You are on page 1of 13

Received: 1 March 2019 Revised: 28 May 2019 Accepted: 13 August 2019

DOI: 10.1002/asjc.2253

REGULAR PAPER

Adaptive robust dynamic surface control for uncertain


strict‐feedback nonlinear systems using fuzzy logic systems

Sanyan Chen1 | Yuchao Wang1 | Wenrui Shi1 | Hansheng Wu2

1
College of Mechanical and Electrical
Abstract
Engineering, Sichuan Agricultural
University, China In this paper, an adaptive robust stabilization problem is dealt with for a class
2
Department of Information Science, of uncertain strict‐feedback nonlinear systems in the presence of unknown
Prefectural University of Hiroshima,
structure uncertainties, external disturbances, and unknown time‐varying vir-
Japan
tual control coefficients. It is not required to know the upper bounds of external
Correspondence disturbances, as well as the upper and lower bounds of unknown time‐varying
Yuchao Wang, College of Mechanical and
virtual control coefficients. The controller is designed by adopting
Electrical Engineering, Sichuan
Agricultural University, Yaan 625014, backstepping. In addition, to avoid suffering from the problem of ‘explosion
China. of terms’, a dynamic surface control approach is employed by introducing the
Email: wangyc0918@yahoo.co.jp
first‐order low‐pass filter. Furthermore, at every step of the backstepping design
Funding information procedure, fuzzy logic systems are used to approximate the unknown structure
Education Department of Sichuan Prov- uncertainties. In particular, the norms of weight matrices and the upper bounds
ince, Grant/Award Number: 18ZB0456
of approximation errors of fuzzy logic systems are supposed to be unknown. It
is also shown that the proposed controller can guarantee the uniform bounded-
ness of uncertain strict‐feedback nonlinear systems. Finally, the simulation for
a single‐link manipulator actuated by a brush DC motor is carried out to illus-
trate the validity of the proposed controller.

K E YWO R D S
adaptive robust control, backstepping design, dynamic surface control, fuzzy logic systems

1 | I N T RO D U C T I O N stability of uncertain nonlinear systems. In [10], an adap-


tive robust state feedback controller has been developed
It is generally known that there always exist uncertainties to make the closed‐loop system globally uniformly expo-
in engineering systems owing to modeling errors, mea- nentially convergent to a sphere. Nevertheless, a common
surement errors, changes of external environment condi- assumption of the above methods is that the systems are
tions, and so on [1–5]. Therefore, in the literature, required to satisfy the matched uncertainties. As a break-
extensive efforts have been devoted to the problem of through in the field of nonlinear control, a backstepping
robust stabilization for dynamical systems with signifi- design is proposed to obtain adaptive robust stabilization
cant uncertainties. Many robust state (or output) feedback or model asymptotical tracking for dynamic systems that
controllers have been developed (see, e.g. [6–8] and the can be transformable to uncertain strict‐feedback nonlin-
references therein) in order to generate satisfactory per- ear systems [11,12]. It is a powerful tool that can relax the
formance of control systems. In [9], a class of adaptive requirement of matched conditions. Therefore, in recent
robust controllers has been presented to guarantee the years, there have been many studies into the
---------------------------------------------------------------------------------------------------------------------------------
© 2019 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd

Asian J Control. 2021;23:761–773. wileyonlinelibrary.com/journal/asjc 761


762 CHEN ET AL.

backstepping design approach [13–19]. Nevertheless, it is used for the modeling and controller design for uncertain
noted that the virtual control coefficients in [13–19] are nonlinear systems [48–53]. However, the upper bounds of
set equal to one. In reality, it is impossible to decide vir- the norms of ideal weight vectors are needed in [48], and
tual control coefficients in many cases [20–27]. Unfortu- prior knowledge of the approximation errors is required
nately, both upper bounds and lower bounds of the in [49]. Moreover, in [48–53], the elements of the ideal
unknown time‐varying virtual control coefficients are weight vector are updated online one by one. Obviously,
supposed to be known [20–22], while the unknown this will produce a certain amount of computational
time‐varying virtual control coefficients are assumed to burden.
be upper bounded by the known smooth functions in In this paper, motivated by the previous works, the
[23] and [24]. Although the upper bounds of the adaptive robust stabilization problem is developed for a
unknown time‐varying virtual control coefficients are class of uncertain strict‐feedback nonlinear systems in
unknown, it is assumed that the unknown time‐varying the presence of unknown structure uncertainties, external
virtual control coefficients have known lower bounds disturbances, and unknown time‐varying virtual control
[25–27]. coefficients. Fuzzy logic systems are utilized to approxi-
In traditional backstepping the problem of ‘explosion of mate the unknown structure uncertainties. Meanwhile,
terms’ is a significant disadvantage, that is, the differentia- the adaptation laws are employed to estimate the norms
tion of the virtual controllers is included in the process of of weight matrices, the upper bounds of approximation
controller design. As a result, with the increasing of the errors of fuzzy logic systems, the upper bounds of external
order of the systems, the complexity of the controller rap- disturbances, as well as the upper and lower bounds of
idly increases. In the last two decade, the dynamic surface unknown time‐varying virtual control coefficients. Based
control technique (see, e.g., [28–37] and the references on the dynamic surface control technique and
therein) that utilizes a first‐order low‐pass filter at every backstepping manner, a class of adaptive robust control-
step of the virtual controller design procedure is adopted. lers is synthesized. It is confirmed that the proposed con-
The dynamic surface control technique was first put for- trol scheme can guarantee the uniform boundedness of
ward by Swaroop [38]. The important merits of the uncertain strict‐feedback nonlinear systems. The contri-
dynamic surface control technique include the following. butions of this paper include the following. 1) By compar-
1) The control approach associated with the dynamic sur- ison with the existing literature, such as [20–27], the
face control technique is a solution to the issue of ‘explo- upper and lower bounds of virtual control coefficients
sion of terms’ caused by the repeated differentiation of can be supposed to be unknown. Accordingly, this paper
the virtual controllers. 2) The requirements of the nonlin- removes the restricted assumption that the upper bounds
ear functions in the virtual controllers are released. The or the lower bounds of unknown time‐varying virtual
nonlinear functions employed in the virtual controllers control coefficients are known. 2) The differentiations of
only require to be first order derivable functions, which the virtual controllers are not included in the process of
are irrelevant to the order of systems. 3) The dynamic sur- controller design, and the issue of ‘explosion of terms’ in
face control technique has been applied to several practical the traditional manner of backstepping is avoided. 3) In
nonlinear control systems [39–42], such as autonomous the controller design, fuzzy logic systems are used to
surface vehicles, permanent magnet synchronous motor, model the unknown nonlinear structure uncertainties,
underactuated autonomous underwater vehicles, inte- and the prior knowledge about the weight matrices and
grated missile guidance and autopilot, and so on. Hence, approximation errors of the fuzzy logic systems is not
this technique can be utilized to design the controllers in required. Furthermore, the norm of the ideal weight vec-
this paper. By contrast, in [30] and [34], the time‐varying tor in the fuzzy logic systems is taken as the estimation
virtual control coefficients are known functions; in [31], parameter, not as the elements of the weight vector. For
[33], and [35], neural networks are used to estimate the this reason, compared with [48–53], the computational
terms including negative first power of control coefficients, load is reduced in this paper.
and so the control coefficients are required to be smooth This paper is organized as follows. The preliminaries,
functions; in [32], the negative first power of the deriva- problem formulation, and standard assumptions are
tives of control coefficients are needed in the proposed introduced in Section 2. A class of adaptive robust
controllers. To sum up, they all have strict requirements dynamic surface controllers is proposed and the corre-
on the control coefficients. sponding stability analysis is carried out in Section 3.
On the other hand, it is well known that fuzzy logic The application of the proposed controller to the problem
systems have been extensively exploited for nonlinear of stabilization for a single‐link manipulator actuated by a
systems due to their universal approximation properties brush DC motor is presented in Section 4. Finally, conclu-
[43–47]. As such, fuzzy logic systems have been widely sions are made in Section 5.
CHEN ET AL. 763

