Accepted Manuscript

Anti-oscillation and chaos control of the fractional-order brushless DC

motor system via adaptive echo state networks

Shaohua Luo , Shaobo Li , Farid Tajaddodianfar , Jianjun Hu

PII: S0016-0032(18)30460-5
DOI: 10.1016/j.jfranklin.2018.07.004
Reference: FI 3546

To appear in: Journal of the Franklin Institute

Received date: 21 January 2018

Revised date: 13 May 2018
Accepted date: 10 July 2018

Please cite this article as: Shaohua Luo , Shaobo Li , Farid Tajaddodianfar , Jianjun Hu , Anti-
oscillation and chaos control of the fractional-order brushless DC motor system via adaptive echo
state networks, Journal of the Franklin Institute (2018), doi: 10.1016/j.jfranklin.2018.07.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Anti-oscillation and chaos control of the fractional-order

brushless DC motor system via adaptive echo state networks
Shaohua Luoa, Shaobo Lia,b, Farid Tajaddodianfarc, Jianjun Hud
a
Key Laboratory of Advanced Manufacturing Technology, Ministry of Education, Guizhou University, Guiyang 550025,
China
b
Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China
c
Department of Mechanical Engineering, Erik Jonsson School of Engineering and Computer Science, The University of
Texas at Dallas, Richardson TX 75080, USA
d
Department of Computer Science and Engineering, University of South Carolina, Columbia SC 29201, USA

T
Abstract—This paper proposes anti-oscillation and chaos control scheme for the fractional-order brushless

IP
DC motor system wherein there exist unknown dynamics, immeasurable states and chaotic oscillation.
Aimed at immeasurable states, the high-gain observers with fast convergence are presented to obtain the
information of system states. To compensate uncertainties existing in the dynamic system, a finite-time echo

CR
state network with a weight is proposed to approximate uncertain dynamics while its weight is tuned by a
fractional-order adaptive law online. Meanwhile a fractional-order filter is introduced to deal with the
repeated derivative of the backstepping. Based on the fractional-order Lyapunov stability criterion, the
anti-oscillation and chaos control scheme integrated with a high-gain observer, an echo state network and a

US
filter are proposed by using recursive steps of backstepping. The proposed scheme guarantees the
boundedness of all signals of the closed-loop system in the sense of global asymptotic stability, and also
suppresses chaotic oscillation. Finally, the effectiveness of our scheme is demonstrated by simulation results.
Keywords—Fractional-order brushless DC motor system; Adaptive chaos control; Chaotic oscillation;
AN
Echo state network; High-gain observer

1.Introduction
Due to wide applications, fractional calculus has become an active area of research in recent years [1-6]. It
M

gives a better representation of the memory and hereditary properties by comparing with classical
integer-order models in the field of the physics and engineering [7-10]. There exist primarily three definitions
of fractional calculus such as Grünwald-Letnikov definition, Riemann-Liouville definition and Caputo
definition. Among them Caputo definition is widely applied in real world applications. Wu et al. [11] defined
ED

the Riesz Riemann-Liouville difference wherein space variables are defined on discrete domains. By
application of the Grünwald-Letnikov definition, Özdemir et al. [12] presented an eigenfunction expansion
method for fractional optimal control of 2-dimensional distributed systems. For the fractional-order
Hastings-Powell food chain model, Matouk et al. [13] investigated its dynamical behavior and proposed a
PT

new discretization method by using the Caputo definition. Viewed from the modeling and analysis of
dynamic systems, it is meaningful to consider the fractional-order in both research and practice.
Considering advantages such as high torque per weight, long lifetime, lower noise and high reliability, the
CE

brushless DC motor system has been widely used in the industrial automation and manufacturing engineering
[14-16]. A new study found that chaotic oscillation of the brushless DC motor system triggered intermittent
oscillation, irregular current noise and instability of control performance. Although chaos is beneficial in
many application fields, it can destroy the stability and safety of the brushless DC motor system if measures
are not taken. Therefore, chaos and oscillation controls become one of the key research topics in the motor
AC

system. The Ott, Grebogi and Yorke (OGY) method is deemed to be efficient for controlling chaotic
oscillation [17-19]. But finding an adjustable parameter is not easy in the practice. Ge and Cheng [20]
investigated the chaos synchronization and parameter identification of three-time-scale brushless DC motor
system. A sliding mode control (SMC) scheme of single time-scale brushless DC motor system is proposed
to suppress chaotic oscillation [21]. Based on the differential-geometry control theory, a current time-delayed
feedback method can induce the brushless DC motor system operating in a stable regime [22]. These works
mainly focus on chaotic oscillation control of the integer-order brushless DC motor system. In the field of
fractional calculus, Narmada and Arounas-salame [23] investigated the design and performance evaluation of
the fractional-order controller for the brushless DC motor system. Rajagopal et al. [16] studied the dynamic
properties and chaotic oscillation control of the fractional-order brushless DC motor system. Zhou et al. [14]
proposed two control approaches via a single input to stabilize the fractional-order brushless DC motor

