∑n
+ 𝜓̃ n 𝜓̃ n + Xn X n − 𝑟𝑉 t,n + 𝜎 2 n,i (xi (t)) − 𝜎 2 n,i (xi (t − 𝜏i (t))) .
𝛾n 1 − 𝜏 ∗ i=1 i=1

By using the fact that |𝜑(u)| ≤ p* and Young's inequality along with applying Assumptions (1-3) and using tanh(.) into
the following equation, we have
( ) ( )
. . 1 1 1 1
V n ≤ V n−1 − sn cn − −
2
+ sn 𝑓 n + p + sn gn (x) K Ω𝑢 + X n
∗2 T 2

2 2𝜌n 2 𝜁n 2 𝜁n
[ ( )] r𝜏 (74)
e 0 ∑ 2
n
1 . 1 2 sn
+ Xn H n + 𝜓̃ n 𝜓̃ n + 𝜌n dn − 𝑟𝑉 t,n + 1 − 2tanh 2
𝜎 n,i (xi (t)) ,
𝛾n 2 kn 1 − 𝜏 ∗ i=1
TOORANJIPOUR ET AL. 1147

where kn > 0, cn is a positive parameter, which will be given later, and 𝑓 n is


( ) n
sn gn 2 (x) er𝜏0 2 sn ∑ 2
𝑓 n = 𝑓n (x) + + cn sn + tanh2
𝜎 n,i (xi (t)) . (75)
4 1 − 𝜏 sn
∗ kn i=1

Now, fuzzy logic as previous steps help us to approximate 𝑓 n as



𝑓 n (Zn ) = 𝜃nT Sn (Zn ) + 𝛿n (Zn ) |𝛿n (Zn ) | ≤ 𝜀n , (76)

where Zn = [x, sn ]T ∈ Rn+1 and 𝜀n is an unknown positive constant. Equation (74) is rewritten as follows by using (76),
adaptive laws (46)-(47), and the control input u defined as (45)
( ) ( )
. 3 1 ∑ n
1 ∑ n
∑n
𝜂i
V n ≤ −s1 𝜈M c1 − −
2 2
− 2
si ci − 1 − −r Vt,i − 𝜓̃ i 𝜓̃ i
4 2𝜌1 2 i=2
2𝜌 i
2
i=1 i=1
2𝛾 i
( )

n−1
( 2 ) ∑ n−1
. ( ). ∑ n
1 1
+ X i+1 + s2 i+1 + (gi N (𝜍i ) + 1) 𝜍 i + gn N (𝜍n ) K T Ω + 1 𝜍 n + Xn 2 − (77)
i=1 i=1 i=2 𝜁n 2 𝜁n
n [ ( )]
∑n
er𝜏0 ∑ ∑
n
si ∑n
+ Xi H i + 1 − 2tanh 2
𝜎 𝑗,i (xi (t)) +
2
Δi ,
i=2
1 − 𝜏 ∗ 𝑗=1 i=1 ki i=1

𝜂n
+ 12 an 2 + 12 𝜀n 2 + 12 𝜌n 2 dn + p 2 ≥ 0. Then, (77) can be further written as

𝜓 2

where Δn = 2𝛾n n
( ) ( )
. 3 1 ∑
n
1
Vn ≤ −s21 𝜈M 2 c1 − − − s2i ci − 1 −
4 2𝜌1 2 i=2
2𝜌i 2

n
∑n
𝜂i ∑
n−1
. ( ).
−r Vt,i − 𝜓̃ i 𝜓̃ i + + (gi N (𝜍i ) + 1) 𝜍 i + gn N (𝜍n ) K T Ω + 1 𝜍 n (78)
i=1 i=1
2𝛾i i=1
( ) n [ ( )]
∑n
1 1 ∑ n
er𝜏0 ∑ ∑
n
si 𝜈i ∑n
− Xi 2
− 2 −1 + Xi H i + 1 − 2tanh 2
𝜎 2
𝑗,i (xi (t)) + Δi .
i=2
𝜁i 𝜁i i=2
1 − 𝜏 ∗ 𝑗=1 i=1 ki i=1

Consider the compact set S as follows:


