New Adaptive and Learning Control System Design for Robotic Manipulators - Research Article

Advances in Mechanical Engineering

2019, Vol. 11(5) 1–11
Ó The Author(s) 2019
Adaptive fuzzy backstepping control DOI: 10.1177/1687814019851309
journals.sagepub.com/home/ade
for uncertain nonlinear systems with
tracking error constraints

Min Wan1 and Qingyou Liu1,2

Abstract
This article deals with the design of adaptive fuzzy backstepping control for uncertain nonlinear systems in strict-
feedback form with tracking error constraints. In this article, a fuzzy system is used to approximate the unknown
nonlinear functions and the differential of virtual control law of each subsystem. In order to satisfy the limitation of track-
ing error constraints, the barrier Lyapunov function is introduced. Moreover, by applying the minimal learning para-
meters technique, the number of online parameters update for each subsystem is reduced to only 1. The control
scheme not only ensures the tracking error is not to transgress the constraint bounds but also solves the problem of
‘‘explosion of complexity’’ and greatly reduces the initial control input and the number of the adaptive parameters; this
provides the conditions for the practical application. The simulation results show the effectiveness of the proposed
method.

Keywords
Uncertain nonlinear system, fuzzy system, backstepping, barrier Lyapunov function, minimal learning parameters

Date received: 3 August 2018; accepted: 3 April 2019

Handling Editor: James Baldwin

Introduction approximate uncertain continuous nonlinear functions.

The advantage of sliding mode controller is that it has
In nonlinear system modeling, due to modeling error, strong robustness to disturbances and unmodeled
unknown physical phenomena (friction in mechanical dynamics, so it has also been widely applied.12–15
system), load variation, and random disturbance, sys- However, all the above methods require that the system
tem uncertainty is unavoidable. Because of the existence meets an important condition, that is, the unknown
of high non-linearity and uncertainty of the nonlinear nonlinearity and the control input appear in the same
system, it is very difficult for the design of the control- equation of the state space model, which is usually
ler.1–4 In the past few decades, the adaptive control regarded as the matching condition.
technology for feedback linearization of nonlinear sys-
tems has made remarkable progress.4–6 The feedback
linearization requires the uncertainty to satisfy the lin-
ear parameterized condition. However, most of the sys- 1
School of Mechatronic Engineering, Southwest Petroleum University,
tems in practice cannot be linearized. Chengdu, P.R. China
LX Wang and JM Mendel7 proposed adaptive fuzzy 2
Xihua University, Chengdu, P.R. China
control using fuzzy system to approximate the
Corresponding author:
unknown control law or the unknown nonlinear func- Min Wan, School of Mechatronic Engineering, Southwest Petroleum
tion. In the literature,8–11 an adaptive neural network University, Chengdu 610500, P.R. China.
controller is designed using neural network to Email: [email protected]

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering

In the actual system, there is a large class of non- (MIMO) or single input single output (SISO) nonlinear
linear systems that do not meet the matching condition, systems with state constraints and time-varying delays.
such as the system of the mechanical hand which is dri- Gao et al.41 described an adaptive neural control for a
ven by the motor. For mismatched uncertainty non- class of nonlinear pure-feedback systems with time-
linear systems, the backstepping control is very effective varying full-state constraints.
and has achieved great success.16–20 Traditional back- Based on the above results, a novel adaptive fuzzy
stepping control needs to do repeated differentiations backstepping control method is studied in this article.
of the virtual control law of the former subsystem. If The main advantages of this proposed control method
there are nonlinear functions in the virtual controllers, are listed as follows:
repeated differentiations will lead to the problem of
‘‘explosion of complexity’’ with the increase in the order 1. An adaptive fuzzy backstepping control design
of the system. This makes high-order systems face great is addressed for a class of strict-feedback non-
difficulties in controller implementation. If the system linear system which is more general for practical
has parameters or structural uncertainties and external applications, in the presence of uncertain non-
disturbances, it will further lead to the difficulty in the linear function, unknown control gain, output
application of backstepping control. constraints, and external disturbance.
Swaroop et al.21 and Zhang and Ge22 proposed a 2. In each subsystem, only one fuzzy system is
dynamic surface control (DSC) method, which can used to approximate the unknown control gain,
avoid the problem of repeated differentiations using n the unknown nonlinear function, and the differ-
first-order low pass filters and has been widely used. ential of the virtual control of the previous sub-
But DSC cannot deal with the uncertainty problem. system. At the same time, compensate for all
Since fuzzy systems and neural networks can approxi- uncertainties and avoid inherent ‘‘explosion of
mate arbitrary nonlinear functions with arbitrary preci- complexity’’ problem. By applying MLPA, the
sion, many literatures have combined them with DSC number of online parameter updates of fuzzy
or backstepping control in recent years.23–30 Liu et al.31 logic system for each subsystem is reduced to
were the first to propose an adaptive backstepping only 1.
finite-time fault-tolerant control for strictly feedback 3. To prevent output from violating the con-
switched nonlinear systems based on neural networks. straints, we employ a BLF. The semi-globally
With the development of adaptive backstepping and uniformly ultimately boundedness (SGUUB) of
DSC design in nonlinear systems, many fuzzy or neural all the signals of the closed-loop system are pro-
adaptive control methods with the minimal-learning- ven. The tracking error converges to an ade-
parameters algorithm (MLPA) have been reported in quately small bound. The effectiveness of the
Yang and colleagues.32,33 Previous works23,34 described proposed control is demonstrated by a simula-
a direct adaptive fuzzy backstepping control with tion example.
MLPA of uncertain nonlinear systems in the presence
of input saturation. This article is organized as follows: problem state-
Output constraints are important constraints for ment and preliminaries is described in section ‘‘Problem
many industrial systems. Ignoring output constraints statement and preliminaries.’’ In section ‘‘Fuzzy system
can lead to performance degradation, hazards, or sys- and its approximation,’’ fuzzy system and its approxi-
tem damage. In order to solve the output constraints mation is presented. Control design and the adaptive
problem, the barrier Lyapunov function (BLF) has law are presented in section ‘‘Adaptive fuzzy backstep-
received extensive attention, because this function ping control design.’’ Stability analysis is proposed in
grows to infinity when its related state is close to a cer- section ‘‘Stability analysis.’’ The simulation results and
tain limit. By maintaining the boundaries of BLF, any conclusion are given in sections ‘‘Simulations’’ and
violation of output constraints can be prevented.33,35–37 ‘‘Conclusion,’’ respectively.
Liu et al.,38 Li and Li,39 Li et al.,40 and Gao et al.41
extended the output constraints to full-state constraints
Problem statement and preliminaries
based on BLFs for various nonlinear systems. Liu
et al.38 described an adaptive control-based BLF for Many mechanical systems in practical engineering have
stochastic nonlinear systems with full-state constraints. the strict-feedback structure (1), such as robotic manip-
Li and colleagues39,40 described approximation-based ulators, autonomous quadrotor helicopters, hydraulic
neural control for multiple input multiple output actuators, and induction motors
Wan and Liu 3