2 | P RO BL EM FO RM U L A TI O N A N D 8
>
> g2i
P R E L I MI N A R I E S >
> ; i ¼ 1;
>
> 2g2i
>
>
>
< g2
2.1 | Nonlinear control problem * 1
ϕi : ¼ i
þ 2; 2 ≤ i ≤ n − 1;
>
> 2g 2 2g
>
> i i
Consider the following uncertain strict‐feedback nonlin- >
>
>
> 1
ear system: : 2g2 ;
> i ¼ n::
i
x_ 1 ¼ g1 ð~
x 1 Þx 2 þ f 1 ð~
x 1 Þ þ d1 ðtÞ (3b)
x_ 2 ¼ g2 ð~
x 2 Þx 3 þ f 2 ð~
x 2 Þ þ d2 ðtÞ
where ω*i are the optimal fuzzy parameter
⋮ ⋮
vectors and εi are the upper bounds of
x_ n−1 ¼ gn−1 ð~
x n−1 Þx n þ f n−1 ð~
x n−1 Þ þ dn−1 ðtÞ approximation errors. Here, it is obvious that
x_ n ¼ gn ð~
x n ÞuðtÞ þ f n ð~
x n Þ þ dn ðtÞ the parameters ψ*i and ϕ*i are still unknown
(1) positive constants.

where x~i ðtÞ ¼ ½x 1 ; x 2 ; … ; x i ⊤ ∈ Ri ; i ¼ 1; 2; … ; n are Remark 1. Actually, the system (1) describes
the system state vectors, u(t)∈R is the control input, gi the control problems of power systems,
(·), f i(·) and di(·),i=1,2,…,n are all unknown functions, motors, and so on [1,2]. The dynamic charac-
where gi(·) represent time‐varying virtual control coeffi- teristics of the system, as well as the distur-
cients, f i(·) represent structure uncertainties that are bances caused by changes of external
assumed to be continuous and differentiable, and di(·) environment conditions, are often uncertain
represents external disturbances assumed to be continu- and time‐dependent. Therefore, the control
ous and bounded. coefficients and disturbances of the system
Control target: Utilizing a dynamic surface control are time‐varying and bounded. Here, it is
technique and fuzzy logic systems to synthesize a state not required to know the upper bounds of
feedback controller such that the state x~n ðtÞ involved in the disturbances. Meanwhile, the upper and
nonlinear system (1) is uniformly bounded. lower bounds of virtual control coefficients
In order to make the control scheme feasible, some can be allowed to be unknown, and gi(·)
assumptions are put forward for nonlinear system (1). away from zero is the controllable condition
Assumption 1. The time‐varying virtual of system (1). Hence, Assumptions 1 and 2
control coefficients gi(·),i=1,2,…,n are contin- are reasonable.
uous and bounded with known sign. Without
loss of generality, it is assumed that gi(·)>0. 2.2 | Description of fuzzy logic systems
In addition, for any i∈{1,2,…,n}, there exist
the unknown positive constants g iand gi Fuzzy logic control is a kind of nonlinear control based
such that 0 < gi ≤ gi ð·Þ ≤ gi < ∞:Here, on fuzzy set theory, fuzzy linguistic variables and fuzzy
from this assumption, it is obvious that logic reasoning. The experts' control experience and
knowledge are described as control rules through linguis-
gi ð·Þ tic variables, and then these rules are used to control the
≥ 1; i ¼ 1; 2; … ; n (2)
gi systems. In this paper, fuzzy logic systems are imple-
mented to estimate the nonlinear structure uncertainties
Assumption 2. For any i∈{1,2,…,n}, there f i(·) in the controller design procedure. The knowledge
 satisfy
exists the unknown positive constant di base for fuzzy logic systems is constructed from fuzzy
the following inequality IF–THEN rules as follows.
 i < ∞;
jdi ðtÞj ≤ d i ¼ 1; 2; … ; n
Rl : If x 1 is F l1 and x 2 is F l2 and … and x n is F ln ;
Without loss of generality, some definitions
Then y is Gl ; l ¼ 1; 2; … ; N:
are also given.
!
* 1 1 * 2  ; i ¼ 1; 2; … ; n
ψi : ¼ ‖ω ‖ þ εi þ d where x(t)=[x1,x2,…,xn]⊤ is the input, y is the output, N is
gi 4gi i i
the rules number. F li and Gl are fuzzy sets associated with
(3a)
the fuzzy functions μF li ðx i Þ and μGl ðyÞ, respectively.
764 CHEN ET AL.

The fuzzy logic systems with center‐average where k1 and σ1 are any given positive constants and π1(t)
defuzzifizer, product inference, and singleton fuzzifizer b ðtÞ þ ‖ρ ð~
is defined as π 1 ðtÞ: ¼ ψ x 1 Þ‖2 and ρ ð~
x 1 Þ will be
1 1 1
are in the following form: b ðtÞ and ϕ
given later. Furthermore, ψ b ðtÞ are the estimates
1 1
  *
of the unknown constants ψ1 and ϕ*1 , and they are
N n
∑l¼1y l ∏i¼1 μF li ðx i Þ
  updated through the following adaptation laws
yðxÞ ¼ (4)
N n
∑l¼1 ∏i¼1 μF li ðx i Þ
b ðtÞ
dψ 1 b ðtÞ þ γ jS1 j
¼ −γ 11 σ 1 ψ (8a)
where y l ¼ maxy∈R μGl ðyÞ. dt 1 11

Define the fuzzy basis functions as


b ðtÞ

n
1 b ðtÞ þ γ 12 jS1 j2
¼ −γ 12 σ 1 ϕ (8b)
∏i¼1 μF li ðx i Þ dt 1
ρl ðxÞ ¼  
N n
∑l¼1 ∏i¼1 μF li ðx i Þ where γ11 and γ12 are any given positive constants.
Moreover, letting ψ~ 1 ðtÞ ¼ ψb 1 ðtÞ−ψ*1 and ϕ~ 1 ðtÞ ¼ ϕb 1 ðtÞ−ϕ*1 ,
Denoting ω⊤ ¼ ½y1 ; y2 ; … ; yN  ¼ ½ω1 ; ω2 ; … ; ωN ; it is held that
ρðxÞ ¼ ½ρ1 ðxÞ; ρ2 ðxÞ; … ; ρN ðxÞ⊤ , then equation (4) can
be rewritten into the following form: ~ ðtÞ
dψ 1 ~ ðtÞ þ γ jS1 j−γ σ 1 ψ*
¼ −γ 11 σ 1 ψ (9a)
1 11 11 1
dt

yðxÞ ¼ ω ρðxÞ (5)
~ ðtÞ
dϕ 1 ~ 1 ðtÞ þ γ 12 jS1 j2 − γ 12 σ 1 ϕ*
¼ −γ 12 σ 1 ϕ (9b)
Lemma 1. [53]. Let f (x) be a continuous dt 1

function defined on a compact set Ω. Then


for any constant ϵ>0, there exists a fuzzy logic Inspired by the dynamic surface control technique, a
system (5) such as supx∈Ω j f ðxÞ−ω⊤ ρðxÞj ≤ ϵ: new variable ξ2d is introduced. The estimated value ξ2d
Based on Lemma 1, given a compact set x∈Ωx, is obtained by passing ξ2 through a first‐order low‐pass
we know that the smooth function f (x) can be filter with time constant τ2. In this way, we can avoid
expressed as direct differentiation of ξ2 . The equation of the first‐order
low‐pass filter is as follows.
f ðxÞ ¼ ω*⊤ ρðxÞ þ εðxÞ (6)

where ω* is an optimal fuzzy parameter vector, τ 2 ξ_ 2d þ ξ 2d ¼ ξ 2 ; ξ 2d ð0Þ ¼ ξ 2 ð0Þ (10)


and the approximation error ε(x) satisfies
‖εðxÞ‖ ≤ ε with the constant ε > 0. Then, we define the errors S2 and y2 as