1
 ACCEPTED MANUSCRIPT

system based on the generalized Gronwall inequality and Mittag-Leffler function. However, all the works
above depend heavily on precise mathematic model without uncertainties, as well as do not involve
immeasurable states limited by the cost and characteristics of physical sensors.
Jaeger et al. [24] discussed the optimization and application of echo state networks with leaky-integrator
neurons. Han and Lee [25] investigated the precise positioning of dynamic systems with nonsmooth
nonlinearities by using fuzzy wavelet echo state networks and dynamic surface sliding mode control.
Venayagamoorthy [26] applied echo state networks to online design of a wide-area monitor for a
multi-machine power system. Know then, the echo state network is an effective tool to compensate system
uncertainties. In engineering applications, state variables are immeasurable for nonlinear systems under the
influence of internal and external conditions. So an observer is an effective resolution for this issue [27]. By
using the measurements of stator currents and voltages, a sliding mode observer is designed to estimate the

T
phase-to-phase trapezoidal back-EMF in the brushless DC motor [28]. Ren et al. [29] proposed a fuzzy
adaptive high-gain-based observer backstepping control scheme for SISO nonlinear systems with dynamical

IP
uncertainties. The backstepping control is widely applied in controlling uncertain integer-order nonlinear
systems [30, 31]. Its controller is established by the „intermediate virtual controls‟ in a recursive way.
However, this approach faces with the problem of „explosion of complexity‟. To tackle this problem,

CR
first-order filters [32], tracking differentiators [33], neural networks [34] or fuzzy systems [35, 36] are
introduced in controller design. Unfortunately, these works are at the phase of preliminary application in the
field of fractional-order dynamic systems. Meanwhile they are unavailable for the fractional-order brushless
DC motor system because the complicated process of fractional derivative. Liu et al. [37] designed an

US
adaptive fuzzy backstepping controller for a class of uncertain fractional-order nonlinear systems by
application of the Caputo definition. However, this scheme enhances complexity and difficulty in course of
dSPACE real-time simulation system and physical controller without considering virtual control data. To our
best knowledge, there are sporadic articles about chaos control of the fractional-order brushless DC motor
AN
system.
Motivated by the literature above, this work presents an anti-oscillation and chaos control scheme for the
fractional-order brushless DC motor system characterized by unknown dynamics, parameter variation,
immeasurable states and chaotic oscillation. To explain clearly the contributions of this work, the following
list is worked out:
M

1) The fractional-order high-gain observers are designed to accurately estimate immeasurable states
online. It can eliminate the effect from the cost and characteristics of physical sensors.
2) An echo state network with a weight and an adaptive law is used to compensate uncertainties. It can
lower the requirement of precise mathematic model and accurate system parameters. It also can
ED

benefit fast solving and reduce the computation load.

3) To handle the explosion of complexity in the backstepping, a fractional-order filter is introduced to
tackle this problem.
4) Integrating echo state networks, filter, backstepping and high-gain observer into anti-oscillation and
PT

chaos controllers in the domain of fractional calculus not only guarantees the boundedness of all
signals, but also suppresses chaotic oscillation.
The remaining parts of this work are organized as follow. The fractional-order brushless DC motor system
CE

and preliminaries are presented in Section 2. The anti-oscillation and chaos controllers are given in Section 3.
Simulation results are shown in Section 4. Finally, Section 5 concludes this work.

2.Fractional-order brushless DC motor system and preliminaries

AC

An integer-order brushless DC motor system is shown in Fig.1. It belongs to one type of electromechanical
systems. The coupled equations between electrical and mechanical dynamics are given as [20, 38, 39]
d

 dt J
 
 t kt iq   Ld  Lq  iq id   b  TL 
n 1
J
 diq
   Riq  nt  Ld id  kt   vq 
1
 (1)
 dt Lq
 di
 d 
 dt
1
L
  Rid  nt Lq iq  vd 
d