{ ( n ) }
1 ∑ 2 1 2 ∑n
S= s + 𝜓̃ + 2 Vt,i + Xi 2
≤D , (79)
2 i=1 i 𝛾i i i=2

where D > 0 is a known constant. According to the work of Zhang et al,19 since H i , i = 2, … , n is a continuous function,
there exists a positive constant Mi such that |H i | ≤ Mi on the compact set S; thus, we have

| | 1 Mi 2 1
|Xi H i | ≤ Xi 2 2 + 𝜇i 2 , i = 2, … , n, (80)
| | 2 𝜇i 2
where 𝜇i is a positive constant which will be defined later. Substituting (80) into (78) results in
( ) ( )
. 3 1 ∑ n
1 ∑ n
∑n
𝜂i
V n ≤ −s1 𝜈M c1 − −
2 2
− 2
si ci − 1 − −r Vt,i − 𝜓̃ i 𝜓̃ i +
4 2𝜌1 2 i=1
2𝜌 i
2
i=1 i=1
2𝛾 i
( )

n−1
. ( ). ∑n
1 1 M2
+ (gi N (𝜍i ) + 1) 𝜍 i + gn N (𝜍n ) K T Ω + 1 𝜍 n − Xi 2 − 2 − 1 − i2 (81)
i=1 i=2
𝜁i 𝜁i 2𝜇i
[ ( )]
er𝜏0 ∑ ∑ ∑ ∑1
n n n n
si
+ 1 − 2tanh2
𝜎 2
𝑗,i (x i (t)) + Δ i + 𝜇i 2 .
1 − 𝜏 ∗ 𝑗=1 i=1 ki i=1 i=2
2

1 Mi 2 𝜁i
Let ci > 1 + 2𝜌i 2
, i = 2, … , n and 𝜇i 2 > 2(1−𝜁i )
, i = 2, … , n. Then, (81) can be rewritten as

n [ ( )]
. ∑
n
( ). er𝜏0 ∑ ∑
n
si
V n ≤ −CVn + gi′ N (𝜍i ) + 1 𝜍 i + Λ + 1 − 2tanh2
𝜎 2 𝑗,i (xi (t)) , (82)
i=1
1 − 𝜏 ∗ 𝑗=1 i=1 ki
1148 TOORANJIPOUR ET AL.

( ( ) ( ) ( M𝑗 2
))
where C = min 2𝜈M 2 c1 − 34 − 2𝜌1 2 , 2 c𝑗 − 1 − 2𝜌1 2 , r, 𝜂i , 2 𝜁1 − 1
𝜁𝑗 2
−1− 2𝜇𝑗 2
, i = 1, … , n, j = 2, … , n,
∑n ∑n 1 𝑗 𝑗

Λ = i=1 Δi + i=2 12 𝜇i 2 ≥ 0, and gi′ is defined as


{ ( )
gi x i i≠n
gi′ = T
(83)
gn K Ω i = n.

It can be clearly observed from (82) that the last term may be negative or positive, which depends on the size of si .
According to Lemma 3, three cases will be raised as follows.

Case 1. If si ∈ Ωcsi , i = 1, … , n, then |si | < 0.8814ki , with ki being the positive design parameter. Consequently, si is
bounded. Therefore, x1 , 𝜓 1 and 𝜍 1 as a function of s1 are bounded too. Following the similar way qi , Xi , 𝜓 i , and 𝜍 i ,
i = 2, … , n, are also bounded. Therefore, all the signals in the closed-loop system are bounded.

Case 2. If si ∉ Ωcsi , i = 1, … , n, then from Lemma 3 and using the fact that 𝜎 2 j,i (xi (t)) ≥ 0, j = 1, … , n, i = 1, … , n,
we have
n [ ( )]
∑n
∑ si
1 − 2tanh2
𝜎 2 𝑗,i (xi (t)) ≤ 0. (84)
𝑗=1 i=1
ki

By using (84), (82) becomes


. ∑
n
( ).
V n ≤ −CVn + gi′ N (𝜍i ) + 1 𝜍 i + Λ. (85)
i=1

Multiplying both sides of (85) by eCt , then by integrating from that over [0,t], we have
t

n
[ ′( ) ] .
Vn (t) ≤ c0 + e
−𝐶𝑡
gi xi N (𝜍i ) + 1 𝜍 i e𝐶𝜏 d𝜏, (86)
∫ i=1
0
( )
Λ
where c0 = C
+ Vn (0) − ΛC e−𝐶𝑡 . According to Lemma 2, if inequality (86) holds, then Vn (t),
t∑ .
∫0 ni=1 [gi′ (xi )N (𝜍i ) + 1] 𝜍 i e𝐶𝜏 d𝜏, and 𝜍(t) must be bounded. As a result, in Case 2, all the closed-loop signals are
SGUUB.

Case 3. If k of si ∈ Ωcsi , i ∈ {l1 , l2 , … , lk }, while rest of s𝑗 ∉ Ωcs𝑗 , j ∈ {lk+1 , lk+2 , … , ln }. In this case, the whole system
is divided into the two subsystems, for those si ∈ Ωcsi , i ∈ {l1 , l2 , … , lk } and s𝑗 ∉ Ωcs𝑗 , j ∈ {lk+1 , lk+2 , … , ln }. The
first subsystem according to the case 1 can be easily proved that all the closed-loop signals are bounded. According to
the work of Wang et al,49 if the Lyapunov function candidate is constructed for the second subsystem, it can be proved
in a similar way to the Case 2 that all the closed-loop signal in the second subsystem are also bounded and SGUUB.
Now, the second part of Theorem 1 will be proved. Equation (86) can be rewritten as