8
> x_ 1 (t) = g1 (x1 (t))x2 (t) + f1 (x1 (t)) + v1 (t)
>
>
> x_ 2 (t) = g2 (x1 (t), x2 (t))x3 (t) + f2 (x1 (t), x2 (t)) + v2 (t)
>
>
< ..
. ð1Þ
>
> x_ (t) = gi (x1 (t), x2 (t), . . . , xi (t))xi + 1 (t) + fi (x1 (t), x2 (t), . . . , xi (t)) + vi (t)
>
>
i
>
> x_ (t) = gn (x1 (t), x2 (t), . . . , xn (t))u(t) + fn (x1 (t), x2 (t), . . . , xn (t)) + vn (t)
: n
y(t) = x1 (t)

where 1 ł i ł n  1. x1 , x2 , . . . , xn are the state vari-

ables. u 2 R, y 2 R are input and output of the system,
respectively. vi (t) is external disturbance with unknown in the set h 2 N and C, M are positive constants then v
bound. gi () and fi () are unknown smooth functions. remains bounded and j(t) 2 Z, 8t 2 ½0, ‘).

Assumption 1. The signs of gi () are known, and there Lemma 2. For all jjj\1, and any positive integer p, the
exist positive constants gim and giM such that following inequality holds
0\gim ł jgi ()j ł giM , i = 1, 2, . . . , n. Without losing
generality, we can assume 0\gim ł gi () ł giM . 1 j2p
log \ ð5Þ
1  j2p 1  j2p

Assumption 2. There exist constants gid .0,

i = 1, 2, . . . , n such that jg_ i ()j ł gid . Lemma 3. The following inequality holds for any 1.0
and any r 2 R
 
Assumption 3. The reference signal yd (t) is a sufficiently r
0 ł jrj  r tanh ł 0:27581 = 10 ð6Þ
smooth function of t, and yd (t), y_ d (t) are known and 1
bounded.

Assumption 4. External disturbance vi (t) is bounded by Fuzzy system and its approximation
the positive unknown constant viM , that is, Fuzzy system with product inference, singleton fuzzi-
jvi (t)j ł viM . fier, and center-average defuzzifier is a universal
approximator. If the fuzzy rule has the following form:
IF x1 is F1j , and x2 is F2j , and . and xn is Fnj , THEN
Control objective. The control objective is to design an
y is Bj (j = 1, 2, . . . , N ), where x = ½x1 , x2 , . . . , xn T 2 Rn
adaptive control scheme such that the output y(t) tracks
is the system input, y represents the output of the sys-
the desired trajectory yd (t) and the tracking error should
tem, Fij (i = 1, 2, . . . , n) and Bj stand for fuzzy sets, N
always remain within given constraints while ensuring
stands for the number of fuzzy rules, the output of
the boundedness of all the closed-loop signals.
fuzzy system can be expressed as

Lemma 1. Let Z :¼ fj 2 R : jjj\1g  R and P

N Q
n
^uj mji (xi )
N = Rl 3 Z  Rl + 1 be open sets. Consider the system j=1 i=1
h_ = H(t, h), where h :¼ ½v, jT 2 N , and y(x) = ð7Þ
N Q
P n
H :¼ R+ 3 N ! R l+1
is piecewise continuous in t and mji (xi )
j=1 i=1
locally lipschitz in h, uniformly in t, on R+ 3 N .
Suppose that there exist functions U :¼ Rl 3 R+ ! R+ where ^uj = max Bj (y), mji (xi ) and Bj (y) denote member-
and V1 :¼ Z ! R+ , continuously differentiable and y2R
ship functions with respect to fuzzy sets Fij and Bj .
positive definite in their respective domains, such that Define fuzzy base functions as
V1 (j) ! ‘ as jjj ! 1 ð2Þ Q
n
mji (xi )
g 1 (kvk) ł U (v, t) ł g 2 (kvk) ð3Þ jj (x) = i=1
, j = 1, 2, . . . , N ð8Þ
N Q
P n
where g1 and g 2 are class K‘ functions. Let mji (xi )
j=1 i=1
V (h) :¼ V1 (j) + U (v), and j(0) 2 Z. If the inequality
holds j(x) = ½j1 (x), j2 (x), . . . , jN (x)T is the fuzzy basis func-
tion vector, and u ^ = ½^u1 , ^u2 , . . . , ^uN T is the weight para-
∂V meter vector, thus the final output of the fuzzy system
V_ = H ł CV + M ð4Þ
∂h equation (7) can be rewritten as
4 Advances in Mechanical Engineering