2
 ACCEPTED MANUSCRIPT

where id , iq , , nt are the direct-axis current (A), quadrature- axis current (A), rotor angular speed (rad/s)
and number of permanent pole pairs, respectively, vd denotes the direct-axis voltage (V), vq is the
quadrature-axis voltage (V), J , R , Ld , Lq denote the moment of inertia (kgm2), winding resistance (Ω),
direct-axis fictitious inductance (H) and quadrature-axis fictitious inductance (H), respectively, kt  3 / 2ke ,
b , ke , TL represents for the permanent magnet flux (Wb), viscous damping coefficient (N/rad/s), permanent
magnet flux constant and additional term, respectively.
Define time constants 1 , 2 and 3 in the forms of 1  JR / kt2 , 2  Lq / R and 3  Ld / R . Then the
previous Eq.(1) for the brushless DC motor system becomes

T
 dx1
 1 dt  x2  x2 x3  x1  TL

IP

 dx2
 2   x2  x1  x1 x3  uq (2)
 dt

CR
 dx3
 3  x1 x2  x3  ud
 dt
where non-dimensional variables are given as
nt Lq Lq Lq Lq Lq Rb
x1  iq , x3  vq , ud  

TL 
R

k
nt Lq
2
, x2 

TL ,
kt

Ld

Lq
kt
id , uq 

US
kt R kt R
vd ,  nt2 ,  (1  )nt2 ,

, TL is the normalized load torque, uq and ud denote the normalized quadrature-axis

kt2
,
AN
t

and direct-axis stator voltages.

M
ED
PT
CE

Fig.1. The brushless DC motor system and its commutation.

AC

3
 ACCEPTED MANUSCRIPT

60

40

20

x1
0

-20

-40
5
5

T
0
0
-5 -5
x2 x3

IP
Fig.2. Chaotic attractor of the brushless DC motor system.
The brushless DC motor system appears chaotic oscillation when system parameters are set as 1 1 ,

CR
2  6.45 , 3  7.125 , u q  4.017 , ud  15.305 , TL  2.678 ,  16 ,  1.516 and  1.6 [20, 38, 39].
All initial values of state variables are set to 0. Fig.2 shows the chaotic attractor of the brushless DC motor
system with system parameters falling into a certain area.
Considering the rotor angular speed, quadrature-axis and direct-axis currents are of the fractional-order 1 ,

 C D0,t x1  x2  x2 x3  x1 


1

1 1 1
TL US
the coupled system (2) is decomposed into two associated sub-systems as

1
AN
 D 1x   x  x  x x  1 u
1 1 1
(3)
 C 0,t 2 2
2
2
1
2
1 3
2
q

 1 1 1
 C D0,t 1 x3  x1 x2  x3  ud

M

3 3 3

Fig.3 shows that different fractional-orders 1  0.95 , 0.9 , 0.8 of the brushless DC motor system can
change the running state from chaos to order under the same conditions. Obviously, the system has chaotic
characteristic. Fig.4 illustrates the phase diagram for different . It can be concluded that the chaotic
ED

oscillation of the brushless DC motor system is very sensitive to any parameter variation like parameter .
To facilitate the controller design, the Caputo definition which is one of the most frequently used
definitions for fractional calculus is introduced here. The Caputo fractional derivative is given by [40]
 h   s 
PT

n
1 t
 
  n  1  0 t  s  1
 n 1
ds, n  1  1  n
C D0, t h  t    , (4)
1

 n
 n h t  , 1  n
CE

 t
dt is the Gamma function, n is an integer, h   t  is the n -order derivative of

where  n  1   0 et t n  1 1
n

the function h  t  .
AC

4
 ACCEPTED MANUSCRIPT

 1=0.95  1=0.9

50 20

0 0

x1

x1
-50 -20
5 5 5 5
0 0 0 0
-5 -5 x2 -5 -5 (b) x
x2 (a) x3  1=0.8 3

20

T
0
1
x
-20

IP
5 5
0 0
-5 -5 (c) x3
x2

CR
Fig.3. Chaotic attractor of the brushless DC motor system for different fractional-order.
For any continuous function h  t  , the following equality exists [40]

C D0,t h2 t   2h t C D0,t h t  .
1 1
(5)

C
US
For h  0   0 and 0  p, q  1 , the additive law of exponents in the field of the fractional calculus exists
[40]
D0,pt  C D0,qt h  t   C D0,qt  C D0,pt h  t   C D0,pt  q h  t  . (6)
AN
Lemma 1 [41]: The nonlinear fractional-order system C D0,t I  t   O  t  with 0   1 , I t   and
O t   can be transformed into the following continuous frequency distributed model
z  , t  
  z  , t   O t  , I t     z ,t d , (7)
t
M

sin   and z
where ,    ,t  denote the frequency, weighting function and true state of the
system, respectively.
ED

4 4

3 =3.0 3 =2.36
x2

x2

2 2
PT

1 1

0 0
-10 0 10 20 -10 0 10 20
CE

(a) x1 (b) x1

6 10
=2.1 =1.6
4 5
AC
x2

2 0
x

0 -5

-2 -10
-20 0 20 40 -50 0 50
(c) x1 (d) x1

Fig.4. Phase diagram with 1  0.95 .