S2 ¼ x 2 − ξ 2d (11a)
3 | MAIN RESULTS
y2 ¼ ξ 2d − ξ 2 ¼ −τ 2 ξ_ 2d (11b)
In this section, we design the adaptive robust controller
using dynamic surface control technique and fuzzy logic
Then, S_ 1 can be expressed as
systems in a step‐by‐step manner for the uncertain non-
linear system (1). The design process consists of n steps.
The virtual controllers ξi,i=2,3,…,n are designed at the S_ 1 ¼ g1 ð~
x 1 ÞS2 þ g1 ð~
x 1 Þy2 þ g1 ð~
x 1 Þξ 2 þ f 1 ð~
x 1 Þ þ d1 ðtÞ
previous n−1 steps, and then ξid,i=2,3,…,n are obtained (12)
by using ξi,i=2,3,…,n as the inputs through a first‐order
low‐pass filter. Finally, the actual controller u(t) is intro- For any i∈{1,2,…,n}, according to (6), we know that for
duced at step n. an arbitrarily small positive number εi , there exists an
Step 1: optimal fuzzy parameter vector ω*i such that f i ð~ x iÞ ¼
*⊤
We define S1=x1. Then, the time derivative of S1 can be ωi ρi ð~
x i Þ þ εi ð~
x i Þ, and the approximation error εi ð~
x iÞ
expressed as S_ 1 ¼ g1 ð~
x 1 Þx 2 þ f 1 ð~
x 1 Þ þ d1 ðtÞ: satisfies the inequality ‖εi ð~ x i Þ‖ ≤ εi .
Therefore, S_ 1 becomes
Choose the virtual controller as
S_ 1 ¼ g1 ð~
x 1 ÞS2 þ g1 ð~
x 1 Þy2 þ g1 ð~
x 1 Þξ 2
1 b ðtÞS1 − π 21 ðtÞS1 (13)
ξ 2 ¼ − k 1 S1 − ϕ (7) þ ω*⊤
π 1 ðtÞjS1 j þ σ 1 1 ρ1 ð~
x 1 Þ þ ε1 ð~
x 1 Þ þ d1 ðtÞ
1
2
CHEN ET AL. 765

From (11), it is derived that By using (3b), (7), and (9), it is derived that

y2 _
y_ 2 ¼− − ξ2 g ð~x 1Þ 1
τ2 V_ 1 ðtÞ ≤ ϕ*1 S21 þ S22 þ y22 þ 1 S1 − k1 S1
g1 2
y ∂ξ ∂ξ b_ ∂ξ 2 b_ (14)

¼ − 2 − 2 S_ 1 − 2 ϕ 1− ψ b ðtÞS1 − π 1 ðtÞS1
2
τ 2 ∂S1 b
∂ϕ 1 b 1
∂ψ −ϕ
1 1
π 1 ðtÞjS1 j þ σ 1
(19)
~ ðtÞ þ ψ* þ ‖ρ ð~
þ ðψ 1 x 1 Þ‖ ÞjS1 j
2
It is clear that there exists the continuous function η2(·) 1 1

such that − σ1ϕ ~ ðtÞS2 − σ 1 ϕ


~ 2 ðtÞ þ ϕ ~ ðtÞϕ*
1 1 1 1 1
y2 b Þ ≤ 
b1 ; ψ ~ 2 ðtÞ−σ 1 ψ
~ ðtÞψ*
jy_ 2 þ j ≤ η2 ðS1 ; S2 ; y2 ; ϕ η2 − σ1 ψ 1 1 1
τ2 1

From (2), we can readily conclude that


where η2 is the maximum of |η2(·)| on compact region
ΩS1 × ΩS2 × Ωy2 × Ωb × Ωb . Here, ΩS1 ; ΩS2 ; Ωy2 ; Ωb 1 b 1 ðtÞS2
ϕ1 ψ1 ϕ1 V_ 1 ðtÞ ≤ ϕ*1 S21 þ S22 þ y22 − k1 S21 − ϕ 1
2
and Ωb are appropriate compact sets. S1 ; S2 ; y2 ; ϕ b
ψ1 1 σ 1 π 1 ðtÞjS1 j 1 ~2
  þ − σ 1 ϕ 1 ðtÞ
and ψb 1 will be suppressed: ΩSi ¼ Si jSi j ≤ S ; ∃S > 0: π 1 ðtÞjS1 j þ σ 1 2 (20)
   i i
1 ~2 ~ * ~
⊂ Rði ¼ 1; 2Þ, Ωy2 ¼ y2 jy2 j ≤ y2 ; ∃y2 > 0: ⊂ R, Ωb ¼ − σ1 ϕ 1 ðtÞ−σ 1 ϕ 1 ðtÞϕ1 þ ϕ 1 ðtÞS1
2
ϕ 2
n  o n  1

b 1 jϕ
ϕ b1 j ≤ ϕ
 ; ∃ϕ
 > 0: ⊂ R, and b 1 jψ
Ωb ¼ ψ b1j ≤ ψ
 ; 1 ~2 1 ~2 ~ *
1 1 ψ1 1 − σ1ψ 1 ðtÞ− σ 1 ψ 1 ðtÞ−σ 1 ψ 1 ðtÞψ1
 > 0:g ⊂ R.
2 2
∃ψ 1
For the uncertain nonlinear subsystem described by (9) ab
By using the fact that 0 < ≤ b; ∀a; b > 0, (20)
and (12), we consider the Lyapunov function candidate of aþb
can be further written into
the following form
1 1 ~2
1 2 1 −1 ~ 2 1 ~2 V_ 1 ðtÞ ≤ S22 þ y22 − k1 S21 þ σ 1 − σ 1 ϕ 1 ðtÞ
V 1 ðtÞ ¼ S1 þ γ 11 ψ 1 ðtÞ þ γ −1 ϕ ðtÞ (15) 2 2
2g1 2 2 12 1 1 1 ~2 1
(21)
* 2
þ σ 1 ðϕ*1 Þ2 − σ 1 ψ 1 ðtÞ þ σ 1 ðψ1 Þ
The time derivative of V1(t) along the trajectories of (9) 2 2 2
and (12) results in Step i: (2 ≤ i ≤ n−1)
Taking the derivative of the error Si=xi−ξid, it is
1 ~ 1 ðtÞ
dψ ~ ðtÞ
dϕ derived that S_ i ¼ gi ð~ x i Þ þ di ðtÞ− ξ_ id :
x i Þx iþ1 þ f i ð~
V_ 1 ðtÞ ¼ S1 S_ 1 þ γ −1 ~ þ γ −1 ~
11 ψ 1 ðtÞ 12 ϕ 1 ðtÞ
1
(16)
g1 dt dt Then, similar to step 1, we choose the virtual controller
as
Based upon Assumption 2 and (13), it is deduced that
b i ðtÞSi − π i ðtÞSi
2
1
ξ iþ1 ¼ − ki Si − Si − ϕ (22)
g ð~x 1Þ g ð~
x 1Þ g ð~x 1Þ 2 π i ðtÞjSi j þ σ i
V_ 1 ðtÞ ≤ 1 S1 S2 þ 1 S1 y2 þ 1 S1 ξ 2
g1 g1 g1
! where ki and σi are any given positive constants and πi(t)
1 1 * 2  1 jS1 j b ðtÞ þ ‖ρ ð~
is defined as π i ðtÞ: ¼ ψ b ðtÞ
x i Þ‖2 : Furthermore, ψ
þ ‖ω ‖ þ ε 1 þ d i i i
g1 4g1 1 (17) b ðtÞ are the estimates of the unknown constants ψ*
and ϕ i i
þ ‖ρ1 ð~x 1 Þ‖2 jS1 j and ϕ*i , and they are updated through the following adap-
dψ~ ðtÞ ~ ðtÞ
dϕ tation laws
þ γ −1 ~ þ γ −1 ~
11 ψ 1 ðtÞ 12 ϕ 1 ðtÞ
1 1
dt dt
b ðtÞ