^T j(x)
y(x) = u ð9Þ Control law and adaptive law design
Step 1. Define ^f1 = (f1  y_ d )=g1 . u
^T j1 (x1 ) is a fuzzy sys-
1
According to the universal approximation theorem
tem for approximating nonlinear function ^f1 . u1 is the
of fuzzy system, if ^f (x) is a continuous function defined optimal parameter vector and
on the compact set O, and using fuzzy system 2 e1 is the minimum
approximation error. Y1 = u1  , Y ^ 1 is the estimate of
^T j(x) to approximate ^f (x), there exist optimal
y(x) = u ~ 1 = Y  Y ^ 1.
Y1 . Moreover, Y 1  
parameter vector u such that sup j^f (x)  u j(x)j ł e, Let D1 = e1 + g11 v1 . Since je1 j ł e1M , g11  ł g1m 1
x2O
and jv1 j ł v1M , there exist an unknown constant D1 .0
for any given small constant e,7 where 0\e ł eM . ^ 1 is the estimate of D1 . Moreover
such that jD1 j ł D1 . D
~  ^
D1 = D1  D1 .
Define je1 j = jy  yd j\K, where K.0 is the tracking
Adaptive fuzzy backstepping control
error constraint.
design Now a BLF can be chosen as
Backstepping control scheme
1 K2 1 ~2 1 ~2
For convenience, symbols xi are introduced, where V1 = log 2 + Y1 + 0 D ð15Þ
2g1 2
K  e1 2g1 2g 1 1
xi = ½x1 , x2 , . . . , xi T 2 Ri , i = 1, 2, . . . , n. The uncertain
nonlinear system (1) can be reconstructed as follows: where g1 .0 and g 01 .0 are the positive design
For subsystem 1: x_ 1 ðtÞ = g1 (x1 )x2 ðtÞ + f1 (x1 ) + v1 (t), parameters.
we can define e1 = y  yd . A virtual control signal a1 is Consider m = (1=(K 2  e21 )). The time derivative of
introduced, and then we can have V1 is equal to
e_ 1 = y_  y_ d = g1 (x1 )e2 + g1 (x1 )a1 + f1 (x1 ) + v1 (t)  y_ d e1 e_ 1 g_ 1 K2 1 ~ ^_
ð10Þ V_ 1 = 2 2
 2
log 2 2
 Y 1 Y1
g1 (K  e1 ) 2g1 K  e1 g 1
where e2 = x2  a1 . 1 ~ ^_
 0 D 1 D1
For subsystem 2, a virtual control signal a2 is intro- g1
duced, and then we can have g_ K2
= me1 (e2 + a1 + ^f 1 + g11 v1 )  12 log 2
2g1 K  e21
e_ 2 = x_ 2  a_ 1 = g2 (x2 )e3 + g2 (x2 )a2  a_ 1 + f2 (x2 ) + v2 1 ~ ^_ 1 ~ ^_
ð11Þ  Y 1 Y 1  0 D1 D1
g1 g1
where e3 = x3  a2 . = me1 (e2 + a1 + u1 T j 1 + e1 + g11 v1 )
For subsystem k, a virtual control signal ak is intro- g_ 1 K2 1 ~ ^_ 1 ~ ^_
 log  Y 1 Y1  0 D 1 D 1
duced, and then we can have 2g1 2 2 2
K  e1 g 1 g1
= me1 (e2 + a1 + u1 T j 1 + D1 )
e_ k = x_ k  a_ k1 = gk (xk )ek + 1 + gk (xk )ak
ð12Þ g_ 1 K2 1 ~ ^_ 1 ~ ^_
 a_ k1 + fk (xk ) + vk  2
log 2
 Y 1 Y1  0 D 1 D 1
2g1 2
K  e1 g 1 g1
where ek + 1 = xk + 1  ak . ð16Þ
Define en = xn  an1 for the final subsystem, and
then we can have By applying Young’s inequality, we have

e_ n = x_ n  a_ n1 = gn (xn )u + fn (xn ) + vn  a_ n1 ð13Þ 1 2 2  T

me1 u1 Tj 1 ł m e 1 Y1 j 1 j 1 + t ð17Þ
4t
As a result, system (1) can be rewritten as the follow-
ing form where t.0 is any given positive constant.
By applying Lemma 2, we have

e_ k = gk (xk )ek + 1 + gk (xk )ak  a_ k1 + fk (xk ) + vk
e_ n = gn (xn )u  a_ n1 + fn (xn ) + vn K2 e2
log 2
\ 2 1 2 , je1 j\K ð18Þ
K2
 e1 K  e 1
ð14Þ

where a0 = yd , 1 ł k ł n  1. Substituting equations (17) and (18) into equation

Equation (14) shows that the control object can be (16) results in
achieved as long as the appropriate virtual control law
ak and control law u are designed.
Wan and Liu 5