C 2 . ∑( n
).
s1 ≤ CVn ≤ −V n + gi′ N (𝜍i ) + 1 𝜍 i + Λ. (87)
2 i=1

By integrating both sides of (87), we have


t t
C ∑n
( ′ ).
s d𝜏 ≤ −Vn (t) + Vn (0) +
2
g N (𝜍i ) + 1 𝜍 i d𝜏 + Λt. (88)
2∫ 1 ∫ i=1 i
0 0

t[ ] .
Since Vn (t) and ∫0 gn′ (x (𝜏)) N (𝜍) + 1 𝜍d𝜏 are bounded, one can write
( )
⎛2 t n (
∑ ′ ). ⎞
t ⎜ C −Vn (t) + Vn (0) + ∫ gi N (𝜍i ) + 1 𝜍 i d𝜏 ⎟
1 ⎜ 0 i=1 Λ⎟
lim s1 d𝜏 ≤ lim ⎜
2
+ 2 ⎟. (89)
t→∞ t ∫ t→∞ t C⎟
0

⎜ ⎟
⎝ ⎠
TOORANJIPOUR ET AL. 1149

Rewrite (89) as
t
1 2
lim s21 d𝜏 ≤ Λ. (90)
t t→∞ ∫ C
0

As a result, the tracking error will finally converge to a neighborhood of origin in the mean square sense.

Remark 6. According to (90), it shows that the transfer error s1 depends on 𝜈 M , 𝜂 i , 𝜉 i , Mj , 𝛾 i , and ai , i = 1, … , n,
j = 2, … , n. By reducing ai and increasing 𝜈 M and 𝛾 i , the size of s1 will be reduced. Since 𝜂 i exists at the same time
in the numerator and denominator of (90), reducing or increasing of this parameter needs more attention. Moreover,
𝜂 i cannot be too small since it is possible that the parameter estimations may be not enough to prevent drifting.
Conversely, by increasing 𝜂 i , the control energy may lead the bigger value. Consequently, from implementation point
of view, the design parameters should be tuned to achieve the desired performance behavior.
From the preceding discussion of Cases 1 to 3 and the convergence of tracking error to the neighborhood of zero, it
can be concluded that all the closed-loop signals of the system (1) are bounded and tracking error eventually converge
to the neighborhood of origin, under Assumptions 1 to 5. This completes the proof of Theorem 1. Although this
theorem just shows the bounded value for the closed-loop signals, the quality of response is going to be under control
by using the prescribed functions implemented in the structure of the controller.
{
Remark 7. For any given initial condition compact set Ω0 = si (0) , X𝑗 (0) , 𝜓̃ i (0) , i = 1, … , n, 𝑗 = 2, … , n } , we can
always construct a larger compact set Ω than Ω0 , including Ωcsi , Ci , i = 1, … , n, and Cd in which FIS approximation
is valid and the functions 𝜎 i, j (.) are bounded.

Remark 8. Conservatism issue plays an important role in stability analysis and control design, so setting up some
technical discussions to achieve a rigorous stability analysis seems to be necessary. In some papers such as the work
of Zhou et al,28 by using the fact 0 < SiT (Zi ) Si (Zi ) ≤ 1, the nonlinear membership functions are ignored from final
analysis and synthesis conditions. As a result, these methods bring about conservatism issue. However, in this paper,
the influence of nonlinear membership functions has been considered to obtain less conservative design. One reason
for having conservatism in this paper is applying the DSC technique. This technique not only prevents the “explo-
sion of complexity” but also relaxes some restrictive requirements desired signals and on plant functions. However,
one of the disadvantages of using DSC is the conservative basis of this approach as detailed in the work of Pan and
Yu.50 Although the uncertainties related to unknown parameters added to the conservatism issue in the procedure
of control design, the error converges to the unknown neighborhood of the origin, and the prescribed performance
technique applied in this paper brings about predefined decaying bounds to the output of the system. In other words,
the prescribed performance method reduces the additive influence of conservatism applied in the control design.

4 S I M UL ATION R E SU LT S

In this section, some simulation examples are given to show the effectiveness of our proposed scheme.

