This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access
Adaptive Finite Time Control of Nonlinear
Quantized Systems with Actuator Dead-Zone
Yan Zhang, Fang Wang, Jianhui Wang, and Yuanyuan Huang
Abstract—This paper investigates a finite-time control prob-
lem of nonlinear quantized systems with actuator dead-zone in
a non-strict feedback form. By combing a simplified dead-zone
model and the sector-bound characteristic of a hysteretic quan-
tizer, the control difficulty of coexistence of unknown actuator
dead-zone and control signal quantization effect can be overcome.
By applying the approximation ability of neural network systems,
an integrated adaptive controller is constructed, which can
compensate the unknown control gain. The designed neural
controller can ensure the transient performance of nonlinear
quantized systems with actuator dead-zone in finite-time. Based
on the Bhat and Bernstein theorem, the finite-time stability of
system is proved. Finally, a numerical example is given to verify
the validity of the proposed approach.
Index Terms—Adaptive neural control, backstepping tech-
nique, unknown dead-zone, nonlinear quantized systems, finite-
time stability.
I. I NTRODUCTION
Over the past few decades, the adaptive fuzzy or neural
control based on approximation has received extensive atten-
tion. In [1]- [12], fuzzy logical or neural network systems
were applied to model the unknown nonlinear functions, and
the adaptive controllers were designed by combing adaptive
technique with backstepping. Although some achievements
have been made in [1]- [12], the impact of quantization on the
systems was neglected. Therefore, the problem of quantized
Y. Zhang is with the School of Mathematics and Systems Science, Shan-
dong University of Science and Technology, Qingdao 266590, China (e-mail:
717774522@qq.com).
F. Wang is with the College of Mathematics and Systems Science, Shandong
University of Science and Technology, Qingdao 266590 (e-mails: sandy-
wf75@126.com).
J. H. Wang is with the College of Automation, Guangdong University
of Technology, Guangzhou 510006, China, and also with the School of
Mechanical and Electric Engineering, Guangzhou University, Guangzhou
510006, China.(e-mail: jhwang@gzhu.edu.cn)
Y. Y. Huang is with the College of Computer and Communication Engi-
neering, Changsha University of Science and Technology, Changsha 410000
(e-mails: snailhyy@126.com).
2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
control for linear and non-linear programs has also caught the
attention of many scholars. The stability of a category of linear
quantized systems was investigate in [13]- [14]. In addition,
the quantized control of a kind of uncertain nonlinear systems
was studied in [15]- [19]. Compared with the schemes in [13]-
[14], the system models in [15]- [19] do not need to be fully
known to the designer. By applying backstepping skill, some
novel approaches were proposed in [20]- [23] for the input
quantization to compensate the unknown control gain. Unlike
some existing control strategies for input quantized nonlinear
systems, the proposed control schemes in [20]- [23] do not
require the global Lipschitz condition and the quantization
parameters can be unknown to the designer. In [24], a new
control approach was presented for a kind of stochastic non-
linear systems with unknown hysteresis. It is noteworthy that
the adaptive fuzzy or neural control in [22] and [24] requires
the systems to have pure feedback form. In theory, when the
system does not meet these situations, the system can not be
controlled by the controller. Furthermore, for a more general
category of non-strict feedback systems with quantized input,
the issue of an adaptive tracking was discussed in [25]. By
applying the Lyapunov function approach, for a category of
switched stochastic systems with quantized input, a systematic
adaptive control method was developed in [26]. Although the
progress have been made for quantized systems in [13]- [26],
the problem of actuator dead-zone was not considered.
It is well known that the actuator dead-zone can worse
the systems performance. In order to deal with the dead-zone
nonlinearity, many approaches have been brought forward in
[27]- [31] to design the controllers. Among them, there are
two main methods are usually employed. One method, as
shown in [32]- [35], to compensate for the impact by creating
a slick inverse of dead-zone. The another method, as described
in [36]- [45], facilitates the control design by establishing a
streamlined dead zone model. The above achievements are in
view of the implicit assumption that the control commands
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

transmitted between the physical actuator and the controller highlighted as follows.
does not present quantization phenomena. The work [46] Compared with the existing researches on nonlinear
investigates the issue of stochastic nonlinear quantized systems quantized systems with actuator dead-zone, a novel finite-time
with actuator dead zone. However, the aforementioned control control strategy is developed. The proposed control scheme
researches in [13]- [46] focus on the problem of the infinite ensures the transient performance of quantized system with
time stability. In theory, the control schemes in [13]- [46] can actuator dead-zone in finite-time. On paper, a valid finite-time
not ensure the transient response of the system in finite time. solution is obtained for nonlinear quantized systems with
actuator dead-zone. Furthermore, according to the structural
As a class of time-optimal control, the finite-time control properties of radial basis function (RBF) neural networks
has been caught more and more attention by scholars in recent (NNs), the structure of the plants in this manuscript is in
years. The key characteristic of the finite-time control is to non-strict feedback form. Hence, the control scheme is more
make the system reach equilibrium in finite time and keep challenging and the control system is more common. The rest
equilibrium thereafter. A sliding controller was constructed of this manuscript is arranged as follows. The second part
in [47] by establishing a terminal mode sliding manifold. In provides necessary preparation and problem statement. The
order to solve the flutter phenomena caused by discontinuous third part addresses a tracking problem of non-strict feedback
controller, the Lyapunov theory of finite-time stability has been nonlinear quantized systems with actuator dead zone, and the
built for the first time in [48]- [49]. Then, the Lyapunov sta- finite time convergence is proved. The fourth part verifies the
bility theory in [48]- [49] is utilized to handle the subsequent effectiveness of the controller through an example. The fifth
finite-time control issues for the nonlinear systems in [50]- part gives the conclusion and points out the future research
[51]. In order to ensure the finite-time stability of nonlinear direction.
systems, some control programs were designed in [52]- [54]
for a kind of nonlinear systems with hysteretic characteristics. II. P REREQUISITES A ND P ROBLEM F ORMULATION
By employing more appropriate Lyapunov-Krasovskii func-
A. Prerequisites
tional, the finite-time stability of time-varying delay systems
Lemma 1 (see [63]). Considering the nonlinear system
was discussed, see [55]- [58]. Furthermore, a new finite-
ρ = f (ρ, ς), for smooth positive definite function V (ρ) ∈ C 1 ,
time control strategy was presented in [59] for a category of
if there exist scalars c > 0, d > 0, and0 < σ < 1 satisfying
nonlinear system with actuation failures. The proposed control
that
scheme in [59] depends neither on the accurate model of
the system nor actuation failures. In addition, for a kind of V̇ (ρ) ≤ −cV σ (ρ) + d, t ≥ 0 (1)
nonlinear systems with dead-zone, several finite-time tracking
then, the system ρ̇ = f (ρ, ς) is SGPFS.
control programs were investigates in [60]- [65]. In [66],
Remark 1. Lemma 1 gives an important criterion of
a new adaptive finite-time output-feedback control method
SGPFS, which will be used to prove that the system is semi-
was proposed for a kind of nonlinear quantized systems with
global and practically finite-time stable.
unmeasurable states. By using the nonlinear decomposition
Lemma 2 (see [67]). For ιk ∈ R, k = 1, ..., n, 0 < p ≤ 1,
of hysteretic quantizer, the problem of quantization in finite-
one has
time was solved successfully in [67]. However, the works
(∑
n )p ∑
n (∑
n )p
[66]- [67] do not take into consideration the influence of the |ιk | ≤ |ιk |p ≤ n1−p |ιk | . (2)
actuator dead-zone on the control performance. To the best of k=1 k=1 k=1