 
 0
1 where M1 = (s1 =2g 1 )Y2 0 0 2
1 + (s1 =2g 1 )D1 + D1 d + t.
V_ 1 \me1 e2 + a1 + me1 Y1 jT1 j 1 + D1
4t Step 2. Define ^f2 = (f2  a_ 1 )=g2 . u
^ j2 ðx2 Þ is a fuzzy
T
2
g1d 1 ~ ^_ 1 ~ ^_ system for approximating nonlinear function ^f2 . u2 is
+t+ 2
me21  Y 1 Y1  0 D1 D1
2g1m g1 g1 the optimal parameter vector and e2 is the minimum
   2
1 ^ 1 jT j + D ^ 1 tanh me1 approximation error. Y = u  , Y
2
^ 2 is the estimates
2
ł me1 e2 + a1 + me1 Y 1 1
4t d ~ 2 = Y  Y
of Y2 , Y ^ 2.
me  2
+ jme1 jD1  D1 me1 tanh
1 The Lyapunov function can be chosen as
d
g1d 1 ~ g 1 2 2 T ^_ 1
 1 2 1 ~2
+ 2 me1 + Y1 2
m e1 j 1 j 1  Y V2 = V1 + e2 + Y ð28Þ
2g1m g1 4t 2g2 2g2 2
1 ~ 0 me 
+ 0 D1 (g 1 me1 tanh
1
)D ^_ 1 + t where g 2 .0 is the positive design parameter.
g1 d The time derivative of V2 is equal to
ð19Þ
1 g_ 1 ~ ^_
Choose the first virtual control law a1 , the parameter V_ 2 = V_ 1 + e2 e_ 2  22 e22  Y 2 Y2
g2 2g2 g2
adaptive laws Y ^ 1 and D ^_ 1 as follows
me  = V_ 1 + e2 (e3 + a2 + g 1 (f2  a_ 1 ) + g 1 v2 )
2 2
1 ^ 1 jT j  D^ 1
a1 = l1 e1  me1 Y 1 tanh ð20Þ g_ 1 ~ ^_
4t 1 1
d  22 e22  Y 2 Y2
2g2 g2
^_ 1 = g 1 m2 e2 jT j  s1 Y
Y ^1 ð21Þ
4t 1 1 1 = V_ 1 + e2 (e3 + a2 + ^f 2 + g21 v2 ) ð29Þ
  g_ 2 2 1 ~ ^_
^_ 1 = g 0 me1 tanh me1  s0 D
D ^ ð22Þ  e  Y 2 Y2
1
d 1 1 2g22 2 g 2
where l1 .0, s1 .0, and s01 .0 are positive design = V_ 1 + e2 (e3 + a2 + u T j2 + e2 + g 1 v2 )
2 2
parameters. g_ 1 ~ ^_
 22 e22  Y 2 Y2
Based on Lemma 3, we have 2g2 g2
me 
1
jme1 jD1  D1 me1 tanh ł 0:2758dD1 = d0 D1 ð23Þ By applying Young’s inequality, we have
d
Substituting equations (20)–(23) into equation (19) 1 2  T
e2 u2 T j2 ł e Y j j +t ð30Þ
results in 4t 2 2 2 2
 
gd where t.0 is any given positive constant. Then, equa-
V_ 1 \ l1  12 me21 + me1 e2 + D1 d0 tion (29) becomes
2g1m
ð24Þ
s1 ~ ^ s0 1 ~ ^  
+
g1
Y 1 Y1 + 0 D 1 D 1 + t
g1 _V 2 ł V_ 1 + e2 e3 + a2 + 1 e2 Y j T j 2 + D2
2 2
4t
Completion of Young’ inequality in equations (25) d
g 1 ~ ^_
and (26) + 22 e22  Y 2 Y2 + t
2g2m g2
s1 ~ ^ s1 ~ 2 s1 2  
Y1 Y1 ł Y1 + Y ð25Þ 1 ^ T
g1 2g1 2g 1 1 = V_ 1 + e2 e3 + a2 + e2 Y 2 j2 j2
4t
s0 1 ~ ^ s0 1 ~ 2 s0 1 2 g d
1 ~ g 2 2 T _
^

D1 D 1 ł D + D ð26Þ + 22 e22 + e2 D2 + Y 2 e j j  Y 2 +t
g0 1 2g 0 1 1 2g 0 1 1 2g2m g2 4t 2 2 2
Substituting equations (25) and (26) into equation ð31Þ
(24) leads to  
where D2 = e2 + g21 v2 . Since je2 j ł e2M , g21  ł g2m 1

  
and jv2 j ł v2M , there exist an unknown constant D2 .0
gd s1 ~ 2 ^ 2 is the estimate of D2 . Moreover,
V_ 1 \ l1  12 me21 + me1 e2  Y such that jD2 j ł D2 . D
2g1m 2g 1 1 ~  ^
D2 = D2  D2 .
s0 1 ~ 2 s1 2 s0 1 Choose the virtual control law a2 and parameter
 0 D 1+ Y1 + 0 D2 + D1 d0 + t
2g 1 2g1 2g 1 1 adaptive law Y ^ 2 as follows
 
gd s 1 ~ 2 s0 1 ~ 2
= l1  12 me21 + me1 e2  Y  D + M1
2g1m 2g1 1 2g0 1 1 1 ^ T
a2 = l2 e2  me1  e2 Y 2 j 2 j 2 ð32Þ
ð27Þ 4t
6 Advances in Mechanical Engineering

  k  
^_ 2 = g 2 e2 jT j2  s2 Y
Y ^2 ð33Þ gd X 1 gd
4t 2 2 V_ k ł l1  12 me21  lk   k2 e2i
2g1m i=2
2 2gkm
where l2 .0 and s2 .0 are positive design parameters. Xk Xk
By applying Young’s inequality, we have si ~ 2 s0 1 ~ 2
+ ek ek + 1  Yi + Mi  0 D
i=1
2g i i=1
2g 1 1
1 1
 e22 + je2 D2 j ł D2 ð34Þ ð39Þ
2 2 2
Substituting equations (32)–(34) into equation (31) where Mk = (1=2)D2 2
k + (sk =2g k )Yk + t. T
^
Step n, Define fn = (fn  a_ n1 )=gn . u ^n j n (xn ) is a
results in
fuzzy system for approximating nonlinear function ^fn .
  un is the optimal parameter vector
gd  and
2 en is the mini-
V_ 2 ł V_ 1  l2  22 e22  me1 e2 + e2 e3 mum approximation error. Yn = un  , Y ^ n is the esti-
2g2m  ~  ^
s mates of Yn . Yn = Yn  Yn .  
2
+ je2 D2 j + Y~ 2Y
^2 +t Let Dn = en + gn1 vn . Since jen j ł enM , gn1  ł gnm 1
,
g2 
  and jvn j ł vnM , there exist an unknown constant Dn .0
_ 1 g2d such that jDn j ł Dn . D^ n is the estimate of D . Moreover,
ł V 1  l2   2 e22  me1 e2 n
2 2g2m ~  ^
Dn = Dn  Dn .
1 2 s2 ~ 2 s2 2 The Lyapunov function is chosen as
+ e2 e3 + D 2  Y + Y +t
2 2g 2 2 2g 2 2
  ð35Þ 1 2 1 ~2
1 gd Vn = Vn1 + en + Y ð40Þ
ł V_ 1  l2   22 e22  me1 e2 2gn 2g n n
2 2g2m
s2 ~ 2 where g n .0 is positive design parameter.
+ e2 e3  Y + M2
2g 2 2 The time derivative of Vn is equal to
   