5
 ACCEPTED MANUSCRIPT

3.Anti-oscillation and chaos controllers

3.1.Echo state networks
Echo state networks can approximate any given real continuous function on a compact set as one of
universal approximators. The equations of a continuous leaky integrator echo state networks are defined as
[24, 25]

 X  aX  win u  wX  w fbY
  
 , (8)
Y  G w0 X
T
 
where denotes the time constant, u is the external input, a represents for the leaking decay rate, G   and

T
  are the output activation function and Gaussian basis function, X , Y and w0 denote the reservoir

IP
activation state, output vector and output weight matrices, respectively, win , w and w fb are the input, internal
and feedback connection weight matrices, respectively.
t    X1 ,
T
The Gaussian basis functions about echo state networks for , X  with being the node

CR
number of output layer are comprised by
  X i  i 2 
 
i  X   exp    , i  1, , , (9)

 i 

where i , i  1,

t  
, are the center and

T
t   X  , where
i , i  1, ,
US are the width of the Gaussian basis function.
The estimation value fˆesn  X  of an unknown nonlinear function f esn  X  is expressed as
fˆesn  X , *
AN
denotes the weight vector. Define the optimal weight vector
 
*
 arg min  sup f esn  X   fˆesn  X , , (10)

 X DX 
where and DX denote the compact set of suitable bounds of  t  and X .
M

For any given constant  0 , there exists echo state network such that
sup f esn  X   fˆesn  X , *   . (11)
X DX
ED

Define a parameter estimation error as t   * t   t  and an optional approximation error as

t   f  X   fˆ  X , *
 , where  0 is the upper bound of  t  .
PT

3.2.High-gain observer design

Carry out the matrix transform for (3)
 C D0 ,t 1 x  Ax  Ky  f  x   Bu


 y  Cx
CE

 (12)
 C D0,t 1
x3  f3  x   ud / 3

x   k / 1 1 
T
0   f1  x     x1 / 1  x2 x3 / 1  TL / 1 
where x   1  , A   1  ,  , u    , f  x    = ,
AC

C  
 x2   k2 0  0  uq   f 2  x      x1  x2  / 2  x1 x3 / 2 
 k1  0 0 
K  ,B  , f3  x   x1 x2 / 3  x3 / 3 , x   x1 , x2 , x3  and ki , i = 1,2 , denote the coefficients
 k2  1 / 2 0
of the Hurwitz polynomials as p  s   s  k1 s  k2 .
2

Note A is a Hurwitz matrix. For any given matrix QT  Q  0 , there is a matrix PT  P  0 such that
AT P  PA  2Q . (13)
The information of variables like x1 , x2 and x3 are not available because of constraints from physical
conditions, so it is necessary to design high-gain observers to estimate them

6
 ACCEPTED MANUSCRIPT

 C D0,t 1 xˆ  Axˆ  lK  y  yˆ   Kyˆ  fˆ  xˆ ,   Bu



 yˆ  Cxˆ
 (14)
 D 1 xˆ  k l  x  xˆ   fˆ  x̂,   u /
C 0 ,t 3 3 3 3 3 d 3

where fˆ  xˆ ,    fˆ1  xˆ1 , 1  fˆ2  xˆ1 , xˆ2 , 2   

T
, l  diag l , l 2 , l is a constant gain, fˆ3  xˆ ,   fˆ3  xˆ3 , 3  ,
xˆ   xˆ1 , xˆ 2 , xˆ3  , k3  0 , xˆ   xˆ1 , xˆ2  .
T

To benefit fast solving, a transform related the number of weight vector for echo state networks is
employed as [33]

T
1 1
i t  i  x  i t  i  x i  x 
T T
, (15)
2 2

IP
where i  t  = iT  t  i  t  , i  1,2,3 .
One has

CR
f  x   fˆ  xˆ ,   f  x   fˆ  xˆ , *   fˆ  xˆ , *   fˆ  xˆ , 
1 1
 t  T  xˆ   xˆ   E   t  ,
 (16)
2 2

where
1 1
f 3  x   fˆ3  xˆ ,   3  t  3  xˆ3  3  xˆ3    3  t  ,
2
T

2
t    1 t  , 2 t  ,  xˆ   diag  1  xˆ  , 2  xˆ  ,
US E is an identity matrix and  t    1  t  , 2  t 
T
.
AN
Introduce observation errors ei  xi  xˆi , i  1,2,3 . Define scaled observation errors i  ei / l i 1 , i  1,2 and
3  e3
. From (12), (14) and (16), it has