From Assumption 1 and (3a), we have
i b ðtÞ þ γ jSi j
¼ −γ i1 σ i ψ (23a)
i i1
dt

g 2 g 2 g ð~x 1Þ
V_ 1 ðtÞ ≤ 12 S21 þ S22 þ 12 S21 þ y22 þ 1 S1 ξ 2 b i ðtÞ

4g1 4g1 g1 b ðtÞ þ γ i2 jSi j2
¼ −γ i2 σ i ϕ (23b)
i
  dt
þ ψ*1 þ ‖ρ1 ð~ x 1 Þ‖2 jS1 j (18)
where γi1 and γi2 are any given positive constants.
~ ðtÞ
dψ ~ ðtÞ
dϕ Moreover, letting ψ~ i ðtÞ ¼ ψb i ðtÞ−ψ*i and ϕ~ i ðtÞ ¼ ϕb i ðtÞ−ϕ*i , it
þ γ −1 ~ þ γ −1 ~
11 ψ 1 ðtÞ 12 ϕ 1 ðtÞ
1 1
dt dt is held that
766 CHEN ET AL.

~ ðtÞ
dψ Based upon on Assumptions 1–2, (3), (22), (24), and
i ~ ðtÞ þ γ jSi j−γ σ i ψ*
¼ −γ i1 σ i ψ (24a) (27), it is derived that
i i1 i1 i
dt
~ ðtÞ
dϕ i ~ ðtÞ þ γ i2 jSi j2 − γ i2 σ i ϕ*
¼ −γ i2 σ i ϕ g 2 g
dt i i (24b) V_ i ðtÞ ≤ V_ i−1 ðtÞ þ i 2 S2i þ S2iþ1 þ y2iþ1 þ i Si
2gi gi
Considering the first‐order low‐pass filter, it can be

b ðtÞSi − π i ðtÞSi
2
1
obtained × − k i Si − Si − ϕ
2 i
π i ðtÞjSi j þ σ i
 
τ iþ1 ξ_ ðiþ1Þd þ ξ ðiþ1Þd ¼ ξ iþ1 ; ξ ðiþ1Þd ð0Þ ¼ ξ iþ1 ð0Þ (25) þ ψ~ ðtÞ þ ψ þ ‖ρ ð~
*
x i Þ‖ jSi j
2
i i i

1 1 ~ ðtÞS2 − σ i ϕ
~ 2 ðtÞ þ ϕ ~ ðtÞϕ*
Then, we define the errors Si+1 and yi+1 as þ 2 S2i þ 2 y2i − σ i ϕ i i i i i
2gi 2τ i
Siþ1 ¼ x iþ1 − ξ ðiþ1Þd (26a) ~ 2 ðtÞ−σ i ψ
−σ i ψ ~ ðtÞψ*
i i i
(31)
yiþ1 ¼ ξ ðiþ1Þd − ξ iþ1 ¼ −τ iþ1 ξ_ ðiþ1Þd (26b) From (2), we can readily conclude that

Then, S_ i can be expressed as i



1 1 ~2
V_ i ðtÞ ≤ S2iþ1 þ ∑ − k j S2j þ y2jþ1 þ σ j − σ j ϕ ðtÞ
2 2 j
S_ i ¼ gi ð~
x i ÞSiþ1 þ gi ð~x i Þyiþ1 þ gi ð~x i Þξ iþ1 j¼1

(27) 1 1 ~2 1
x i Þ þ di ðtÞ− ξ_
* 2
þ ω*⊤ ρ ð~ x i Þ þ εi ð~ þ σ j ðϕ*j Þ2 − σ j ψ ðtÞ þ σ j ðψ Þ (32)
i i id 2 2 j 2 j

i 1
From (26), it is deduced that þ ∑ 2 y2j
j¼22τ j
yiþ1 _
y_ iþ1 ¼− − ξ iþ1
τ iþ1 Step n:
y ∂ξ ∂ξ b_ ∂ξ iþ1 b_ (28) Taking the derivative of the error Sn=xn−ξnd, which
¼ − iþ1 − iþ1 S_ i − iþ1 ϕ i− ψ
τ iþ1 ∂Si b
∂ϕ i b i
∂ψ yields
i

It is clear that there exists the continuous function ηi+1 S_ n ¼ gn ð~ x n Þ þ dn ðtÞ− ξ_ nd


x n ÞuðtÞ þ f n ð~ (33)
(·) such that
yiþ1  To stabilize (33), we propose the following controller
jy_ iþ1 þ b;ψ
j ≤ ηiþ1 ðS1 ; … ; Siþ1 ; y2 ; …; yiþ1 ; ϕ b ≤ η iþ1
τ iþ1 i i
1 b ðtÞSn − π 2n ðtÞSn
uðtÞ ¼ − k n Sn − Sn − ϕ (34)
where ηiþ1 is the maximum of |ηi+1(·)| on compact region 2 n
π n ðtÞjSn j þ σ n
ΩS1 × … × ΩSiþ1 × Ωy2 × … × Ωyiþ1 × Ωb × Ωb . Here, ΩS1 ; … ;
ϕ ψ i i
where kn and σn are any given positive constants and πn(t)
ΩSiþ1 ; Ωy2 ; … ; Ωyiþ1 ; Ωb and Ωbψ are appropriate compact
ϕi i b ðtÞ þ ‖ρ ð~
is defined as π n ðtÞ: ¼ ψ b ðtÞ
x n Þ‖2 :Moreover, ψ
n n n
sets. b and
Siþ1 ; yiþ1 ; ϕ b i will be suppressed:
ψ
  i
  and ϕb n ðtÞ are the estimates of the unknown constants
  
ΩSiþ1 ¼ Siþ1 jSiþ1 j ≤ Siþ1 ; ∃Siþ1 > 0:g ⊂ R, Ωyiþ1 ¼ yiþ1 jyiþ1 j
n  o ψ*n and ϕ*n , and they are updated through the following
≤ yiþ1 ; ∃yiþ1 > 0:g ⊂ R, Ωbϕ ¼ ϕb i jϕb i j ≤ ϕ i ; ∃ϕ i > 0: ⊂ R, and adaptation laws
i

n  o
b jψ
Ωb ¼ ψ bj≤ψ ; ∃ψ
 > 0: ⊂ R.
ψi i i i i b ðtÞ
dψ n b ðtÞ þ γ jSn j
¼ −γ n1 σ n ψ (35a)
n n1
Consider the following Lyapunov function candidate dt