4.1 Validation examples


Example 1. In this section, the proposed method is validated and compared through the following SISO nonlinear
system with time-varying delays, a dead-zone nonlinearity, and disturbances terms20
( )
⎧x. 1 = x1 e−0.5x1 + 1 + x1 2 x2 + 2x1 2 (t − 𝜏1 ) + d1 (t)
⎪.
⎨x2 = x1 x2 2 + (3 + cos (x1 x2 )) D (u) + 0.2x2 (t − 𝜏2 ) sin x2 (t − 𝜏2 ) + d2 (t) , (91)

⎩𝑦 = x1 (t) ,

where x1 and x2 denote the state variables. The external disturbances are defined as d1 (t) = 0.1 sint and
d2 (t) = 0.1e−3t , the unknown time-varying delays are considered as 𝜏 1 = 5 and 𝜏 2 = 10, and the desired trajectory is
1150 TOORANJIPOUR ET AL.

yd = 0.5(sint+ sin (0.5t)). The dead-zone nonlinearity D(u(t)) is also defined as


⎧ (1 − 0.3 sin (u)) (u − 0.5) , u > 0.5,

D (u (t)) = ⎨ 0, −0.25 ≤ u ≤ 0.5, (92)

⎩ (0.8 − 0.2 cos (u)) (u + 0.25) , u < −0.25.

The control objective is to design a controller based on DSC strategy and prescribed performance method for the
system (1) using FLS and Nussbaum functions such that all the signals in the closed-loop system are SGUUB and the
tracking error converges to the small residual set. Furthermore, all the error surfaces are strictly evolved within prede-
fined bounds. According to Theorem 1, the adaptive laws and the control input uare chosen as (29)-(30), (42)-(43), and
(45). The design parameters are chosen as a1 = 0.4, a2 = 0.5, 𝛾 1 = 20, 𝛾 2 = 40, 𝜂 1 = 𝜂 2 = 0.1, and 𝜁 2 = 0.01. The performance
function is selected as follows: ( )
h (t) = h0 − h∞ e−𝑛𝑡 + h∞ (93)
with the parameters given in Table 1. The initial conditions are given as x1 (0) = 0, x2 (0) = 0, 𝜍 1 (0) = 1.9, 𝜍 2 (0) = 9, 𝜓̂ 1 (0) =
0.01, and 𝜓̂ 2 (0) = 0.5.
Due to reliability and robustness of Gaussian membership functions, they are used in many articles such as in other
works.9,12,28,51,52 Choosing fuzzy membership function as
( )
−0.5 (x − 2 + l)
𝜇l (x) = exp l = 1, 2, 3. (94)
4
The adjusting parameters related to membership functions are chosen by trial-and-error to achieve the best response. The
tracking performance of the proposed scheme has been depicted in Figure 2. Compared with the tracking result in the
work of Wang and Hu,20 the proposed method has better tracking performance in the transient and steady state response.
As it is obvious from Figure 2, the reference signal has been tracked in less than 0.4 second, in spite of existing uncertain-
ties in the system such as completely unknown control directions, unknown external disturbances, and more importantly,
the coupling of time-varying delays and dead-zone nonlinearity. The corresponding control input, which has been illus-
trated in Figure 3, is bounded and applicable, and compared with control input u,20 the proposed method has a lower
overshoot.

TABLE 1 Performance function parameters


values
h𝟎 h∞ n 𝜹min 𝜹max
Example 1 0.2 0.05 1 0.5 1.1
Example 2 0.3 0.02 2 0.5 1.1

0.8

0.6

0.4

0.2

-0.2

-0.4 0.4

-0.6 0.2

-0.8 0
0 0.2 0.4 0.6 0.8

0 2 4 6 8 10 12 14 16 18 20
Time(sec)

FIGURE 2 System output y and reference signal yd [Colour figure can be viewed at wileyonlinelibrary.com]
TOORANJIPOUR ET AL. 1151

FIGURE 3 The variation of the control input u versus time [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 4 Variation of e(t) and the performance bounds versus time [Colour figure can be viewed at wileyonlinelibrary.com]

As it can be seen in Figure 4, the error e(t) evolves strictly within the prescribed performance bounds.
Example 2. In this example, another comparison study based on the proposed control scheme and the method in
the work of Zhou et al40 is provided. In spite of considering unknown time-varying delays and dead-zone nonlinearity
in the aforementioned work,40 this method suffers from “explosion in terms” due to using conventional backstep-
ping method. Moreover, unknown control directions have not been considered in the aforementioned method. The
following nonlinear nonaffine system with time-varying delays and dead-zone nonlinearity is considered40 :

⎧x. 1 = 1−e−x1 + x2 3 + x2 e−1−x1 2 + 0.5x1 2 (t − 𝜏1 (t)) x2 (t − 𝜏2 (t)) ,


−x

⎪. 1+e 1 ( )
⎨x2 = x1 2 + 0.15D(u) + 0.1 1 + x2 2 D (u) + sin (0.1D (u)) + x1 (t − 𝜏1 (t)) x2 (t − 𝜏2 (t)) ,
3 (95)

⎩𝑦 = x1 (t) ,

where x1 and x2 denote the state variables. The unknown time-varying delays are considered as
𝜏 1 = 𝜏 2 = t − 0.5(1+ sin(t)), and the desired trajectory is yd = 0.5 sin (t)+ sin (0.5t). The dead-zone nonlinearity D(u(t))
is also defined as

⎧ (1 − 0.2 sin (u)) (u − 2.5) , u > 2.5,



D (u (t)) = ⎨ 0, −1.5 ≤ u ≤ 2.5, (96)

⎩ (0.8 − 0.1 cos (u)) (u + 1.5) , u < −1.5.