gd 1 gd
ł l1  12 me21  l2   22 e22 1 g_ 1 ~ ^_
2g1m 2 2g2m V_ n = V_ n1 + en e_ n  n2 e2n  Y n Yn
gn 2gn gn
X si
2 X 2
s0 1 ~ 2
+ e2 e3  ~2 +
Y M  D g_ 1 ~ ^_
2g i i
i
2g 0 1 1 = V_ n1 + en (u + ^f n + gn1 vn )  n2 e2n  Y n Yn
i=1 i=1 2gn gn
where M2 = (1=2)D2 2 = V_ n1 + en (u + un T jn + en + gn1 vn )
2 + (s2 =2g 2 )Y2 + t.
Step k(k = 3, . . . , n  1). Define ^fk = (fk  a_ k1 )=gk . g_ n 2 1 ~ ^_
^k T jk is a fuzzy system for approximating nonlinear  en  Y n Yn
u 2gn 2 gn
function ^fk . uk is the optimal parameter vector  
 and2 ^
ek is
_ 1  T
the minimum approximation error. Yk = uk  , Y k is ł V n1 + en u + en Yn jn jn + Dn
~ k = Y  Y ^ k. 4t
the estimates of Yk , Y k   g_ n 2 1 ~ ^_
Let Dk = ek + gk1 vk . Since jek j ł ekM , gk1  ł gkm 1
,  2 en  Yn Yn + t
and jvk j ł vkM , there exist an unknown constant Dk .0  2gn gn
 
such that jDk j ł Dk . D ^ k is the estimate of D . Moreover, 1 ^ T gd
k _
ł V n1 + en u + en Yn jn jn + n2 e2n + en Dn
~k = D  D
D  ^ k. 4t 2gnm
k
Choose the following Lyapunov function 1 ~ g n 2 T _^

+ Yn e j j  Yn + t
gn 4t n n n
1 2 1 ~2
Vk = Vk1 + e + Y ð36Þ ð41Þ
2gk k 2g k k
The virtual control law ak and parameter adaptive Choose the first virtual control law u and parameter
^ k are chosen as ^ n as follows
adaptive law Y
law Y

1 ^ T 1 ^ T
ak = lk ek  ek1  e k Yk j k j k ð37Þ u = ln en  en1  e n Yn j n j n ð42Þ
4t 4t

^_ k = g k e2 j T j  sk Y
Y ^k ð38Þ ^_ n = g n e2 jT jn  sn Y
Y ^n ð43Þ
4t k k k 4t n n
where lk .0, g k .0, and sk .0 are positive design where ln .0 and sn .0 are positive design parameters.
parameters. Substituting equations (42) and (43) into equation
Substituting equations (37) and (38) into equation (41) results in
(36) results in
Wan and Liu 7

 
gnd Select the positive coefficients li as
Vn ł Vn1  ln  2 e2n  en1 en + jen Dn j
_ _
2gnm
ð44Þ
sn ~ ^ g1d
+ Yn Y n + t l1 = a1 + 2
ð50Þ
gn 2g1m

Completion of Young’ inequality in equations (25) 1 gd

li = ai + + i2 , (i = 2, 3, . . . , n) ð51Þ
and (26) 2 2gim
sn ~ ^ sn ~ 2 sn 2
Yn Yn ł Y + Y ð45Þ where a1 and ai are positive constants.
gn 2gn n 2g n n
Substituting equations (50) and (51) into equation
1 1 (49) results in
 e2n + jen Dn j ł D2 ð46Þ
2 2 n
X
n Xn
si ~ 2
Substituting equations (45) and (46) into equation V_ ł a1 me21  ai e2i  Y
(44) results in i=2 i=1
2g i i
 d
 s0 1 ~ 2 Xn
_V n ł V_ n1  ln  1  gn e2  en1 en  sn Y
~2  0 D + Mi
2
2 2gnm n
2g n 2 2g 1 1 i=1

sn 2 1 e21 X
n Xn
si ~ 2
2
+ Yn + D2 +t = a1  ai e i  Y
2gn 2 n 2 2
K  e1 i = 2 i=1
2gi i
  Xn
1 gd sn ~ 2 s0 1
= V_ n1  ln   n2 e2n  en1 en  Y + Mn  ~2 +
D Mi
2 2gnm 2g n n 2g 0 1 1
i=1
  Xn   ð52Þ
gd 1 gd e21 X
n Xn
si ~ 2
ł l1  12 me21  li   i2 e2i ł a1 log  ai e2i  Yi
2g1m i=2
2 2gim K 2  e21 i = 2 2g i
i=1
Xn
si ~ 2 s 0 1 ~ 2 Xn
s0 1 Xn
 Yi  0 D 1 + Mi  ~2 +
D Mi
1
i=1
2g i 2g 1 i=1 2g 0 1 i=1
ð47Þ 1 K2 Xn
1 2
ł 2a1 g1m log 2  2ai gim e
where Mn = (sn =2g n )Y2 2 2g1 K  e21 i = 2 2gi i
n + (1=2)Dn + t.
According to the virtual control laws and control Xn
si ~ 2 s0 1 ~ 2 Xn