1 2 1 −1 ~ 2 1 ~2
V i ðtÞ ¼ V i−1 ðtÞ þ S þ γ ψ ðtÞ þ γ −1 ϕ ðtÞ (29) b ðtÞ

2gi i 2 i1 i 2 i2 i n b n ðtÞ þ γ n2 jSn j2
¼ −γ n2 σ n ϕ (35b)
dt
The time derivative of Vi(t) is deduced as
where γn1 and γn2 are any given positive constants.
~ Moreover, letting ψ~ n ðtÞ ¼ ψb n ðtÞ−ψ*n and ϕ~ n ðtÞ ¼ ϕb n ðtÞ−ϕ*n ,
1
V_ i ðtÞ ¼ V_ i−1 ðtÞ þ Si S_ i þ γ −1 ~ dψ i ðtÞ
i1 ψ i ðtÞ it is held that
gi dt
(30)
~
þ γ −1 ~ dϕ i ðtÞ ~ ðtÞ

i2 ϕ i ðtÞ n ~ ðtÞ þ γ jSn j−γ σ n ψ*
¼ −γ n1 σ n ψ (36a)
dt dt n n1 n1 n
CHEN ET AL. 767

~ ðtÞ
dϕ where
n ~ ðtÞ þ γ n2 jSn j2 − γ n2 σ n ϕ*
¼ −γ n2 σ n ϕ (36b)
n n
dt
μ¼ min min ðk j gj Þ;
ðγ j1 σ j Þ; min
j¼1; 2; … ; n j¼1; 2; … ; n
Consider the following Lyapunov function candidate
min ðγ j2 σ j Þ; min ð−2τ j N j Þ
1 2 1 −1 ~ 2 1 ~2 j¼1; 2; … ; n j¼2; … ; n
V n ðtÞ ¼ V n−1 ðtÞ þ S þ γ ψ ðtÞ þ γ −1 ϕ ðtÞ (37)

2gn n 2 n1 n 2 n2 n n 1 1 n 1
Δ ¼ ∑ σ j þ σ j ðϕ*j Þ2 þ σ j ðψ*j Þ2 þ ∑ 
ηj
j¼1 2 2 j¼24ð1 − mÞ
The time derivative of Vn(t) is deduced as
2τ 2j þ 1 − 2m
n
Nj ¼
1 1 ~2 2τ 2j
V_ n ðtÞ ≤ ∑ − k j S2j þ σ j − σ j ϕ ðtÞ
j¼1 2 2 j

Solving (42) results in 0 ≤ V ðtÞ ≤
Δ Δ
þ ðV ð0Þ− Þe−μt :
1 1 ~2 1 μ μ
þ σ j ðϕ*j Þ2 − σ j ψ ðtÞ þ σ j ðψ*j Þ2 (38) (
2 2 j 2 1 1
Letting δ ¼ min min ð Þ; min ð γ i1 Þ; :
n−1 n y2j i¼1; 2; … ; n 2g i¼1; 2; … ; n 2
þ ∑ y2jþ1 þ ∑ i

n 
2
j¼1 j¼22τ j 1 ~ 2 ðtÞþ
min ð γ i2 Þ , We have ∑i¼1 S2i þ ψ i
i¼1; 2; … ; n 2

~ 2 ðtÞÞ ≤ δ−1 Δ þ V ð0Þ :Therefore, all signals


The block diagram of the proposed adaptive robust
ϕ
control scheme is shown in Figure 1. i
μ
Then, we have the following Theorem 1. in the closed‐loop system remain bounded,
Theorem 1. Consider the uncertain strict‐ and the state x~n ðtÞ and ψ~ i ðtÞ; ϕ~ i ðtÞ; i ¼ 1; ···; n
feedback nonlinear system described by (1). eventually converge to the compact set
Suppose that Assumptions 1 and 2 are satis- n
 
fied. Under the adaptive robust control scheme Ωs ≜ x~n ðtÞ; ψ 1 ðtÞ; ···; ψ n ðtÞ; ϕ 1 ðtÞ; ···; ϕ n ðtÞ ∑ S2i
~ ~ ~ ~
given in (34) with (35), the state x~n ðtÞis uni-
i¼1


 Δ
formly bounded in the presence of the nonlin- ~ 2
~ −1
þ ψ i ðtÞ þ ϕ i ðtÞ ≤ δ
2
þ V ð0Þ
ear uncertainties, external disturbances, and μ
unknown time‐varying virtual control
coefficients. Thus, we complete the proof of Theorem 1.

Proof From the backstepping design proce- Remark 2. It is clear that for the sufficiently
dure above, we consider the Lyapunov func- small time constants τj of the first‐order
tion candidate as follows: low‐pass filter, there always exists a positive
constant m such that Nj ≤ 0. Furthermore, m
V ðtÞ ¼ V n ðtÞ þ V y ðtÞ (39) is not required to be known for the system
designers. In this paper, the selections of time
where constants τj are not restricted by ηj , that is, it

n 1 2 does not matter whether ηj are known.


V y ðtÞ ¼ ∑ yj (40)
j¼2 2τ j
Remark 3. It is noteworthy that the lower
Based on (26) and (28), we can easily obtain and upper bounds of gi(·) are not required
in the proposed controllers. They are only

!
1n n 1 y2j used for stability analysis in the Proof,
V_ y ðtÞ ¼∑ yj y_ j ≤ ∑ − 2 þ jyj j
ηj whereas the lower and upper bounds of gi(·)
j¼2 τ j j¼2 τj τj
! are required in [20–22], and the lower
n m 2 m−1 2 bounds of gi(·) are needed in the proposed
≤ ∑ − 2 yj þ 2 yj
j¼2 τj τj controllers in [25–27]. Moreover, although
! the problem of ‘explosion of terms’ is avoided
n 1−m 2 1
þ∑ y þ 
η ; 0<m<1 via employing the dynamic surface control
j¼2 τ 2j j 4ð1 − mÞ j approach in [30–35], in [30] and [34], the
(41) time‐varying virtual control coefficients are
According to (38) and (41), V_ ðtÞ is known functions; in [31],[33], and [35], neu-
ral networks are used to estimate the terms
V_ ðtÞ ≤ −μV ðtÞ þ Δ (42) including the negative first power of control
768 CHEN ET AL.