The initial conditions are given as x1 (0) = 0.2, x2 (0) = 0.5. The control parameters a1 , a2 , 𝛾 1 , 𝛾 2 , 𝜂 1 , 𝜂 2 , 𝜁 2 and fuzzy
membership functions, and their parameters are defined as Example 1, and the performance function parameters
1152 TOORANJIPOUR ET AL.

FIGURE 5 Variation of e(t) and the performance bounds versus time [Colour figure can be viewed at wileyonlinelibrary.com]

are also applied as Table 1. Figure 5 is provided to compare the proposed method with the work of Zhou et al.40 As
observed from this Figure, the error obtained by the proposed method completely satisfies the prescribed bounds, but
the error of method proposed by the aforementioned work40 violates the constraints as shown in Figure 5.

Remark 9. The selection of control parameters plays an important role in the resulting behavior of the system. On
the other hand, according to Examples 1 and 2, by getting the same control parameters, the resulting behavior is
completely satisfactory. Therefore, these two examples show a robustness behavior to the system change.

4.2 Numerical example


Consider the following nonaffine SISO nonlinear system:
. −x1
⎧x1 = 1−e
1+e−x1
+ (2 + sin x1 ) x2 + 0.5x1 2 (t − 𝜏1 (t)) cos x1 (t − 𝜏1 (t)) + d1 (t) ,
⎪. D(u)
⎨x2 = x1 x2 + (2 + sin (x1 x2 )) √|D(u)|+0.1 + x1 (t − 𝜏1 (t)) x2 (t − 𝜏2 (t)) + d2 (t) ,
2
(97)

⎩𝑦 = x1 (t) ,
where x1 and x2 denote the state variables. The external disturbances are defined as d1 (t) = 0.5 cos t and d2 (t) = 0.2 sin t, the
unknown time-varying delays are considered as 𝜏 1 (t) = 0.2(1+ cos t) and 𝜏 2 (t) = 0.3(1+ sin t), and the desired trajectory is
yd = sin t+ cos (0.5t). The dead-zone nonlinearity D(u(t)) is also defined as (92). The control objective is the same as the
previous example. The design parameters are chosen as a1 = 0.5, a2 = 0.5, 𝛾 1 = 20, 𝛾 2 = 40, 𝜂 1 = 𝜂 2 = 0.1, and 𝜁 2 = 0.01.
The performance function are selected as (93) with the parameters given in Table 2.
The initial conditions are given as x1 (0) = 0.5, x2 (0) = −0.2, 𝜍 1 (0) = 1.9, 𝜍 2 (0) = 10, 𝜓̂ 1 (0) = 0.01, and 𝜓̂ 2 (0) = 0.5.
Fuzzy membership functions are chosen as (94). In Figure 6, the system output tracks the reference signal.
The control input u is illustrated in Figure 7.
Figure 8 depicts the error and the prescribed bounds, which shows the effectiveness of the proposed method.
The variations of adaptive FLS parameters 𝜓 1 and 𝜓 2 have been shown in Figure 9. In Figure 10, the variation of
Nussbaum functions and their adaptive gains have been depicted.
As it has been expected, all the closed-loop signals which have been depicted in Figures 6 to 10 are SGUUB.

TABLE 2 Performance
function parameters values
h𝟎 h∞ n 𝜹min 𝜹max
1 0.1 1 0.6 1.1
TOORANJIPOUR ET AL. 1153

1.5 x
1.5
y

1 1

0.5
0.5 0 0.2 0.4 0.6 0.8

-0.5

-1

-1.5

0 2 4 6 8 10 12 14 16 18 20
Time(sec)

FIGURE 6 System output y and reference signal yd [Colour figure can be viewed at wileyonlinelibrary.com]

100
50

50 0
-50
control input

0 0.5
0

-50

-100
0 5 10 15 20
Time(sec)

FIGURE 7 The variation of control input u(t) versus time [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 8 Variation of e(t) and the performance bounds versus time [Colour figure can be viewed at wileyonlinelibrary.com]
1154 TOORANJIPOUR ET AL.