our knowledge, up to now, the finite-time quantized control Lemma 3 (see [68]). For ∀ξ ∈ R and ∀ϵ > 0, the
problem of a category of nonlinear systems with actuator following relationship can be obtained:
(ξ )
dead-zone has not been studied, although its potentialities in
0 ≤ |ξ| − ξ tanh ≤ κϵ, κ = 0.2785. (3)
networked control systems. Considering the above discussion, ϵ
˙
this manuscript is devoted to solving the finite-time control Lemma 4 (see [38]). For θ̂(t) = −γ θ̂(t) + κν(t), if
problem of non-strict feedback nonlinear quantized systems ν(t) > 0 and θ̂(t0 ) ≥ 0 is satisfied, one has θ(t) ≥ 0 for
with actuator dead-zone. The contributions of this paper are ∀t ≥ t0 , where γ > 0 and κ > 0 denote the design parameters.

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

Lemma 5 (see [69]). For any real variables υ and ω, the unknown actuator dead zone of the finite-time deterministic
following inequality is satisfies: system. And this study will provide a stability analysis scheme
τ γ −τ for the actuator nonlinearity based on quantitative control
|υ|τ |ω|γ ≤ λ|υ|τ +γ + λ γ |ω|τ +γ . (4)
τ +γ τ +γ system.
where τ > 0, γ > 0 and λ > 0 represent the design Assumption 1. There exist smooth monotonously in-
parameters. creasing functions βi : R+ −→ R+ , i = 1, 2, . . . , n, under
Remark 2. Same as the finite-time studies in [1], [2], the initial value of βi (0) = 0 such that
[66], [67], Lemma 2 and Lemma 5 will be used widely to
|fi (x)| ≤ βi (∥x∥).
cope with the inequality (56) and (58).
What needs to be pointed out is that βi is a monotone
B. Problem description increasing function, which satisfies βi (0) = 0, that means
∑n ∑n
Think about the non-strict feedback system as follows: βi ( k=1 ak ) ≤ k=1 βi (nak ), where ak > 0. There is
 a smooth function hi (s) that makes βi (s) = shi (s). which

 ẋi = xi+1 + fi (x), 1 ≤ i ≤ n − 1,

results in

ẋn = Γ(Q(u)) + fn (x), (5) ∑
n ∑
n

 y=x . βi ( ak ) ≤ nak hi (nak ) (8)
1
k=1 k=1

where x = [x1 , x2 , . . . , xn ]T ∈ Rn represents the state vector, This inequality will be widely applied in the following
y ∈ R represents the system output, u ∈ R represents processes.
the control signal. fi (.) : Rn → R represents an unknown Assumption 2. There are design parameters ϖ > 0 and
smooth nonlinear function. Γ(Q(u)) denotes the system input ϖ̄ > 0, such that
influenced by quantization and actuator dead zone, which can
be designed as follows: 0 < ϖ ≤ kl ≤ ϖ̄, 0 < ϖ ≤ kr ≤ ϖ̄.


 kr (Q(u) − br ), Q(u) ≥ br ,
 Because the existence of quantization and dead zone
Γ(Q(u)) = 0, bl < Q(u) < br , (6) increases the difficulty of controller design, it is necessary to


 k (Q(u) − b ), Q(u) ≤ b . introduce the following Lemma.
l l l

where bl and br stand for the breakpoints of the actuator Lemma 6 (see [46]). Γ(Q(u)) can be broken down as
nonlinearity. The parameters kr represents the right slope of below:
the actuator dead zone and the parameters kl denotes the left Γ(Q(u)) = H̄(u)u + Ḡ(t). (9)
slope of the actuator dead zone. Q(u(t)) denotes the hysteretic,
where
which can be expressed as follows:


 ui sgn(u), ui
< |u| ≤ ui , u̇ < 0, or ϖ(1 − δ) ≤ H̄(u) ≤ ϖ̄(1 + δ),

 1+δ

 ui < |u| ≤ 1−δ
ui
|Ḡ(t)| ≤ ϖ̄ max{|br |, |bl |} + ϖ̄umin .

 , u̇ > 0. (10)



 ui (1 + δ)sgn(u), ui < |u| ≤ 1−δ , u̇ < 0,
ui

ui (1+δ)
1−δ < |u| ≤
Q(u) = ui
 1−δ , u̇ > 0. C. Neural network systems



 0, 0 ≤ |u| < 1+δ , u̇ < 0, or
umin



 Since system (5) contains the unknown function fi (.) :
1+δ ≤ |u| ≤ umin , u̇ > 0.
umin



 − R n
→ R, therefore, radial basis function (RBF) neural
Q(u(t )), othercases.
(7) networks (NNs) will be adopted to approximate fi (.). An
where ui = ρ1−i umin (i = 1, 2, . . .), ρ ∈ (0, 1) represents a expression can be obtained as follows:
measure of quantization density, the parameter umin > 0, and
f (X) = ΦT ξ(X).
δ = 1−ρ
1+ρ . Therefore, Q(u) ∈ U = {0, ±ui , ±ui (1 + δ), i =
1, 2, . . .}. where X ∈ Ωx ⊂ Rq represents the input vector,
Remark 3. What needs to be pointed out is, compared Φ = [ϕ1 , ..., ϕl ]T ∈ Rl denotes the weight vector of
with the existing quantitative research, this paper considers the RBF NNs, l(> 1) denotes the number of nodes, ξ(X) =

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

[ξ1 (X), ..., ξl (X)]T ∈ Rl represents the basis function vector, denotes the basis function vector of RBF NNs. ci > 0, ai > 0,
and ξi (X) as shown below: and σ = 2l−1
2l+1 (l > 2, l ∈ n) represent the design parameters.
(X − vi ) (X − vi )
T The controller can be expressed as follows:
ξi (X) = exp[− ], 1 ≤ i ≤ l.
ζ2 cn 2σ−1 zn θ̂n ξnT (Zn )ξn (Zn )
u=− z − , (14)
where vi = [vi1 , ..., viq ] T
is the center of the receiving field, 1−δ n 2a2n (1 − δ)
and ζ denotes the width of the Gaussian function. where cn > 0 and an > 0 are two parameters.
As shown in [11], f (X) is a continuous function define The adaptive law of θ̂ is chosen as:
on a compact set Ωx . For ∀ε > 0, there exists an RBF NNs qi
˙
Φ∗T ξ(X), makes the following formula holds θ̂i = 2 zi2 ξiT ξi − γi θ̂i , θ̂i (0) ≥ 0, i = 1, 2, . . . , n (15)
2ai

f (X) = Φ∗T ξ(X) + δ(X), ∀X ∈ Ωx . (11) where qi > 0 and γi > 0 denote the design parameters.
Assumption 3. For the desired trajectory yd and its i-th
where δ(X) is the approximation error, Φ∗ represents the ideal (i)
derivative yd , i = 1, 2, ..., n, we have
weight vector, and the inequality |δ(X)| < ε holds. when the
(i)
number of nodes l is large enough, one has |yd | ≤ d, i = 0, 1, 2, ..., n

Φ∗ = arg min { sup |f (X) − ΦT ξ(X)|}. where d ≥ 0 denotes a constant.