FIGURE 1 The block diagram of the proposed adaptive robust control scheme

coefficients, which requires the first order dif- to be updated online are reduced in the vir-
ferentiability of control coefficients; in [32], tual controllers ξi+1 with i=1,···,n−1 and true
the negative first power of the derivatives of controller u.
control coefficients are needed in the pro-
posed controllers. Remark 6. It should be mentioned that the
size of convergent region Ωs can be suffi-
Remark 4. Being different from [48], the ciently small by reasonably selecting the con-
upper bounds of the norm of ideal weight trol gain ki,i=1,···,n, the parameters of
vector are not needed. Meanwhile, compared adaptation laws γi1,γi2,σi,i=1,···,n and time
with [49] and [53], the upper bounds of the constants τi,i=2,···,n. Obviously, Δ can be
approximation errors are not required to be reduced by decreasing σi, which makes the
known. Additionally, in the controller size of convergent region Ωs sufficiently
design, the norm of the ideal weight vector small. However, in this situation, from the
in the fuzzy logic systems is taken as the esti- definition of the positive constant μ, the cor-
mation parameter, not as the elements of the responding positive constants γi1,γi2, some
weight vector. As a result, by comparison other tunable parameters in the adaptation
with [48–53], the computational load is laws, should be large enough to maintain
severely reduced. the invariance of the the positive constant μ.
At the same time, it should be noted that
Remark 5. As a matter of fact, in the tradi- decreasing the value of the σi may give rise
tional manner of backstepping, ξn and the to a lower convergence rate in the terms of
(higher order) derivatives of ξi with i=2,···,n the adaptation laws. Besides, larger control
−1 would have to emerge in true controller gain ki may lead to better control perfor-
u, causing u to be much more complex. For mance in general. However, if the controller
higher dimensional systems, we would find gain ki is too large, oversaturation will occur.
that the contrast is greater. By fusion of Consequently, in practical application, design
dynamic surface control technique, the pro- parameters should be carefully adjusted so
posed control scheme is much simpler than that better stability can be guaranteed for
[47] and [54] since the differentiation of vir- the systems.
tual controller ξi,i=2,3,…,n is replaced by
ξ_ ; i ¼ 2; 3; … ; n, and ξid are obtained by
id
using ξi as the inputs through a first‐order 4 | SIMULATION RESULTS
low‐pass filter. Hence, simpler algebraic
In this section, we present a practical example to validate
operations can take the place of differential
the performance of our proposed scheme, and the control
operations. This can be explained more
performance is compared using the method of [27].
clearly in equations (10), (11), (25), and (26).
Similar to [54–56], the stabilization of a single‐link
Thus, the repeated differentiations of virtual
manipulator actuated by a brush DC motor is considered.
controllers ξi with i=2,···,n−1 are avoided.
Its dynamic model can be expressed as follows.
Moreover, compared with [54], not only the
issue of ‘explosion of terms’ is better solved,
but also the number of adaptive parameters  q€ þ N
M  sinðqÞ þ B
 q_ ¼ I þ ΔðIÞ (43a)
CHEN ET AL. 769

For the controllers and adaptation laws, the


 ¼ −RI − K B q_ þ V þ ΔðV Þ
Lİ (43b) parameters are chosen as k1 = 0.5, k2 = 0.5,
2 2 2 k3 = 0.1, γ12 = 100, γ21 = γ22 = 10, γ31 = γ32 = 100,
where  ¼ J þ md þ Md þ 2Mδ ; N
M  ¼ mdg þ Mdg and σ1 = 0.4, σ2 = 0.5, σ3 = 0.2. It is worth men-
K r 3K r Kr 5K r 2K r Kr
tioning that because of f 1(·) = 0, the approx-
 ¼ B . J is the motor rotor inertia, m is the link mass,
B
Kr imation is unnecessary. Then, the time
M is the load mass, d is the link length, δ is the radius constants of the low‐pass filter and the initial
of load, g is the gravity coefficient, B is the coefficient of conditions are set as τ 2 ¼ 0:02; τ 3 ¼ 0:04;
viscous friction at the joint, Kr is the coefficient which b ð0Þ ¼ 0:5;
x 1 ð0Þ ¼ x 2 ð0Þ ¼ x 3 ð0Þ ¼ 0:8; ϕ1
characterizes the electromechanical torque coupling, L  b ð0Þ ¼ 1; ϕ
b 2 ð0Þ ¼ ψ b ð0Þ ¼ 0:4.
b 3 ð0Þ ¼ ψ
ϕ 2 3
is the armature inductance, q(t) is the angular motor rotor With our proposed algorithm and the above
(load) position, I(t) is the motor armature current, ΔI is parameter settings, the simulation results
the current disturbance, V(t) is the input control voltage, are shown in Figure 2.
ΔV is the input control voltage disturbance.
Setting _
x 1 ðtÞ ¼ qðtÞ; x 2 ðtÞ ¼ qðtÞ; x 3 ðtÞ ¼ IðtÞ and Case B. Using the controller of [27] to con-
uðtÞ ¼ VðtÞ, (43) can be transformed into trol this system. First, the lower bound values
10 25
x_ 1 ¼ x 2 (44a) are chosen as g1 ¼ 1, g2 ¼ and g3 ¼ .
11 2
Then, to approximate the structure uncer-
1 
N 
B ΔI
x_ 2 ¼  x 3 −  sinðx 1 Þ−  x 2 þ  (44b) tainties f i(·), i = 2,3, we take the basis func-
M M M M tions of RBFNN as
!
1 KB R 1 ‖~
x i − cj ‖ 2
x_ 3 ¼  u −  x 2 −  x 3 þ  ΔV (44c) ρij ð~
x i Þ ¼ exp − ; i ¼ 2; 3; j ¼
L L L L δ2ij
It is obvious that nonlinear system (44) satisfies 1; 2; ···; Li , where L2=L3=11, c2=[cj1cj2]⊤,
unmatched condition. The system parameters and distur- c3=[cj1cj2cj3]⊤, and the centers cj1,cj2 and cj3
bances are given as M ¼ 1 þ 0:1 sinð2tÞ; N
 ¼ 10 þ sinð2tÞ; are, respectively, spaced in [−0.50.5],
 ¼ 1 þ 0:1 sinð2tÞ; L
B  ¼ 0:05 þ 0:03 sinð2tÞ; R ¼ 0:8; K B ¼ 5; [−0.50.5] and [−0.50.5] with widths
ΔðIÞ ¼ 0:08 sinð2tÞ; ΔðV Þ ¼ 0:03 cosð2tÞ. δ2j=δ3j=0.1. For the controller and adapta-
Case A. Using the proposed method to con- tion laws of [27], the parameters are taken
trol this system. To estimate the structure as k1 = k2 = k3 = 2.5, γ12 = γ21 = γ22 =
uncertainties, we choose the following fuzzy γ3 = m1 = m2 = 1, and σ 1 ðtÞ ¼ 10expð−0:08tÞ;
membership functions. σ 2 ðtÞ ¼ 10expð−0:08tÞ and σ 3 ðtÞ ¼
100expð−0:08tÞ. Finally, the initial condi-
2 1 tions are set as x 1 ð0Þ ¼ x 2 ð0Þ ¼ x 3 ð0Þ ¼
μF 1 ðx 1 ; x 2 Þ ¼ ∏
i¼1 1 þ exp½5ðx i þ 0:6Þ 0:8; ϕ b ð0Þ ¼ ϕ
b ð0Þ ¼ 0:5; ψ b ð0Þ ¼ b
κ 1 ð0Þ ¼ 1,
1 2 2
2  and ψb ð0Þ ¼ b κ 2 ð0Þ ¼ 0:4. With the method of
μF l ðx 1 ; x 2 Þ ¼ ∏exp − ðx i þ 0:8−0:2lÞ2 ; l ¼ 2; ···; 6 3
i¼1 [27], the simulation results are shown in
2 1 Figure 3.
μ F 7 ðx 1 ; x 2 Þ ¼ ∏
i¼1 1 þ exp½ −5ðx i − 0:6Þ It can be seen from Figures 2 and 3 that both
3 1 the proposed controller and the controller of
μ F 1 ðx 1 ; x 2 ; x 3 Þ ¼ ∏
i¼1 1 þ exp½5ðx i þ 0:6Þ [27] can solve the stabilization problem of
3  system (44). However, the proposed control-
μF l ðx 1 ; x 2 ; x 3 Þ ¼ ∏exp − ðx i þ 0:8−0:2lÞ2 ; l ¼ 2; ···; 6 ler illustrates a better performance than the
i¼1
3 1 controller of [27] in terms of convergence
μF 7 ðx 1 ; x 2 ; x 3 Þ ¼ ∏ time. From Figures 2(a) and 3(a), the conver-
i¼1 1 þ exp½−5ðx i − 0:6Þ
gence time of our proposed controller is
Then, we can obtain about 1.8 seconds, which is shorter than the
convergence time of [27]. Furthermore, it
μF l ð~
x iÞ
ρl ð~
x iÞ ¼ 7
; i ¼ 2; 3 l ¼ 1; …; 7 can be observed from Figures 2(b), 2(c),
∑ μF j ð~
x iÞ 2(d) and 3(b) that the number of adaptive
j¼1 parameters to be updated online are reduced
ρð~
x i Þ ¼ ½ρ1 ð~
x i Þ; ρ2 ð~ x i Þ⊤ ;
x i Þ; ···; ρ7 ð~ i ¼ 2; 3 from 6 to 5.
770 CHEN ET AL.