2.5

1.5

0.5

0
0 5 10 15 20
Time(sec)

FIGURE 9 The adaptive parameters of 𝜓 1 and 𝜓 2 [Colour figure can be viewed at wileyonlinelibrary.com]

20

1
0
N
1
-20 2
N
2
-40

-60

-80

-100
0 5 10 15 20
Time(sec)

FIGURE 10 The adaptive parameters 𝜍 1 and 𝜍 2 and the corresponding Nussbaum gain N 1 and N 2 [Colour figure can be viewed at
wileyonlinelibrary.com]

5 CO N C LUSION S

In this paper, an adaptive prescribed performance fuzzy backstepping controller has been designed for a class of SISO
nonaffine nonlinear systems with unknown time-varying delays and dead-zone input nonlinearity. In order to ease the
complexity in terms caused by the derivatives of the virtual controller in each step, the DSC strategy is applied by intro-
ducing a first-order filter passed through the virtual controller. Unknown nonlinear functions existing in the system are
approximated by FLS, and the prescribed performance is achieved by using the proper performance functions. Nuss-
baum functions are used in this control scheme to relax the problem of unknown control gains. By constructing proper
Lyapunov-Krasovskii functions, the effect of existing time-varying delays in the system is addressed. It is proved that all
the closed-loop signals are SGUUB and the transformed error s1 converges to the neighborhood of the origin with the
prescribed decaying bounds. Finally, the simulation results demonstrate the superiority of the proposed method over
the existing approaches to tackle the unknown time-varying delays, dead-zone input nonlinearity, and unknown control
directions in the nonaffine system (1). In the future, we will develop this method for an observer-based controller for SISO
systems with unknown dead-zone input and time-varying delays.

ORCID

Pouria Tooranjipour https://orcid.org/0000-0002-4238-4730


Ramin Vatankhah https://orcid.org/0000-0002-5529-5454
Mohammad Mehdi Arefi https://orcid.org/0000-0003-3986-8205
TOORANJIPOUR ET AL. 1155