Φ∈Rl X∈Ωx
Lemma 8. For zi = xi − αi−1 , i = 1, 2, . . . , n, one can
Lemma 7 (see [11]). Let ξ(X) = [ξ1 (X), ..., ξl (X)]T get the relation as follows:
denotes the basis function vector of an RBF NNs, and X =
∑
n
[x1 , ..., xn ]T denotes the input vector. For ∀m ≤ n, let Xm = ∥x∥ ≤ |zi |ωi (zi , θ̂i ) + d (16)
[x1 , ..., xm ], we have i=1

where ωj (zi , θ̂i ) = 1 + ci zi2σ−2 + 1

θ̂ ξ T (Zi )ξi (Zi ), i
2a2i i i
=
∥ξ(X)∥2 ≤ ∥ξ(Xm )∥2 .
1, 2, . . . , n − 1, ωn (zn , θ̂n ) = 1.
Remark 4. The above lemma provides a simple but useful Proof. Let x = [x1 , . . . , xn ]T , z = [z1 , . . . , zn ]T , and
structural properties of an RBF NNs. It should be noted that in α = [α1 , . . . , αn ]T . Then, one has
the process of backstepping design, lemma 7 is very important ∑
n ∑
n

to deal with the entire state variable. ∥x∥ ≤ |xi | = |zi + αi−1 |
i=1 i=1
∑n ∑
n−1
III. A DAPTIVE FUZZY TRACKING CONTROL DESIGN ≤ |zi | + |αi | + yd (17)
i=1 i=1
In this part, an adaptive controller for the system (5) will
∑n ∑
n−1
1
be designed. In order to start the backstepping design process, ≤ |zi | + |yd | + (ci zi2σ−2 + θ̂i ξiT (Zi )ξi (Zi ))|zi |
i=1 i=1
2a2i
Φi (Xi ) will be employed to approximate the unknown non-
∑
n
linear function, define a constant θi = ∥Φi ∥2 , i = 1, 2, · · · , n, ≤ |zi |ωi (zi , θ̂i ) + d. (18)
where Φi represents the unknown weight vector of RBF NNs. i=1

Define θ̂i as the estimate of θi , θ̃i = θi − θ̂i as an estimation Step 1. For the non-strict feedback system (5), the
error. Then, define the error variables as follows: derivative of z1 is

z1 = x1 − yd , ż1 = ẋ1 − ẏd = x2 + f1 (x) − ẏd . (19)

zi = xi − αi−1 , i = 2, · · · , n. (12) Think about Lyapunov function candidate as follows:
where αi represents the virtual control function, it can be z12 θ̃2
V1 = + 1 , (20)
expressed as follows: 2 2q1
1 where q1 > 0 represents a design parameter.
αi = −ci zi2σ−1 − θ̂i zi ξiT (Zi )ξi (Zi ), (13)
2a2i Differentiating V1 yields
¯ (i)T ¯
where ξi (Zi )(Zi = [x̄Ti , θ̂iT , ȳd ]T with θ̂i = ˙
θ̃1 θ̂1
T
[θ̂1 , θ̂2 , . . . , θ̂i ] ),
(i)
ȳd =
(1) (i)
[yd , yd , . . . , yd ]T , i = 1, 2, . . . , n. V̇1 = z1 (z2 + α1 + f1 (x) − ẏd ) − . (21)
q1

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

According to Lemma 5, Assumption 1, (8) and Lemma By combining Lemma 5 and Lemma 7, the following
8, the following inequality can be obtained: inequality can be obtained:
(∑
n )
z1 f¯1 = z1 (ΦT1 ξ1 (X1 ) + δ1 (X1 ))
z1 f1 ≤ |z1 |β1 (∥x∥) ≤ |z1 |β1 |zl |ωl (zl , θ̂l ) + d
l=1 ≤ |z1 |(∥Φ1 ∥∥ξ1 (X1 )∥ + ε1 )
∑
n ( ) 1 2 T
≤ |z1 |β1 (n + 1)|zl |ωl (zl , θ̂l ) + |z1 |β1 ((n + 1)d) ≤ z θ1 ξ (Z1 )ξ1 (Z1 )
2a21 1 1
l=1
1 1 1
n ∑ n
+ a21 + z12 + ε21 . (27)
≤ z12 + zl2 β̄12 (zl , θ̂l ) + |z1 |β1 ((n + 1)d), (22) 2 2 2
2
l=1
where θ1 = ∥Φ1 ∥2 , Z1 = (x1 , yd , ẏd ), and a1 > 0 denotes a
where design parameter.

1 Furthermore, according to θ̃1 = θ1 − θ̂1 , the following

β̄12 (zl , θ̂l ) = (n + 1)2 ωl2 (zl , θ̂l )h21 ((n + 1)|zl |ωl (zl , θ̂l )). inequality can be obtained:
2
γ1 γ1 2 γ1 2
Then, the following inequality can be obtained by using θ̃1 θ̂1 ≤ − θ̃1 + θ , (28)
q1 2q1 2q1 1
Lemma 3 to the term |z1 |β1 ((n + 1)d) in (22)
Substituting (15), (27)–(28) into (25), the following in-
( z β ((n + 1)d) )
|z1 |β1 ((n + 1)d) ≤ tanh
1 1 equality holds:
ϵ1
γ1 2 ∑ 2 2
n
×z1 β1 ((n + 1)d) + κϵ1 . (23)
V̇1 ≤ −c1 z12σ − θ̃ + zl β̄1 (zl , θ̂l )
2q1 1
l=2
Applying Lemma 5 to terms z1 z2 , one has ∑∑
n−1 k
z22
−z12 β̄j2 (z1 , θ̂1 ) + + λ1 , (29)
z2 z2 2
k=2 j=1
z1 z2 ≤ 1 + 2 (24)
2 2
where
Substituting (13), (22)–(24) into (21), we have a21 ε2 γ1 2
λ1 = κϵ1 + + 1+ θ .
2 2 2q1 1
1 1
V̇1 ≤ − z12 − c1 z12σ − 2 θ̂1 z12 ξ1T (Z1 )ξ1 (Z1 )
2 2a1 Step i(2 ≤ i ≤ n − 1). Let us assume that
θ̃1 θ̂1 ∑ 2 2
˙ n
+z1 f¯1 − + zl β̄1 (zl , θ̂l ) ∑
i−1
1 1 2
q1 Vi−1 = ( zj2 + θ̃j ),
l=2
j=1
2 2q j
∑∑
n−1 k
z22
−z12 β̄j2 (z1 , θ̂1 ) + + κϵ1 , (25) satisfies
2
k=2 j=1
i−1 (
∑ γj 2 ) ∑ ∑ ∑ 2 2
i−1 s n