(A) (B)

(C) (D)
FIGURE 2 Case A: Using the proposed method to control this system [Color figure can be viewed at wileyonlinelibrary.com]

(A) (B)
FIGURE 3 Case B: Using the controller of [27] to control this system [Color figure can be viewed at wileyonlinelibrary.com]

5 | CONCLUSIONS fuzzy logic systems. Also, it was proved that the proposed
controller can guarantee the uniform boundedness of
In this paper, we have dealt with the adaptive robust sta- uncertain strict‐feedback nonlinear systems. Finally, the
bilization problem for a class of uncertain strict‐feedback application to the problem of stabilization for a single‐
nonlinear systems in the presence of unknown structure link manipulator actuated by a brush DC motor is pre-
uncertainties, external disturbances, and unknown time‐ sented to demonstrate the utilization of the proposed
varying virtual control coefficients. The upper bounds of controller.
external disturbances, as well as the upper and lower There are several diverse opportunities for future
bounds of unknown time‐varying virtual control coeffi- research. First, the results of this paper could be extended
cients, were not required to be known. We designed the to uncertain strict‐feedback nonlinear systems with full‐
controller by adopting the backstepping manner. Further- state constraints. Second, the signs of the control coeffi-
more, to avoid suffering from the problem of ‘explosion of cients are assumed to be known, the relaxation of this
terms’, a dynamic surface control approach was employed condition would be of interest. Third, the control gains
by introducing the first‐order low‐pass filter. Particularly, ki, the adaptive parameters γi1,γi2 and σi and time con-
we exploited the fuzzy logic systems to approximate the stants τi should be carefully selected by the designers of
unknown structure uncertainties without a priori knowl- the systems so that better stability can be guaranteed for
edge on the weight matrices and approximation errors of the systems. How to choose these parameters to obtain
CHEN ET AL. 771

the best control performance would be of considerable 15. S. Tong and Y. Li, Adaptive fuzzy output feedback control for
interest. switched nonlinear systems with unmodeled dynamics, IEEE
Trans. Cybern. 47 (2017), 295–305.
ACKNOWLEDGEMENT 16. S. Li, X. Liu, and X. Dong, Non‐certainty equivalent practically
adaptive control for high‐order lower triangular systems, IMA J.
Scientific Research Foundation of the Education Math. Control. Inf. 32 (2015), 809–822.
Department of Sichuan Province, China,Grant Number, 17. H. Wu, Simple adaptive robust control schemes of uncertain strict‐
18ZB0456. feedback nonlinear time‐delay systems, IET Control Theory Appl.
11 (2017), 2222–2231.
ORCID 18. Y. Li and G. Yang, Event‐triggered adaptive backstepping control
for parametric strict‐feedback nonlinear systems, Int. J. Robust
Yuchao Wang https://orcid.org/0000-0002-6541-0663 Nonlinear Control 28 (2018), 976–1000.
19. H. Wang et al., Direct adaptive neural control of nonlinear strict‐
feedback systems with unmodeled dynamics using small‐gain
REFERENCES
approach, Int. J. Adapt Control Signal Process. 30 (2016),
1. S. Li, X. Liu, and X. Liu, The coordinated immersion and vari- 906–927.
ance control of power systems with excitation and steam‐valve, 20. Y. Yu and Y. Zhong, Robust backstepping output tracking control
J. Dyn. Syst. Meas. Control. 140 (2018), 071014. for SISO uncertain nonlinear systems with unknown virtual con-
2. D. Xu et al., Adaptive command‐filtered fuzzy backstepping con- trol coefficients, Int. J. Control. 83 (2010), 1182–1192.
trol for linear induction motor with unknown end effect, Inf. Sci. 21. B. Chen, X. Liu, and C. Lin, Observer and adaptive fuzzy control
477 (2019), 118–131. design for nonlinear strict‐feedback systems with unknown virtual
3. H. Liang et al., Adaptive distributed observer approach for cooper- control coefficients, IEEE Trans. Fuzzy Syst. 26 (2018), 1732–1743.
ative containment control of nonidentical networkst, IEEE Trans. 22. H. V. Le and K. T. Chu, Robust control of positive fractional‐order
Syst. Man Cybern. Syst. Hum. 49 (2019), 299–307. interconnected systems with heterogeneous delays, Asian J. Con-
4. W. Shen, H. Huang, and J. Wang, Robust backstepping sliding trol 21 (2019), 596–608.
mode controller investigation for a port plate position servo system 23. W. Chen et al., Globally stable adaptive backstepping neural net-
based on an extended states observer, Asian J. Control 21 (2019), work control for uncertain strict‐feedback systems with tracking
302–311. accuracy known a priori, IEEE Trans. Neural Netw. Learn. Syst.
5. H. Liang, Z. Zhang, and C. K. Ahn, Event‐triggered fault detection 26 (2015), 1842–1854.
and isolation of discrete‐time systems based on geometric tech- 24. S. S. Ge, F. Hong, and T. H. Lee, Adaptive neural control of non-
nique, IEEE Trans. Circuits Syst. II‐Express Briefs (2019). linear time‐delay systems with unknown virtual control
https://doi.org/10.1109/TCSII.2019.2907706 coefficients, IEEE Trans. Syst. Man Cybern. Part B Cybern.
6. E. Lavretsky and K. A. Wise, Robust and adaptive control, 34 (2004), 861–872.
Spring‐Verlag, New York, 2013. 25. Y. Yu and Y. Zhong, Semi‐global robust output tracking for non-
7. H. K. Khalil, Nonlinear systems, 3rd Edition, Prentice‐Hall, linear uncertain systems in strict‐feedback form, IET Control
Upper Saddle River, New Jersey, 2002. Theory Appl. 6 (2012), 751–759.
8. R. A. Freeman and P. V. Kokotovic, Robust nonlinear control 26. H. Wu, Adaptive robust stabilization of uncertain nonlinear
design: state space and Lyapunov techniques, Birkhauser, Boston dynamical systems: an improved backstepping approach, Int. J.
MA, 1996. Control. 91 (2018), 114–131.
9. H. Wu, Continuous adaptive robust controllers guaranteeing uni- 27. Y. Wang and H. Wu, Adaptive robust backstepping control for a
formly ultimate boundedness for uncertain nonlinear systems, Int. class of uncertain dynamical systems using neural networks, Non-
J. Control. 72 (1999), 115–122. linear Dyn. 81 (2015), 1597–1610.
10. S. Xu and G. Feng, Further results on adaptive robust control of 28. D. Swaroop et al., Dynamic surface control for a class of nonlin-
uncertain time‐delay systems, IET Control Theory Appl. 2 ear systems, IEEE Trans. Autom. Control 45 (2000), 1893–1899.
(2008), 402–408. 29. Z. J. Yang, Robust control of nonlinear semi‐strict feedback sys-
11. M. Krstic, I. Kanellakppoulos, and P. V. Kokotovic, Nonlinear tems using finite‐time disturbance observers, Int. J. Robust
and adaptive control design, Wiley, New York, 1995. Nonlinear Control 27 (2017), 3582–3603.
12. I. Kanellakopoulos, P. V. Kokotovic, and A. S. Morse, Systematic 30. Y. Pan and H. Yu, Dynamic surface control via singular perturba-
design of adaptive controllers for feedback linearizable systems, tion analysis, Automatica 57 (2015), 29–33.
IEEE Trans. Autom. Control 36 (1991), 1241–1253. 31. M. Chen, G. Tao, and B. Jiang, Dynamic surface control using neu-
13. F. Wang and X. Zhang, Adaptive finite time control of nonlinear ral networks for a class of uncertain nonlinear systems with input
systems under time‐varying actuator failures, Systems, 2018. saturation, IEEE Trans. Neural Netw. Learn. Syst. 26 (2015),
https://doi.org/10.1109/TSMC.2018.2868329 2086–2097.
14. G. Lai et al., Robust adaptive fuzzy control of nonlinear systems 32. S. Han and J. Lee, Adaptive fuzzy backstepping dynamic surface
with unknown and time‐varying saturation, Asian J. Control 17 control for output‐constrained non‐smooth nonlinear dynamic
(2014), 791–805. system, Int. J. Control. Autom. Syst. 10 (2012), 684–696.
772 CHEN ET AL.