REFERENCES
1. Zinober A, Owens D. Nonlinear and Adaptive Control. Vol. 281. Berlin, Germany: Springer-Verlag Berlin Heidelberg; 2003. https://doi.
org/10.1007/3-540-45802-6
2. Kanellakopoulos I, Kokotovic PV, Morse AS. Systematic design of adaptive controller for feedback linearizable systems. IEEE Trans Autom
Control. 1991;36(1):1241-1253.
3. Kanellakopoulos I. Adaptive Control of Nonlinear Systems (PhD Diss). University of Illinois. 1992. https://www.ideals.illinois.edu/handle/
2142/23411
4. Swaroop D, Hedrick JK, Yip PP, Gerdes JC. Dynamic surface control for a class of nonlinear systems. IEEE Trans Autom Control.
2000;45(10):1893-1899.
5. Yip PP, Hedrick JK. Adaptive dynamic surface control: a simplified algorithm for adaptive backstepping control of nonlinear systems. Int
J Control. 1998;71(5):959-979. https://doi.org/10.1080/002071798221650
6. Liu Y-H. Adaptive dynamic surface asymptotic tracking for a class of uncertain nonlinear systems. Int J Robust Nonlinear Control.
2018;28(4):1233-1245. https://doi.org/10.1002/rnc.3947
7. Alessandri A. Adaptive neural network control of robotic manipulators. Automatica. 2004;40(11):2011-2012. https://doi.org/10.1016/j.
automatica.2004.05.014
8. Wang Y, Chai T, Zhang Y. State observer-based adaptive fuzzy output-feedback control for a class of uncertain nonlinear systems.
Information Sciences. 2010;180(24):5029-5040.
9. Zhou S, Feng G, Feng C-B. Robust control for a class of uncertain nonlinear systems: adaptive fuzzy approach based on backstepping.
Fuzzy Set Syst. 2005;151(1):1-20. https://doi.org/10.1016/j.fss.2004.05.008
10. Wang L-X. A Course in Fuzzy Systems. Upper Saddle River, NJ: Prentice Hall PTR; 1997.
11. Arefi MM, Ramezani Z, Jahed-Motlagh MR. Observer-based adaptive robust control of nonlinear nonaffine systems with unknown gain
sign. Nonlinear Dynamics. 2014;78(3):2185-2194. https://doi.org/10.1007/s11071-014-1573-0
12. Tong S, Liu C, Li Y. Robust adaptive fuzzy filters output feedback control of strict-feedback nonlinear systems. Int J Appl Math Comput
Sci. 2010;20(4):637-653. https://doi.org/10.2478/v10006-010-0047-x
13. Chen W, Jiao L, Li R, Li J. Adaptive backstepping fuzzy control for nonlinearly parameterized systems with periodic disturbances. IEEE
Trans Fuzzy Syst. 2010;18(4):674-685. https://doi.org/10.1109/TFUZZ.2010.2046329
14. Arefi MM, Zarei J, Karimi HR. Adaptive output feedback neural network control of uncertain non-affine systems with unknown control
direction. J Franklin Inst. 2014;351(8):4302-4316. https://doi.org/10.1016/j.jfranklin.2014.05.006
15. Park J-H, Kim S-H, Moon C-J. Adaptive fuzzy controller for the nonlinear system with unknown sign of the input gain. Int J Control Autom
Syst. 2006;4(2):178-186.
16. Willems JC, Byrnes GI. Global adaptive stabilization in the absence of information on the sign of the high frequency gain. In: Analysis
and Optimization of Systems. Berlin, Germany: Springer; 1984:49-57. https://doi.org/10.1007/BFb0004944
17. Mårtensson B. Remarks on adaptive stabilization of first order non-linear systems. Syst Control Lett. 1990;14(1):1-7. https://doi.org/10.
1016/0167-6911(90)90073-4
18. Chen W, Zhang Z. Globally stable adaptive backstepping fuzzy control for output-feedback systems with unknown high-frequency gain
sign. Fuzzy Set Syst. 2010;161(6):821-836. https://doi.org/10.1016/j.fss.2009.10.026
19. Zhang L, Tong S, Li Y. Dynamic surface error constrained adaptive fuzzy output-feedback control of uncertain nonlinear systems with
unmodeled dynamics. Neurocomputing. 2014;143:123-133. https://doi.org/10.1016/j.neucom.2014.06.013
20. Wang J, Hu J. Robust adaptive neural control for a class of uncertain non-linear time-delay systems with unknown dead-zone non-linearity.
IET Control Theory Appl. 2011;5(15):1782-1795. https://doi.org/10.1049/iet-cta.2010.0633
21. Tong S, Li Y. Adaptive fuzzy output feedback control of MIMO nonlinear systems with unknown dead-zone inputs. Fuzzy Syst IEEE Trans.
2013;21(1):134-146.
22. Tao G, Kokotovic PV. Adaptive control of plants with unknown dead-zones. IEEE Trans Autom Control. 1994;39(1):59-68. https://doi.org/
10.1109/9.273339
23. Zhou J, Wen C, Zhang Y. Adaptive output control of nonlinear systems with uncertain dead-zone nonlinearity. IEEE Trans Autom Control.
2006;51(3):504-511. https://doi.org/10.1109/TAC.2005.864200
24. Hua C, Zhang L, Guan X. Distributed adaptive neural network output tracking of leader-following high-order stochastic nonlinear
multiagent systems with unknown dead-zone input. IEEE Trans Cybern. 2017;47(1):177-185. https://doi.org/10.1109/TCYB.2015.2509482
25. Wu J, Su B, Li J, Zhang X, Li X, Chen W. Adaptive fuzzy control for full states constrained systems with nonstrict-feedback form and
unknown nonlinear dead zone. Information Sciences. 2017;376:233-247. https://doi.org/10.1016/j.ins.2016.10.016
26. Zhang Z, Duan G, Hou M. Global finite time stabilization of pure-feedback systems with input dead-zone nonlinearity. J Franklin Inst.
2017;354(10):4073-4101. https://doi.org/10.1016/J.JFRANKLIN.2017.01.039
27. Edalati L, Khaki Sedigh A, Aliyari Shooredeli M, Moarefianpour A. Adaptive fuzzy dynamic surface control of nonlinear systems with
input saturation and time-varying output constraints. Mech Syst Signal Process. 2018;100:311-329. https://doi.org/10.1016/j.ymssp.2017.
07.036
28. Zhou W-D, Liao C-Y, Zheng L. Adaptive backstepping control for a class of uncertain nonaffine nonlinear time-varying delay systems with
unknown dead-zone nonlinearity. Abstr Appl Anal. 2014. https://doi.org/10.1155/2014/758176
29. Khajeh Talkhoncheh M, Shahrokhi M, Askari MR. Observer-based adaptive neural network controller for uncertain nonlinear systems
with unknown control directions subject to input time delay and saturation. Information Sciences. 2017;418–419:717-737. https://doi.org/
10.1016/j.ins.2017.08.024
1156 TOORANJIPOUR ET AL.