where V̇i−1 ≤ − cj zj2σ + θ̃j + zl β̄k (zl , θ̂l )

j=1
2qj s=1 k=1 l=i
(n + 2)z1 ∑ ∑∑
i−1 n−1 k
zi2
f¯1 = + z1 β̄12 (z1 , θ̂1 ) − ẏd − zs2 β̄j2 (zs , θ̂s ) + + λi−1 , (30)
2 2
( z β ((n + 1)d) ) s=1 k=i j=1
1 1
+β1 ((n + 1)d) tanh
ϵ1 where
∑∑
n−1 k
+z1 β̄j2 (z1 , θ̂1 ). ∑
i−1
a2j ε2j γj 2
k=2 j=1
λi−1 = (κϵj + + + θj ).
j=1
2 2 2q j

It is clear that f¯1 is a function of β̄j . There exist an RBF And then, let us consider a Lyapunov function candidate
NNs ΦT ξ1 (X1 ), it can be used to approximate f¯1 . Namely,
1 as follows:
for ∀ε1 > 0, 1 1 2
Vi = Vi−1 + zi2 + θ̃ . (31)
2 2qi i
f¯1 = ΦT1 ξ1 (X1 ) + δ1 (X1 ), |δ1 (X1 )| ≤ ε1 . (26) Differentiating zi yields

where X1 = (x1 , . . . , xn , yd , ẏd ). żi = ẋi − α̇i−1 = xi+1 + fi (x) − α̇i−1 . (32)

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

where following inequality can be obtained:

i−1 (
∑
i−1
∂αi−1 ( ∑
i−1 ) ∑ γj 2 )
α̇i−1 = xj+1 + fj (x) +
∂αi−1 ˙
θ̂j V̇i ≤ − cj zj2σ + θ̃ + λi−1
∂xj 2qj j
j=1 ∂ θ̂j
j=1
j=1
1 2 1
∑
i−1
∂αi−1 (j+1)
+ zi+1 − ci zi2σ − zi2 + zi f¯i
+ yd . (33) 2 2
(j)
j=0 ∂yd θ̂i 2 T 1 ˙
− 2 zi ξi (Zi )ξi (Zi ) − θ̃i θ̂i
2ai qi
Differentiating Vi yields ∑i ∑ s ∑ n
+κϵi + zl2 β̄k2 (zl , θ̂l )
s=1 k=1 l=i+1

V̇i = V̇i−1 + zi (zi+1 + αi + fi (x) − α̇i−1 ) ∑

i ∑
n−1 ∑
k

1 ˙ − zs2 β̄j2 (zs , θ̂s ), (39)

− θ̃i θ̂i . (34) s=1 k=i+1 j=1
qi
where
According to Assumption 1, Lemma 5, (8) and Lemma
∑
i−1
∂αi−1 ∑
i−1
∂αi−1 ˙
8, the following inequality can be obtained: f¯i = − xj+1 − θ̂j
j=1
∂xj j=1 ∂ θ̂j

n 2 ∑ 2 2 n ∑ ( ∂αi−1 )2 n + 3
n i−1
zi fi ≤ z + zl β̄i (zl , θ̂l ) + zi + zi
2 i 2 j=1 ∂xj 2
l=1
+|zi |βi ((n + 1)d), (35) (z χ ) ∑∑
n−1 s
i i
∑
i−1 i−1 
∑  +χi tanh + zi β̄j2 (zi , θ̂i )
∂αi−1  ∂αi−1  ϵi
zi fj ≤ |zi |  βj ((n + 1)d) s=1 j=1
∂xj ∂xj
j=1 j=1 ∑
i−1
∂αi−1 (j+1)
i−1 ( −
∑ ∂αi−1 )2
yd .
n (j)
+ zi2 j=0 ∂yd
2 j=1 ∂xj
∑
i−1 ∑
n Similarly, there exist an RBF NNs ΦTi ξi (Xi ), it can be
+ zl2 β̄j2 (zl , θ̂l ). (36) used to approximate f¯i . Namely, for ∀εi > 0, one can get the
j=1 l=1
relation as follows:
1
where β̄j2 (zl , θ̂l ) = 2 (n + 1)2 ωl2 (zl , θ̂l )h2j ((n + f¯i = ΦTi ξi (Xi ) + δi (Xi ), |δi (Xi )| ≤ εi . (40)
1)|zl |ωl (zl , θ̂l )).
(i)
Let where Xi = (x1 , . . . , xn , θ̂1 , . . . , θ̂i , yd , ẏd , . . . , yd ).
By combining Lemma 5 and Lemma 7, which is similar
i−1 
∑ 
 ∂αi−1  to the derivation in (27), the following inequality can be
χi = βi ((n + 1)d) +  βj ((n + 1)d),
j=1
∂xj expressed as:

and using Lemma 3, one can get the relationship as follows: zi f¯i = zi (ΦTi ξi (Xi ) + δi (Xi ))
≤ |zi |(∥Φi ∥∥ξi (Xi )∥ + εi )
(z χ )
i i 1 2 T
|zi |χi ≤ κϵi + zi χi tanh . (37) ≤ z θi ξ (Zi )ξi (Zi )
ϵi 2a2i i i
1 1 1
+ a2i + zi2 + ε2i . (41)
where ϵi > 0 denotes a design parameter. 2 2 2
Meanwhile, applying Lemma 5, the following inequality (2)
where Zi = (x1 , x2 , . . . , xi , θ̂1 , θ̂2 , . . . , θ̂i , yd , ẏd , yd , . . . , yd ),
(i)

can be obtained: θi = ∥Φi ∥2 and ai > 0 denotes a design parameter.

zi2 z2 Furthermore, by θ̃i = θi − θˆi , the following inequality

zi zi+1 ≤ + i+1 . (38) can be obtained:
2 2
γi γi 2 γi 2
θ̃i θ̂i ≤ − θ̃i + θ . (42)
Substituting (13), (30), (33) and (35)–(38) into (34), the qi 2qi 2qi i

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

Substituting (15) and (40)–(42) into (39), one has Similarly, there exist an RBF NNs ΦTn ξn (Xn ), it can be
i (
∑ γj 2 ) used to approximate f¯n . Namely, for ∀εn > 0, one can get
V̇i ≤ − cj zj2σ + θ̃
j=1
2qj j the relation as follows:

∑
i ∑
s ∑
n
f¯n = ΦTn ξn (Xn ) + δn (Xn ), |δn (Xn )| ≤ εn . (51)
+ zl2 β̄k2 (zl , θ̂l )
s=1 k=1 l=i+1
(n)
∑
i ∑ ∑
n−1 k where Xn = (x1 , . . . , xn , θ̂1 , . . . , θ̂n , yd , ẏd , . . . , yd ).
− zs2 β̄j2 (zs , θ̂s ) By combining Lemma 5 and Lemma 7, which is similar
s=1 k=i+1 j=1
2
to the derivation in (27) and (41), the following inequality can
zi+1
+ + λi , (43) be obtained
2
where zn f¯n = zn (ΦTn ξn (Xn ) + δn (Xn ))
∑i
a2j ε2j γj 2
λi = (κϵj + + + θj ). ≤ |zn |(∥Φn ∥∥ξn (Xn )∥ + εn )
j=1
2 2 2q j 1 2
≤ z θn ξnT (Zn )ξn (Zn )
Step n. According to (5), (9), (12), we have 2a2n n
1 1 1
+ a2n + zn2 + ε2n . (52)
żn = H̄(u)u + Ḡ(t) + fn (x) − α̇n−1 , (44) 2 2 2
(n)
where where Zn = (x1 , . . . , xn , θ̂1 , . . . , θ̂n , yd , ẏd , . . . , yd ), θn =
∑ ∂αn−1 ( ) n−1
∑ ∂αn−1 ˙ ∥Φn ∥2 and an > 0 denotes a design parameter.
n−1
α̇n−1 = xj+1 + fj (x) + θ̂j
j=1
∂xj j=1 ∂ θ̂j Applying Lemma 4, (10) and (14), one has
∑
n−1
∂αn−1 1 2
+
(j+1)
yd . (45) zn H̄(u)u ≤ −cn zn2σ − z θ̂n ξnT ξn . (53)
(j) 2a2n n
j=0 ∂yd

Choose the Lyapunov function candidate as follows: Combing (15), (42), and (52)–(53) to (49), we have
1 1 2 ∑n
Vn = Vn−1 + zn2 + θ̃ , (46) γj 2
2 2qn n V̇n ≤ − (cj zj2σ + θ̃j ) + λn , (54)
j=1
2q j
According to (43), the following inequality can be ob-
tained: where
∑(
n−1
γj 2 ) ∑∑
n−1 s
1 1 1 γn 2
V̇n ≤ − cj zj2σ + θ̃ + zn2 β̄k2 (zn , θ̂n ) λn = λn−1 + a2n + ε2n + u2min + θ . (55)
j=1
2qj j s=1 2 2 2 2qn n
k=1
∑∑
n−1 n
1 1 ˙ Theorem 1. Under Assumptions 1-3 and bounded initial
− zs2 β̄k2 (zs , θ̂s ) + zn2 + λn−1 − θ̃n θ̂n
s=1 k=1
2 qn conditions, the non-strict feedback nonlinear quantized system
+zn (H̄(u)u + Ḡ(t) + fn (x) − α̇n−1 ). (47) (5) before the input defined in (6) is considered. The controller
(14) and parameter adaptive law (15) can ensure that all the
Applying Lemma 5 and (10), one has
1 1 signals in the closed-loop system are bounded in probability
zn Ḡ(t) ≤ zn2 + u2min . (48) and the error signal z1 going to stay a compact set, which is
2 2
Substituting (48) into (47), we have defined by
∑
n−1 {  [ ]1/σ 1 }
γj 2  d
V̇n ≤ − (cj zj2σ + θ̃ ) + λn−1 + zn f¯n Ωz = (z, θ)V (t) ≤ , <ϱ<1
2qj j (1 − ϱ)c 2
j=1
1 1 1 ˙
− zn2 + zn H̄(u)u + u2min − θ̃n θ̂n , (49) with c = min{cj , γj , j = 1, 2, . . . , n}, d = λn + (1 −
2 2 qn σ ∑n
σ)σ 1−σ j=1 γj .
where
proof. Applying lemma 5, let ω = 1, τ = σ, v =
∑∑
n−1 s
1 ˜2
f¯n = fn (x) − α̇n−1 + zn β̄k2 (zn , θ̂n ) 2qi θi , γ = 1 − σ, λ = 1
σ, the following relationship can be
s=1 k=1 obtained
∑∑
n−1 n
3
− zs2 β̄k2 (zs , θ̂s ) + zn . (50) (
1 ˜2 σ σ
θi ) ≤ (1 − σ)σ 1−σ +
1 ˜2
θi (56)
s=1 k=1
2 2qi 2qi

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

Combining (54) and (56), one has In order to examine the validity of Theorem 1, let yd =
∑
n ∑
n ( 1 )σ sin(0.5t) + 0.5 sin(1.5t). The corresponding parameters are
V̇n ≤ − cj zj2σ − γj θ̃2 + d
j=1 j=1
2qj j set as δ = 0.5, µmin = 0.2, k1 = kr = 1.2, b1 = br = 0.6.
n (
1 2 )σ
∑n ∑ Consider the following control law and adaption law:
≤ −c zj2σ −c η̃ +d (57)
2qj j 1
j=1 j=1 α1 = −c1 z12σ−1 − θ̂1 z1 ξ1T (Z1 )ξ1 (Z1 ).
2a21
Furthermore, according to Lemma 2, we have
c2 2σ−1 z2 θ̂2 ξ2T (Z2 )ξ2 (Z2 )
(∑
n ( 2 ))σ u = − z − .
z j 1 2 1−δ 2 2a22 (1 − δ)
V̇ ≤ −c + θ̃ +d
j=1
2 2qj j ˙
θ̂i =
qi 2 T
z ξ ξi − γi θ̂i , θ̂i (0) ≥ 0, i = 1, 2.
2a2i i i
≤ −cV + d. σ
(58)
where z1 = x1 − yd , z2 = x2 − α1 .
Based on Bhat and Bernstein, if ∀(z, θ)∈Ωz for ∀t ∈
[ ]1/σ ( ) σ1 In order to construct the basis vector function ξi (Xi ),
[0, tδ ]. We have V ≥ (1−ϱ)c
d
≥ ϱc d
, namely, d ≤
for each input variable, the center of the receptive field is
ϱcV σ for ∀t ∈ [0, tδ ]. Hence, combining (58), for ∀t ∈ [0, tδ ],
taken as v = [−1.5, −1, −0.5, 0, 0.5, 1, 1.5]T and the width of
the following relationship can be obtained: √
Gaussian function is ζ = 2. Select the design parameters
V̇ ≤ −c(1 − ϱ)V σ . (59) as follows: k1 = 8, k2 = 6, a1 = 0.8, a2 = 1, λ1 =
( ) σ1 20, λ2 = 25, γ1 = 1, γ2 = 2. Select the initial condition
combining V ≥ ϱc
d
and (59), we have as [x1 (0), x2 (0)]T = [0.5, −0.3]T , and [θ̂1 (0), θ̂2 (0)]T =
[V (t0 )]1−σ [0.1, 0.2]T . Fig.1-Fig.4 display the corresponding simulation
tδ ≤ . (60)
c(1 − ϱ)(1 − σ) results.
Let
1.5
[V (t0 )]1−σ y
ϑ= , (61) yd
c(1 − ϱ)(1 − σ) 1