laws (20), (32), (37), and (42) and the adaptive para-  Y i  0 D1 + Mi
i=1
2g i 2g 1 i=1
meters laws (21), (22), (33), (38), and (43), not only all
the uncertainties are compensated by fuzzy systems but 0
Pn C = minf2ai gim , s1 , si , i = 1, 2, . . . , ng and
Define
also no repeated differentiation problems exist. Except M = i = 1 Mi .
for the first subsystem, other subsystem has only one So equation (52) can be rearranged as
parameter to learn online. Therefore, if the order of sys-
tem is n, only (n + 1) adaptive learning parameters is V_ ł CV + M ð53Þ
needed to learn online.
where C, M are positive constants. Moreover, based on
Stability analysis the definition of je1 (t)j\K, the initial condition require-

 
ment is K\e1 (0)\K. This is equivalent to e1K(0)\1.
The positive Lyapunov candidate function of the  
 
closed-loop system is considered as Therefore, Lemma 1 ensures that e1K(t)\1, 8t.0 and V
Xn Xn is bounded on ½0, ‘).
1 K2 1 2 1 ~2
V = Vn = log 2 + e i + Y Multiply V_ ł CV + M by eCt on both sides to infer
2g1 2
K  e1 i=2
2gi i=1
2g i i that
1 ~2
+ D ð48Þ
2g0 1 1 eCt V_ ł (  CV + M)eCt ð54Þ
The derivation of V is as follows d
(VeCt ) ł MeCt ð55Þ
  Xn   dt
gd 1 gd
V_ ł l1  12 me21  li   i2 e2i M
2g1m i=2
2 2gim VeCt  V (0) ł (eCt  1) ð56Þ
C
Xn
si ~ 2 s0 1 ~ 2 Xn
 Yi  0 D 1 + Mi ð49Þ M M
2g i 2g 1 0 ł V (t) ł V (0)eCt + (1  eCt ) ł V (0) + ð57Þ
i=1 i=1 C C
8 Advances in Mechanical Engineering

That is, if V (0) ł n, then V (t) ł n + (M=C), 8t.0.

The boundedness of e1 (t)=K and V guarantee that all
signals of the closed-loop control system are semi-
globally uniformly ultimately bounded. From the defi-
nitions of C and M, it is clear that e1 (t) = y(t)  yd (t)
can be made arbitrarily small by appropriate design
parameters.

Simulations
A motor-driven manipulator is used in the simulations,
and the dynamic equation can be written as follows42
8
> x_ 1 = x2
>
>
< x_ = f ðx Þ + g ðx Þx
2 2 2 2 2 3
ð58Þ
>
> x_ 3 = f3 ðx3 Þ + g3 ðx3 Þu + v
>
:
y = x1
Figure 1. The position tracking.
where x1 = u, x2 = u, _ x3 = I, N = mgl + Mgl,
Mt = J + ð1=3Þml2 + ð1=10ÞMl2 D. g is the gravity 8
acceleration constant. f2 and f3 are unknown nonlinear >
> a1 = l1 e1  uT1 j1 (x1 )
>
>
functions. v is the external disturbance. u is connecting >
> T
rod angle. I is electric current. Kt is torque constant. Kb < a2 = l2 e2  e1  u2 j2 (x2 )
>
is back electromotive force (EMF) coefficient. B is vis- u = l3 e3  e2  uT3 j3 (x3 ) ð60Þ
>
>
cous friction coefficient of bearing. D is load diameter. >
> u_ i = gi ei ji (xi )  2si ui , i = 1, 2, 3
>
>
l is connecting rod length. M is load quality. m is con- >
:
e1 = x1  yd , e2 = x2  a1 , e3 = x3  a2
necting rod weight. L is reactance. R is resistance. u is
the control voltage of the motor. J is actuator torque. The control parameters in the two control methods
The desired trajectory is yd = sin t. f2 () are designed as follows: l1 = 3, l2 = 8:5, l3 = 8:5,
= (B=Mt )x2 + (N=Mt ) sin x1 , g2 () = Kt =Mt , f3 () = g 1 = g01 = g 2 = g 3 = 2, s1 = s01 = s2 = s3 = 3, K =
(R=L)x3  (Kb =L)x2 , g3 () = (1=L), v(t) = (4=L) 0:15, and t = 0:5.
sin (t). Choose fuzzy membership functions as
The parameters of the manipulator: B = 0:015, mFil (xi ) = exp½(xi + 2:5  l=2)2 =2, where
L = 0:0008, D = 0:05, R = 0:075, m = 0:01, J = 0:05, l = 1, 2, . . . , 9.
l = 0:6, Kb = 0:085, M = 0:05, Kt = 1, and g = 9:8. Then
The initial state of the manipulator is x(0) = ½0, 0, 0T .
The modified adaptive fuzzy backstepping control mF j (x1 ) 3 mF j (x2 )
1 2
(MAFBC) scheme described above is summed up as j2j (x2 ) = ,
P
9
equation (59). In order to verify the effectiveness, we mF j (x1 ) 3 mF j (x2 )
1 2
compare the control performance with that of conven- j=1

tional adaptive fuzzy backstepping control (CAFBC) mF j (x1 ) 3 mF j (x2 ) 3 mF j (x3 )