33. M. Wang and C. Wang, Learning from adaptive neural dynamic 50. Y. Li, K. Li, and S. Tong, Finite‐time adaptive fuzzy output feed-
surface control of strict‐feedback systems, IEEE T. Neur. Net. back dynamic surface control for MIMO non‐strict feedback
Lear. 26 (2015), 1247–1259. systems, IEEE Trans. Fuzzy Syst. 27 (2019), 96–110.
34. Y. Pan and H. Yu, Composite learning from adaptive dynamic 51. L. Wu, X. He, and D. Zhang, Cooperative adaptive fuzzy control
surface control, IEEE Trans. Autom. Control 61 (2016), for a class of uncertain non‐linear multi‐agent systems with time
2603–2609. delays, J. Control Decis. 4 (2016), 131–152.
35. N. Abasi, M. Kamali, and M. Zekri, Application of RBF neural 52. L. Wu and J. H. Park, Adaptive fault‐tolerant control of uncertain
networks in robust adaptive DSC design of nonlinear systems, switched nonaffine nonlinear systems with actuator faults and
2017 25th Iranian Conference on Electrical Engineering. time delays, IEEE Trans. Syst. Man Cybern. Syst. (2019).
Tehran, Iran, 2017, pp. 877–882. https://doi.org/10.1109/TSMC.2019.2894750
36. T. Zhang, M. Xia, and Y. Yi, Adaptive neural dynamic surface 53. S. Tong, X. He, and Y. Li, Direct adaptive fuzzy backstepping
control of strict‐feedback nonlinear systems with full state con- robust control for single input and single output uncertain nonlin-
straints and unmodeled dynamics, Automatica 81 (2017), ear systems using small‐gain approach, Inf. Sci. 180 (2010),
232–239. 1738–1758.
37. J. Ma, Z. Zheng, and P. Li, Adaptive dynamic surface control of a 54. Y. Wang, L. Xu, and H. Wu, Adaptive robust backstepping output
class of nonlinear systems with unknown direction control tracking control for a class of uncertain nonlinear systems using
gains and input saturation, IEEE Trans. Cybern. 45 (2015), neural network, J. Dyn. Syst. Meas. Control. 140 (2018).
728–741. 55. J. Carroll and D. Dawson, Integrator backstepping techniques for
38. D. Swaroop et al., Dynamic surface control of nonlinear systems, the tracking control of permanent magnet brush DC motors, IEEE
Proceedings of the American Control Conference. Albuquerque, Trans. Ind. Appl. 32 (1995), 248–255.
NM, USA, 1997, pp. 3028–3034. 56. J. Mao et al., Adaptive practical stabilization of a class of uncer-
39. Z. Peng et al., Adaptive dynamic surface control for formations of tain nonlinear systems via sampled‐data control, Nonlinear
autonomous surface vehicles with uncertain dynamics, IEEE Dynamics 92 (2018), 1679–1694.
Trans. Control Syst. Technol. 21 (2013), 7513–520.
40. H. Wang, Y. Li, and K. Liu, Globally stable adaptive dynamic sur-
face control for cooperative path following of multiple
underactuated autonomous underwater vehicles, Asian J. Control AUTHOR BIOGRAPHIES
20 (2018), 1204–1220.
Sanyan Chen is currently a M.S.
41. Q. Du et al., Robust adaptive dynamic surface control of chaos in
degree candidate in the College of
permanent magnet synchronous motor, Phys. Lett. A 363 (2007),
71–77. Mechanical and Electrical Engineer-
42. M. Hou, X. Liang, and G. Duan, Adaptive block dynamic surface
ing, Sichuan Agricultural University.
control for integrated missile guidance and autopilot, Chin. J. Her current research interests include
Aeronaut. 26 (2013), 741–750. adaptive robust control, fuzzy control,
43. L. Wang, Adaptive fuzzy systems and control: design and stability and mobile robots.
analysis, Prentice‐Hall, Englewood Cliffs, NJ, 1994.
44. W. Wang et al., Adaptive fuzzy control for strict‐feedback canoni- Yuchao Wang received his Ph.D. in
cal nonlinear systems with H∞ tracking performance, IEEE Trans. management and information systems
Syst. Man Cybern. Part B Cybern. 30 (2000), 878–885. from Prefectural University of Hiro-
45. D. Zhai et al., Adaptive fuzzy fault‐tolerant control with guaran- shima, Japan, in March 2016. In July
teed tracking performance for nonlinear strict‐feedback systems, 2016, he joined the College of
Fuzzy Sets Syst. 302 (2016), 82–100. Mechanical and Electrical Engineer-
46. Y. Li et al., Adaptive output‐feedback control design with pre- ing, Sichuan Agricultural University, China, where
scribed performance for switched nonlinear systems, Automatica he is an associate professor. His research interests
80 (2017), 225–231.
are nonlinear systems, uncertain systems, time‐delay
47. H. Liang et al., Prescribed performance cooperative control for
systems, stochastic systems, adaptive robust control,
multiagent systems with input quantization, IEEE Trans. Cybern.
and their applications.
(2019). https://doi.org/10.1109/TCYB.2019.2893645
48. X. Li and G. Yang, FLS‐based adaptive synchronization control of
Wenrui Shi is currently a M.S.
complex dynamical networks with nonlinear couplings and state‐
degree candidate in the College of
dependent uncertainties, IEEE Trans. Cybern. 46 (2016),
171–180. Mechanical and Electrical Engineer-
49. S. Tong and Y. Li, Adaptive fuzzy output feedback tracking ing, Sichuan Agricultural University.
backstepping control of strict‐feedback nonlinear systems with His current research interests include
unknown dead zones, IEEE Trans. Fuzzy Syst. 20 (2012), adaptive control, neural network con-
168–180. trol, and agricultural robots.
CHEN ET AL. 773

Hansheng Wu. received his BS and How to cite this article: Chen S, Wang Y, Shi W,
MS in Automatic Control Engineering Wu H. Adaptive robust dynamic surface control for
from Northeast University of Technol- uncertain strict‐feedback nonlinear systems using
ogy, China, in 1982 and 1984, respec- fuzzy logic systems. Asian J Control. 2021;23:
tively, and a Ph.D in Information 761–773. https://doi.org/10.1002/asjc.2253
Engineering from Hiroshima Univer-
sity, Japan, in 1989. He is currently working as a Pro-
fessor in the Department of Information Science,
Prefectural University of Hiroshima, Japan. His
research interests include optimal control, dynamical
games, large‐scale systems, robust control, adaptive
control, and their applications.

You might also like