then 0.5

[ d ]1/σ
V (t) ≤ , ∀t ≥ ϑ. (62) 0

(1 − ϱ)c
−0.5
In view of the definition of V (t), we come to the
√ [ d ]1/2σ
conclusion that for o = 2 (1−ϱ)c , we have |zi | ≤ o −1

and |y − yd | ≤ o. The proof is thus completed. −1.5

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

Remark 5. Current research status of adaptive quanti-

zation control of nonlinear systems, it can only ensure the Fig. 1: y and yd of Example 1.
stability of infinite time. However, its stability in finite time
can be proved by Therefrom 1, i.e. the tracking performance of
the non-strick feedback system can be achieved when t ≥ ϑ. 1
x2
0.8

0.6

IV. S IMULATION E XAMPLE 0.4

0.2
Example. Think about a non-strict feedback nonlinear
0

system as follows: −0.2

−0.4

ẋ1 = x2 + (1 − sin2 x1 )x2 , −0.6

ẋ2 = Γ(Q(u)) − 3.5x2 + x21 x22 , −0.8

−1
0 2 4 6 8 10 12 14 16 18 20

y = x1 , (63) Time(sec)

where Γ(Q(u)) represents the input of the system, and Q(u) Fig. 2: State variables x2 of Example 1.
is defined in (7).

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

0.4 R EFERENCES
θ̂1
0.3

0.2
[1] F. Wang, B. Chen, X. P. Liu, C. Lin, “Finite-time adaptive fuzzy
0.1 tracking control design for nonlinear systems,” IEEE T. Fuzzy Syst.,
0
0 2 4 6 8 10 12 14 16 18 20
vol. 26, no. 3, pp. 1207-1216, Jun. 2018.
[2] B. Chen, X. P. Liu, S. S. Ge, C. Lin, “Adaptive fuzzy control of a
4
θ̂2 class of nonlinear systems by fuzzy approximation approach,” IEEE T.
3
Fuzzy Syst., vol. 20, no. 6, pp. 1012–1021, 2016.
2
[3] F. Wang, B. Chen, Y. M. Sun, C. Lin, “Finite time control of switched
1
stochastic nonlinear systems,” Fuzzy set syst., vol. 365, pp. 140-152,
0
0 2 4 6 8 10
Time(sec)
12 14 16 18 20 2019.
[4] S. C. Tong, K. K. Sun, S. Sui, “Observer-based adaptive fuzzy
Fig. 3: θ̂1 and θ̂2 of Example 1. decentralized optimal control design for strict feedback nonlinear large-
scale systems,” IEEE T. Fuzzy Syst., vol. 26, no. 2, pp. 569–584, 2018.
[5] C. C. Hua, K. Li, X. P. Guan, “Event-based dynamic output feedback
10
u adaptive fuzzy control for stochastic nonlinear system,” IEEE T. Fuzzy
5
Syst., vol. 26, no. 5, pp. 3004–3015, 2018.
0

−5
[6] Y. M. Li, K. W. Li, S. C. Tong, “Finite-time adaptive fuzzy output feed-
−10 back dynamic surface control for MIMO non-strict feedback systems,”
−15
0 2 4 6 8 10 12 14 16 18 20
IEEE T. Fuzzy Syst., vol. 27, no. 1, pp. 96-110, 2019.
[7] C. C. Hua, L. L. Zhang, X. P. Guan, “Distributed adaptive neural
10

5
Γ(Q(u)) network output tracking of leader- following high-order stochastic non-
0 linear multiagent systems with unknown dead-zone input,” IEEE Trans.
−5

−10
−15
−20
0 2 4 6 8 10
Time(sec)
12 14 16
Fig. 4: u and Γ(Q(u)) of Example 1.
V. C ONCLUSION
In this article, a new finite time control design method has
been addressed for a category of non-strict feedback nonlinear
quantized systems with actuator dead-zone. By using the
relationship between the system input and the control signal,
the problem of nonlinear quantized control is transformed
into a conventional control problem of a nonlinear system
with bounded perturbation and unknown control gain. By em-
ploying the structural properties of RBF NNs, a backstepping
design method is extended from strict-feedback systems to a
category of more common nonlinear systems. By applying the
adaptive neural control based on approximation, an integrated
adaptive controller is constructed, which can ensure that the
system output converges into a small enough neighborhood of
the reference signal in finite-time, and all the signals of the
closed-loop system remain bounded. According to the Bhat
and Bernstein theorem, the finite-time stability of the nonlinear
quantized system is proved. However, it is worth noting that
Cybern., vol. 47, no. 1, pp. 177–185, 2017.
[8] Y. M. Sun, B. Chen, C. Lin, H. H. Wang, “Finite-time adaptive control
18 20 for a class of nonlinear systems with nonstrict feedback structure,” IEEE
Trans. Cybern., vol. 48, no. 10, pp. 2774–2782, 2018.
[9] W. S. Lv, F. Wang, “Adaptive tracking control for a class of uncertain
nonlinear systems with infinite number of actuator failures using neural
networks,” Adv. Differ. Equ., doi: 10. 1186/s13662-017-1426-5(5).
[10] F. Wang, B. Chen, C. Lin, X. Li, “Distributed adaptive neural control
for stochastic nonlinear multiagent systems,” IEEE Trans. Cybern., vol.
47, no. 7, pp. 1795–1804, Jul. 2017.
[11] Y. M. Sun, B. Chen, C. Lin, H. H. Wang, S. W. Zhou, “Adaptive
neural control for a class of stochastic nonlinear systems by backstepping
approach,” Inf. Sci., vol. 369, pp. 748–764, 2016.
[12] Y. M. Li, S. C. Tong, “Adaptive neural networks prescribed perfor-
mance control design for switched interconnected uncertain nonlinear
systems,” IEEE Trans. Neural Netw. Learn. Syst., vol. 29, pp. 3059-
3068, 2018.
[13] S. Tatikonda, S. Mitter, “Control under communication constraints,”
IEEE Trans. Autom. Control, vol. 49, no. 7, pp. 1056–1068, Jul. 2004.
[14] H. J. Gao, T. W. Chen, “A new approach to quantized feedback control
systems,”Automatica, vol. 44, no. 9, pp. 534–542, 2008.
[15] J. Liu, N. Elia, “Quantized feedback stabilization of non-linear affine
systems,”Int. J. Control, vol. 77, no. 3, pp. 239–249, 2004.
[16] W. Ren, J. L. Xiong, “Quantized feedback stabilization of nonlinear
systems with external disturbance,” IEEE T. Automat. Contr., vol. 63,
no. 9, pp. 3167-3172, 2018.
[17] J. Zhou, W. Wang, “Adaptive control of quantized uncertain nonlinear
systems,” Ifac papers., vol. 50, no. 1, pp. 10425-10430, 2017.
[18] W. KhanLin, Y. Lin, S. U. Khan, N. Ullah, “Quantized adaptive
decentralized control for interconnected nonlinear systems with actuator
faults,” Appl. Math. Comput., vol. 320, pp. 175-189, 2018.