1 2 3
without considering the output constraints which is j3j (x3 ) =
P
9
described as equation (60). The number of adaptive mF j (x1 ) 3 mF j (x2 ) 3 mF j (x3 )
1 2 3
j=1
parameters in traditional fuzzy backstepping control is
27, and the number of adaptive parameters in our con-
Define the fuzzy base function vectors as: j 1 (x1 ) =
trol method is only 4
8 me  ½j11 (x1 ), j12 (x1 ), . . . , j19 (x1 )T , j 2 (x2 ) = ½j21 (x2 ), j22 (x2 ),
1
>
> ^ T ^
> a1 = l1 e1  4t me1 Y1 j1 j1  D1 tanh d
>
1
. . . , j29 (x2 )T , and j3 (x3 ) = ½j31 (x2 ), j32 (x3 ), . . . ,
>
>
>
>
> a = l e  me  1 e Y j39 (x3 )T .
>
> ^ T
2 2 j2 j2
>
>
2 2 2 1
4t The simulation results are shown in Figures 1–5.
>
>
>
< 1 ^ T Figures 1 and 2 show the performance of the position
u = l3 e3  e2  e3 Y 3 j3 j3
4t me  tracking of the control methods. From Figure 2, we
>
> _ g
>
> ^ 1 2 2 T ^ ^_ 0 1 0 ^
> Y1 = 4t m e1 j1 j1  s1 Y1 , D1 = g 1 me1 tanh d  s 1 D1
>
> can see that the tracking error provided by MAFBC
>
> _
>
> ^ gi 2 T ^ satisfies the constraints, while the tracking error pro-
>
> Yi = 4t ei ji ji  si Yi , i = 1, 2
>
>
>
: vided by CAFBC violates the error constraints. Figures
e1 = x1  yd , e2 = x2  a1 , e3 = x3  a2
3 and 4 show the performance of the speed tracking.
ð59Þ Figure 5 shows the control input signal. It can be seen
Wan and Liu 9

Figure 2. The error of position tracking. Figure 4. The error of speed tracking.

Figure 3. The speed tracking. Figure 5. The control input.

that the initial control input of MAFBC is much less

Table 1. The performance comparison of the control schemes.
than that of CAFBC.
Finally, for a better illustration, the simulation Performance comparison CAFBC MAFBC
results are summarized in Table 1. From Table 1, it is
clearly shown that the proposed control method The number of adaptive parameters 27 4
MAFBC has much less learning parameters, but its Ð t 2 (je(t)j) of position tracking
max 0.33 0.05
e dt of position tracking 1200 6.8
control performance is better than that of CAFBC. Ð0t 2
0 e dt of speed tracking
1700 150
umax 11.0860 3.9831
Conclusion CAFBC: conventional adaptive fuzzy backstepping control; MAFBC:
modified adaptive fuzzy backstepping control.
In this article, a novel adaptive fuzzy backstepping con-
trol scheme has been proposed for a class of nonlinear
systems in strict-feedback form. Using fuzzy systems complexity’’ is avoided. By introducing the BLF, the
and backstepping control, not only uncertainties, such tracking error satisfies the restriction conditions.
as unknown functions and unknown control gains, are Moreover, in order to reduce the number of adaptive
identified, but also the problem of ‘‘explosion of learning parameters, MLPA is used such that only
10 Advances in Mechanical Engineering

(n + 1) parameter is needed to learning online for n 11. Ge SS and Wang C. Adaptive neural control of uncertain
order system. Finally, the stability is proved and all the MIMO nonlinear systems. IEEE T Neural Netw 2004; 15:
signals are guaranteed to be bounded. Some simulation 674–692.
results have demonstrated the effectiveness of the pro- 12. Tran X and Kang H. Adaptive hybrid high-order termi-
posed control scheme. Future research directions are nal sliding mode control of MIMO uncertain nonlinear
systems and its application to robot manipulators. Int J
the extension of the results to nonstrict-feedback sto-
Precis Eng Manuf 2015; 16: 255–266.
chastic MIMO nonlinear systems with uncertainties 13. Sun L and Zheng Z. Finite-time sliding mode trajectory
and unmeasured states. tracking control of uncertain mechanical systems. Asian
J Control 2017; 19: 399–404.
Declaration of conflicting interests 14. Cong B, Liu X and Chen Z. Improved adaptive sliding
mode control for a class of second-order mechanical sys-
The author(s) declared no potential conflicts of interest with
tems. Asian J Control 2013; 15: 1862–1866.
respect to the research, authorship, and/or publication of this
15. Hao L, Park JH and Ye D. Fuzzy logic systems-based
article.
integral sliding mode fault-tolerant control for a class of
uncertain non-linear systems. IET Control Theor Appl
Funding 2016; 10: 300–311.
16. Cai J, Wen C, Su H, et al. Adaptive backstepping control
The author(s) disclosed receipt of the following financial sup-
for a class of nonlinear systems with non-triangular struc-
port for the research, authorship, and/or publication of this
tural uncertainties. IEEE T Autom Control 2017; 62:
article: This work was supported by National Natural Science
5220–5226.
Foundation of China (Grant No. 51775463).
17. Coban R. Dynamical adaptive integral backstepping vari-
able structure controller design for uncertain systems and
ORCID iD experimental application. Int J Robust Nonlin Control
Min Wan https://orcid.org/0000-0001-6123-6883 2017; 27: 4522–4540.
18. Ahn KK, Nam DNC and Jin M. Adaptive backstepping
control of an electrohydraulic actuator. IEEE/ASME T
References Mechatronics 2014; 19: 987–995.
1. Sakthivel R, Santra S and Mathiyalagan K. Reliable 19. Davila J. Exact tracking using backstepping control
robust control design for uncertain mechanical system. J design and high-order sliding modes. IEEE T Autom
Dyn Syst Meas Control 2014; 137: 021003. Control 2013; 58: 2077–2081.
2. Jin M, Lee J, Chang HP, et al. Practical nonsingular ter- 20. Tong S, Li Y, Li Y, et al. Observer-based adaptive fuzzy
minal sliding-mode control of robot manipulators for backstepping control for a class of stochastic nonlinear
high-accuracy tracking control. IEEE T Ind Electron strict-feedback systems. IEEE T Syst Man Cy B Cybern
2009; 56: 3593–3601. 2011; 41: 1693–1704.
3. Xiao B, Yin S and Kaynak O. Tracking control of 21. Swaroop D, Hedrick JK, Yip PP, et al. Dynamic surface
robotic manipulators with uncertain kinematics and control for a class of nonlinear systems. IEEE T Autom
dynamics. IEEE T Ind Electron 2016; 63: 6439–6449. Control 2000; 45: 1893–1899.
4. An H, Zhou L, Wei X, et al. Nonlinear analysis of 22. Zhang TP and Ge SS. Adaptive dynamic surface control
dynamic stability for the thin cylindrical shells of super- of nonlinear systems with unknown dead zone in pure
cavitating vehicles. Adv Mech Eng 2017; 9: 1–15. feedback form. Automatica 2008; 44: 1895–1903.
5. Tall IA. Feedback linearizable feedforward systems: a 23. Edalati L, Khaki Sedigh A, Aliyari Shooredeli M, et
special class. IEEE T Autom Control 2010; 55: al. Adaptive fuzzy dynamic surface control of nonlinear
1736–1742. systems with input saturation and time-varying
6. Semprun KA, Yan L, Butt WA, et al. Dynamic surface output constraints. Mech Syst Signal Pr 2018; 100:
control for a class of nonlinear feedback linearizable sys- 311–329.
tems with actuator failures. IEEE T Neur Netw Learn 24. Chen XW, Zhang JG, Zhou Y, et al. Fuzzy control for
Syst 2017; 28: 2209–2214. vehicle status estimation considering roll stability and its
7. Wang LX and Mendel JM. Fuzzy basis functions, uni- application in target recognition of automobile cruise sys-
versal approximation, and orthogonal least-squares tem. Adv Mech Eng 2017; 9: 1–12.
learning. IEEE T Neural Netw 1992; 3: 807–814. 25. Liu Y. Adaptive dynamic surface asymptotic tracking for
8. Wang H, Shi P, Li H, et al. Adaptive neural tracking a class of uncertain nonlinear systems. Int J Robust Non-
control for a class of nonlinear systems with dynamic lin Control 2018; 28: 1233–1245.
uncertainties. IEEE T Cybernetics 2017; 47: 3075–3087. 26. Chang W, Tong S and Li Y. Adaptive fuzzy backstep-
9. He W, Dong Y and Sun C. Adaptive neural impedance ping output constraint control of flexible manipulator
control of a robotic manipulator with input saturation. with actuator saturation. Neural Comput Appl 2017; 28:
IEEE T Syst Man Cy Syst 2016; 46: 334–344. 1165–1175.
10. Pan Y, Sun T, Liu Y, et al. Composite learning from 27. Liu W, Lim C, Shi P, et al. Backstepping fuzzy adaptive
adaptive backstepping neural network control. Neural control for a class of quantized nonlinear systems. IEEE
Netw 2017; 95: 134–142. T Fuzzy Syst 2017; 25: 1090–1101.
Wan and Liu 11