30. Wu Z, Wu Y, Wu Z-G, Lu J. Event-based synchronization of heterogeneous complex networks subject to transmission delays. IEEE Trans
Syst Man Cybern Syst. 2018;48(12):2126-2134. https://doi.org/10.1109/TSMC.2017.2723760
31. Que H, Fang M, Wu Z-G, Su H, Huang T, Zhang D. Exponential synchronization via aperiodic sampling of complex delayed networks.
IEEE Trans Syst Man Cybern Syst. 2018;1-9. https://doi.org/10.1109/TSMC.2018.2858247
32. Wu Z-G, Shi P, Su H, Chu J. Stochastic synchronization of Markovian jump neural networks with time-varying delay using sampled data.
IEEE Trans Cybern. 2013;43(6):1796-1806. https://doi.org/10.1109/TSMCB.2012.2230441
33. Shen Y, Wu Z-G, Shi P, Shu Z, Karimi HR. H∞ control of Markov jump time-delay systems under asynchronous controller and quantizer.
Automatica. 2019;99:352-360. https://doi.org/10.1016/J.AUTOMATICA.2018.10.056
34. Ramezani Z, Arefi MM, Zargarzadeh H, Jahed-Motlagh MR. Neuro-adaptive backstepping control of SISO non-affine systems with
unknown gain sign. ISA Trans. 2016;65:199-209. https://doi.org/10.1016/j.isatra.2016.08.024
35. Boulkroune A, M'Saad M, Farza M. Fuzzy approximation-based indirect adaptive controller for multi-input multi-output non-affine
systems with unknown control direction. IET Control Theory Appl. 2012;6(17):2619-2629. https://doi.org/10.1049/iet-cta.2012.0565
36. Bechlioulis CP, Rovithakis GA. Robust adaptive control of feedback linearizable MIMO nonlinear systems with prescribed performance.
IEEE Trans Autom Control. 2008;53(9):2090-2099. https://doi.org/10.1109/TAC.2008.929402
37. Tee KP, Ge SS, Tay EH. Barrier Lyapunov functions for the control of output-constrained nonlinear systems. Automatica.
2009;45(4):918-927.
38. Han S-I, Lee J-M. Recurrent fuzzy neural network backstepping control for the prescribed output tracking performance of nonlinear
dynamic systems. ISA Transactions. 2014;53(1):33-43. https://doi.org/10.1016/j.isatra.2013.08.012
39. Chen L. Asymmetric prescribed performance-barrier Lyapunov function for the adaptive dynamic surface control of unknown
pure-feedback nonlinear switched systems with output constraints. Int J Adapt Control Signal Process. 2018;32(10):1417-1439. https://doi.
org/10.1002/acs.2921
40. Zhou Q, Wang L, Wu C, Li H. Adaptive fuzzy tracking control for a class of pure-feedback nonlinear systems with time-varying delay and
unknown dead zone. Fuzzy Set Syst. 2017;329:36-60. https://doi.org/10.1016/J.FSS.2016.11.005
41. Zhang Z, Xu S, Zhang B. Exact tracking control of nonlinear systems with time delays and dead-zone input. Automatica. 2015;52:272-276.
42. Na J. Adaptive prescribed performance control of nonlinear systems with unknown dead zone. Int J Adapt Control Signal Process.
2013;27(5):426-446. https://doi.org/10.1002/acs.2322
43. Zhao Q, Li Y, Lin Y. Adaptive nonlinear output-feedback dynamic surface control with unknown high-frequency gain sign. Int J Control.
2013;86(12):2203-2214. https://doi.org/10.1080/00207179.2013.806822
44. Tooranjipour P, Vatankhah R, Mehdi Arefi M, Montazeri A. Adaptive neural prescribed performance DSC for non-affine SISO nonlinear
systems with external disturbances. IFAC-PapersOnLine. 2018;51(15):748-753. https://doi.org/10.1016/j.ifacol.2018.09.165
45. Apostol TM. Mathematical Analysis. 2nd ed. Reading, MA: Addison-Wesley; 1974.
46. Nussbaum RD. Some remarks on a conjecture in parameter adaptive control. Syst Control Lett. 1983;3(5):243-246.
47. Tong S, Liu C, Li Y. Fuzzy-adaptive decentralized output-feedback control for large-scale nonlinear systems with dynamical uncertainties.
IEEE Trans Fuzzy Syst. 2010;18(5):845-861. https://doi.org/10.1109/TFUZZ.2010.2050326
48. Ge SS, Tee KP. Approximation-based control of nonlinear MIMO time-delay systems. Automatica. 2007;43(1):31-43.
49. Wang M, Liu X, Shi P. Adaptive neural control of pure-feedback nonlinear time-delay systems via dynamic surface technique. Syst Man,
Cybern Part B Cybern IEEE Trans. 2011;41(6):1681-1692. https://doi.org/10.1109/TSMCB.2011.2159111
50. Pan Y, Yu H. Dynamic surface control via singular perturbation analysis. Automatica. 2015;57:29-33. https://doi.org/10.1016/J.
AUTOMATICA.2015.03.033
51. Li Y, Tong S, Li T. Adaptive fuzzy output feedback dynamic surface control of interconnected pure-feedback nonlinear. IEEE Trans Cybern.
2015;45(1):138-149. https://doi.org/10.1109/TCYB.2014.2333738
52. Tong S, He X, Li Y, Zang H. Adaptive fuzzy backstepping robust control for uncertain nonlinear systems based on small-gain approach.
Fuzzy Set Syst. 2010;161:771-796.

How to cite this article: Tooranjipour P, Vatankhah R, Arefi MM. Prescribed performance adaptive fuzzy
dynamic surface control of nonaffine time-varying delayed systems with unknown control directions and dead-zone
input. Int J Adapt Control Signal Process. 2019;33:1134–1156. https://doi.org/10.1002/acs.3014

Вам также может понравиться