how to realize stochastic nonlinear tracking control with finite [19] G. Y. Lai, Z. Liu, Y. Zhang, etc., “Asymmetric actuator backlash
compensation in quantized adaptive control of uncertain networked
time is still a challenging problem, which is a possible future nonlinear systems,” IEEE T. Neur. Net. Lear., vol. 28, no. 2, pp. 294-307,
research topic. 2017.

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access
[20] Z. Liu, F. Wang, Y. Zhang, C. L. Philip, “Fuzzy adaptive quantized
control for a class of stochastic nonlinear uncertain systems,”IEEE
Trans. Cybern., vol. 46, no. 2, pp. 524-534, 2016.
[21] J. Zhou, C. Y. Wen, W. Wang, “Adaptive control of uncertain nonlinear
systems with quantized input signal,” Automatica, vol. 95, pp. 152-192,
2018.
[22] J. Wu , Z. G. Wu, J. Li , G. J. Wang, etc., “Practical adaptive fuzzy
control of nonlinear pure-feedback systems with quantized nonlinearity
input,” IEEE T. Syst. Cybern., vol. 49, no. 3, pp. 638-648, 2019.
[23] F. Wang, Z. Liu, Y. Zhang, C. L. Philip, “Adaptive quantized controller
design via backstepping and stochastic small-gain approach,” IEEE
Trans. Fuzzy Syst., vol. 24, no. 2, pp. 330-343, 2016.
[24] F. Wang, Z. Liu, Y. Zhang, C. L. Philip, “Adaptive fuzzy control for
a class of stochastic pure-feedback nonlinear systems with unknown
hysteresis,” IEEE T, Fuzzy Syst., vol. 24, no. 1, pp. 140-152, 2016.
[25] Z. X. Yu, Y. Dong, S. G. Li, F. F. Li, “Adaptive quantized tracking
control for uncertain switched nonstrict-feedback nonlinear systems with
discrete and distributed time-varying delays.” Int. J. Robust Nonlin., vol.
28, no. 4, pp. 1145-1164, 2018.
[26] F. Wang, B. Chen, C. Lin, G. Li, Y. M. Sun, “Adaptive quantized
control of switched stochastic nonlinear systems,” Neurocomputing, vol.
207, pp. 450-456, 2016.
[27] J. Yang, W. Yang, S. C. Tong, “Decentralized control of switched
nonlinear large-scale systems with actuator dead zone,” Neurocomputing,
vol. 200, pp. 80-87, 2016.
[28] X. H. Su, Z. Liu, G. Y. Lai, “Event-triggered robust adaptive control for
uncertain nonlinear systems preceded by actuator dead-zone,” Nonlinear
Dynam., vol. 93, no. 2, pp. 219-231, 2018.
[29] X. H. Su, Z. Liu, G. Y. Lai, et al., “Direct adaptive compensation
for actuator failures and dead-zone constraints in tracking control of
uncertain nonlinear systems,” Inf. Sci., vol. 417, pp. 328-343, 2017.
[30] Y. X. Li, G. H. Yang, “Adaptive fuzzy decentralized control for a class
of large-scale nonlinear systems with actuator faults and unknown dead
zones,” IEEE T. Syst, Man, Cy-S., vol. 47, no. 5, pp. 729-740, 2017.
[31] W. Yang, S. C. Tong, “Robust stabilization of switched fuzzy systems
with actuator dead zone,” Neurocomputing, vol. 173, pp. 1028-1033,
2016.
[32] G. Y. Lai , Z. Liu , Y. Zhang , “Adaptive fuzzy tracking control of
nonlinear time-delay systems with dead-zone output mechanism based
on a novel smooth model,” IEEE T. Fuzzy Syst., vol. 23, no. 6, pp.
1998–2011, 2015.
[33] S. C. Tong, Y. M. Li, “Adaptive fuzzy output feedback tracking
backstepping control of strict-feedback nonlinear systems with unknown
dead zones,”IEEE T. Fuzzy Syst., vol. 20, no. 1, pp. 168–179, Feb.
2012.
[34] S. C. Tong, T. Wang, Y. M. Li, B. Chen, “A combined backstepping and
stochastic small-gain approach to robust adaptive fuzzy output feedback
control,” IEEE T. Fuzzy Syst., vol. 21, no. 2, pp. 314–327, 2013.
[35] S. C. Tong , Y. M. Li , “Adaptive fuzzy decentralized output feedback
control for nonlinear large-scale systems with unknown dead zone
inputs,” IEEE T. Fuzzy Syst., vol. 21, no. 5, pp. 913–925, 2013.
[36] C. D. Li , J. Q. Yi , G. Q. Zhang , “On the monotonicity of interval
type-2 fuzzy logic systems,” IEEE T. Fuzzy Syst., vol. 22, no. 5, pp.
1197–121, 2014.
[37] Y. M. Li, S. C. Tong, “Hybrid adaptive fuzzy control for uncertain
MIMO nonlinear systems with unknown dead-zones,” Inf. Sci., vol. 328,
pp. 97–114, 2016.
[38] B. Chen, X. P. Liu, K. F. Liu, and C. Lin, “Fuzzy approximation-based
2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
10
adaptive control of nonlinear delayed systems with unknown dead-zone,”
IEEE T. Fuzzy Syst., vol. 22, pp. 237–248, 2014.
[39] H. Q. Wang , X. P. Liu, S. Li, “Robust fuzzy adaptive fault-tolerant
control for a class of non-lower-triangular nonlinear systems with
actuator failures,” Inf. Sci., vol. 336, no. 1, pp. 60-74, 2016.
[40] Y. M. Li, S. C. Tong, and T. S. Li, “Observer-based adaptive fuzzy
tracking control of MIMO stochastic nonlinear systems with unknown
control direction and unknown dead-zones,” IEEE T. Fuzzy Syst., vol.
23, no. 4, pp. 1228–1241, 2015.
[41] Y. M. Li, S. C. Tong, Y. J. Liu, T. S. Li, “Adaptive fuzzy robust
output feedback control of nonlinear systems with unknown dead zones
based on small-gain approach,” IEEE T. Fuzzy Syst., vol. 22, no. 1, pp.
164–176, 2014.
[42] Y. J. Liu , S. C. Tong , “Adaptive neural network tracking control
of uncertain nonlinear discrete-time systems with nonaffine dead-zone
input, IEEE Trans. Cybern., vol. 45, no. 3, pp. 497-505, 2015.
[43] Y. M. Li, S. C. Tong, “Adaptive fuzzy output-feedback control of pure-
feedback uncertain nonlinear systems with unknown dead-zone,” IEEE
T. Fuzzy Syst., vol. 22, no. 5, pp. 1341–1347, 2014.
[44] Q. Zhou, H. Li, C. Wu, L. Wang, C. K. Ahn, “Adaptive fuzzy control
of nonstrict-feedback nonlinear systems with unmodeled dynamics and
input saturation using small-gain approach,” IEEE Trans. Cybern., vol.
47, no. 8, pp. 1979-1989, 2017.
[45] F. Wang , Z. Liu , G. Y. Lai , “Fuzzy adaptive control of nonlinear
uncertain plants with unknown dead zone output,” Fuzzy Sets Syst., vol.
263, no. 1, pp. 27-48, 2015.
[46] F. Wang, Z. Liu, Y. Zhang, C. L. Philip, “Adaptive quantized fuzzy
control of stochastic nonlinear systems with actuator dead-zone,” Inf.
Sci., vol. 370, pp. 385-401, 2016.
[47] C. Tan, X. Yu, Z. Man, “Terminal silding mode observers for a class
of nonlinear system,” Automatic, vol. 46, no. 8, pp. 1401-1404, 2010.
[48] S. P. Bhat, D. S. Bernstein, “Continuous finite-time stabilization of
the translational and rotational double integrators,” IEEE Trans. Autom.
Control, vol. 43, no. 5, pp. 678-682, 1998.
[49] S. P. Bhat, D. S. Bernstein, “Finite-time stability of continuous
autonomous systems,”Siam J. Control Optim., vol. 38, no. 3, pp. 751-
766, 2000.
[50] G. P. Chen, Y. Yang, “Finite-time stability of switched nonlinear
time-varying systems via indefinite Lyapunov functions,”Int. J. Robust
Nonlin., vol. 28, no. 5, pp. 1901-1912, 2018.
[51] R. M. Yang, L. Y. Sun, “Finite-time robust control of a class of non-
linear time-delay systems via lyapunov functional method,”J. Franklin
I., doi: 10.1016/j.jfranklin.2018.08.029.
[52] W. S. Lv, F. Wang, Y. Li, “Finite-time adaptive fuzzy output-feedback
control of MIMO nonlinear systems with hysteresis,” Neurocomputing,
vol. 296, pp. 74-81, 2018.
[53] W. S. Lv, F. Wang, “Finite-time adaptive fuzzy tracking control for a
class of nonlinear systems with unknown hysteresis,” Int. J. Fuzzy Syst.,
vol. 20, no. 3, pp. 782-790, 2018.
[54] W. H. Liu, D. W. C. Ho, S. Y. Xu, B. Y. Zhang, “Adaptive finite-time
stabilization of a class of quantized nonlinearly parameterized systems,”
Int. J. Robust Nonlin, vol. 27, no. 18, pp. 4554-4573, 2017.
[55] L. L. Wang, T. P. Chen, “Finite-time anti-synchronization of neural
networks with time-varying delays,” Neurocomputing, vol. 275, pp.
1595-1600, 2018.
[56] L. H. Zhang, W. H. Qi, Y. G. Kao, X. W. Gao, L. J. Zhao, “New
results on finite-time stabilization for stochastic systems with time-
varying delay,” Int. J. Control. Autom. Syst., vol. 16, no. 2, pp. 649-658,
2018.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2922748, IEEE Access