28. Yin S, Shi P and Yang H. Adaptive fuzzy control of state constraints and unmodeled dynamics. Automatica
strict-feedback nonlinear time-delay systems with unmo- 2017; 81: 232–239.
deled dynamics. IEEE T Cybernetics 2016; 46: 1926–1938. 36. Liu Y and Tong S. Barrier Lyapunov functions for Nuss-
29. Zong Q, Wang F, Tian B, et al. Robust adaptive dynamic baum gain adaptive control of full state constrained non-
surface control design for a flexible air-breathing hyper- linear systems. Automatica 2017; 76: 143–152.
sonic vehicle with input constraints and uncertainty. Non- 37. He W, David AO, Yin Z, et al. Neural network control
linear Dynam 2014; 78: 289–315. of a robotic manipulator with input deadzone and output
30. Wang X, Li H, Zong G, et al. Adaptive fuzzy tracking constraint. IEEE T Syst Man Cy Syst 2016; 46: 759–770.
control for a class of high-order switched uncertain non- 38. Liu YJ, Lu S, Tong S, et al. Adaptive control-based bar-
linear systems. J Frankl Inst 2017; 354: 6567–6587. rier Lyapunov functions for a class of stochastic non-
31. Liu L, Liu Y and Tong S. Neural networks-based adap- linear systems with full state constraints. Automatica
tive finite-time fault-tolerant control for a class of strict- 2018; 87: 83–93.
feedback switched nonlinear systems. IEEE T Cybernetics 39. Li DP and Li DJ. Adaptive neural tracking control for
2019; 49: 2536–2545. an uncertain state constrained robotic manipulator with
32. Yang Y, Feng G and Ren J. A combined backstepping unknown time-varying delays. IEEE T Syst Man Cy Syst
and small-gain approach to robust adaptive fuzzy control 2018; 48: 2219–2228.
for strict-feedback nonlinear systems. IEEE T Syst Man 40. Li DP, Li DJ, Liu YJ, et al. Approximation-based adap-
Cy A Syst Hum 2004; 34: 406–420. tive neural tracking control of nonlinear MIMO
33. Yang Y and Zhou C. Adaptive fuzzy H stabilization for unknown time-varying delay systems with full state con-
strict-feedback canonical nonlinear systems via backstep- straints. IEEE T Cybernetics 2017; 47: 3100–3109.
ping and small-gain approach. IEEE T Fuzzy Syst 2005; 41. Gao T, Liu YJ, Liu L, et al. Adaptive neural network-
13: 104–114. based control for a class of nonlinear pure-feedback sys-
34. Li Y, Tong S and Li T. Direct adaptive fuzzy backstep- tems with time-varying full state constraints. IEEE/CAA
ping control of uncertain nonlinear systems in the pres- J Automat Sin 2018; 5: 923–933.
ence of input saturation. Neural Comput Appl 2013; 23: 42. Wan M and Liu Q. An improved adaptive fuzzy back-
1207–1216. stepping control for nonlinear mechanical systems with
35. Zhang T, Xia M and Yi Y. Adaptive neural dynamic sur- mismatched uncertainties. Automatika 2019; 60: 1–10.
face control of strict-feedback nonlinear systems with full