11

[57] X. Y. Yang, X. D. Li, “Finite-time stability of linear non-

autonomous systems with time-varying delays,” Adv. Differ. Equ-Ny, doi:
10.1186/s13662-018-1557-3.
[58] D. X. Peng, X. D. Li, C. Aouiti, F. Miaadi, “Finite-time synchroniza-
tion for Cohen–Grossberg neural networks with mixed time-delays,”
Neurocomputing, vol. 294, pp. 39-47, 2018.
[59] F. Wang, X. Y. Zhang, “Adaptive finite time control of nonlinear
systems under time-varying actuator failures,” IEEE Trans. Cybern., doi:
10.1109/TSMC.2018.2868329.
[60] W. S. Lv, F. Wang, L. L. Zhang, “Adaptive fuzzy finite-time control
for uncertain nonlinear systems with dead-zone input,” Int. J. Control
Autom. Syst., vol. 16, no. 5, pp. 2549-2558, 2018.
[61] J. H. Liu, C. L. Wang, X. B. Li, X. Cai, “Adaptive finite-time
practical consensus protocols for second-order multiagent systems with
nonsymmetric input dead zone and uncertain dynamics,” J. Franklin I.,
vol. 356, no. 6, pp. 3217-3244, 2019.
[62] Q. Chen, X. M. Ren, J. Na, D. D. Zheng, “Adaptive robust finite-time
neural control of uncertain PMSM servo system with nonlinear dead
zone,” Neural Comput. Appl., vol. 28, no. 12, pp. 3725-3736, 2017.
[63] H. Y. Li, S. Y. Zhao, W. He, R. Q. Lu, “Adaptive finite-time tracking
control of full state constrained nonlinear systems with dead-zone,”
Automatica, vol. 100, pp. 99-107, 2019.
[64] F. Z. Gao, Y. Q. Wu, Y. H. Liu, “Finite-time stabilization for a
class of switched stochastic nonlinear systems with dead-zone input
nonlinearities,” Int. J. Robust Nonlin., vol. 28, no. 9, pp. 3239-3257,
2018.
[65] Z. K. Zhang, G. R. Duan, M. Z. Hou, “Global finite time stabilization
of pure-feedback systems with input dead-zone nonlinearity,” J. Franklin
I., vol. 354, no. 10, pp. 4073-4101, 2017.
[66] F. Wang, B. Chen, C. Lin, et al., “Adaptive neural network finite-time
output feedback control of quantized nonlinear systems,” IEEE Trans.
Cybern., vol. 48, no. 6, pp. 1839–1848, Jun. 2018.
[67] X. Y. Zhang, F. Wang, L. L. Zhang, “Finite time controller design
of nonlinear quantized systems with nonstrict feedback form,” Int. J.
Control Autom. Syst., vol. 17, no. 1, pp. 225-233, 2019.
[68] M. M. Ploycarpou, P. A. Ioannou, “A robust adaptive nonlinear control
design,” Automatica, vol. 32, no. 3, pp. 423–427, 1996.
[69] C. Qian, W. Lin, “Non-Lipschitz continuous stabilizers for nonlinear
systems with uncontrollable unstable linearization,” Syst. Control Lett.,
vol. 42, no. 3, pp. 185–200, 2001.

2169-3536 (c) 2019 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.