C D0,t 1
 lA  F ,
M

1 1 (17)
C D0,t 1
3   k3 l 3  3  t  3T  xˆ3   xˆ3    3 t  ,
2 2
1 1 
 2 1  t  1  xˆ1   xˆ1   2  1  t 
T

ED

 
where F   ,   1 .
  1 1    2
 2 2  t  2  x   x   2  2  t   / l 
T
ˆ ˆ
 
Consider the fractional-order Lyapunov stability criterion
PT

1 T 1
V0  P  32 . (18)
2 2
Then its fractional derivative is computed as
CE

1 1 
C D0 , t V0  diag 1, l  P   t  T  xˆ   xˆ   E   t  
1 T

 2 2  (19)
1 1
 3 3  t  3T  xˆ3   xˆ3   3  3 3  t   l T Q  k3l 2
.
AC

3
2 2
It has

7
 ACCEPTED MANUSCRIPT

 T 1 2 1 1 2 1 2
 P t   2 , 3 3 t  
2 2
 P 3  3 ,
2 2 2

 1 T PE  1 2  1 P 2 , 1  1  1 2 ,
2 4 4 2
3
4 4
3


 t  T T
 
2 1
P  xˆ   xˆ   PPT   t   xˆ   xˆ 
2
T T
 (20)
 2 2 2
 1
max  P  t  ,
2 2
  2

 2 2
1 1 1

T
 3 3  t  3  xˆ3   xˆ3   3  3 t  ,
T 2 2

2 4 4
where  0 ,  xˆ   xˆ   1 , 3  xˆ3   xˆ3   1 , max  P  denotes the largest eigenvalue of P .

IP
T T

From (13) and (20) , (19) is rewritten

 t   P     32  32  t    32  k3l 32 ,

CR
2 l 2 l 2 1 2 1 1 1
C D0 ,t V0  cl  (21)
1

2 2 2  4 4 2

max  P  , min 
Q  is the minimal eigenvalue of Q .
3
where c  min  Q    2

4 2
3.3.Controller design

V1  t  
1  1 
US
Step 1: Define q1  t   xˆ1  t  and select the fractional-order Lyapunov stability criterion

  z12  ,t  d    z2  , t d  V0 ,
2 0 2 1 0
AN
1 1 1
(22)

sin  , sin  .
where 1 0, 1
  1
1
1
  1
1

 t   t   xˆ   t   k l
M

One has C D0,t q1  t  

1
xˆ2  T
1
T
1 1 1 1 1 1 . From Lemma 1, the frequency
1
distributed model is obtained
 z1  , t 
 
ED

   z1  , t   s1 xˆ2  1T  t   1T  t  1  xˆ1    t   k1l ,


1 1
 t (23)
q  t      z  , t  d ,
 1 0 1 1
where s1  / 1 is uncertain, but the upper bound s1 is known.
PT

After defining t   1 t   ˆ1 t  , it is obtained

1

0,t 1  t   C D0,t 1  t  .
ˆ
CD (24)
1 1
CE

Then it follows from the equivalent frequency distributed model that

 z 1  , t 

   z 1  , t  C D0,t1 ˆ1  t  ,
  t (25)
AC

 

 1  t   0   z 1  ,t d .
 1

The fractional-order derivative of V1  t  is derived as

8
 ACCEPTED MANUSCRIPT

 1 
V1  t     1
  z12  ,t  d   1
  z2  1
, t d
0 0
1

 ˆ t  q1  t  
  s1  q2  t   v 2  t    1 1  x 1  x 1
T
 k1l 1  q1  t  (26)
 2 2 


1
1
1 ˆ

1  t  C D0, t 1  t   1 q1  t  1  x  1  x   C D0, t V0 
T 1
1 2
2
1 t . 
where q2  t   xˆ2  t   v2 t  , v2  t  denotes the virtual control.
Choose the virtual control and parameter adaptive law as

T
1  1 ˆ t  
v2 t      c1   q1  t   1  x  1  x   1  k1l ,
1 T
(27)
 

IP
1
s1   2 2 
D 1 ˆ t  
C 0, t 1 1 
q t   x   x   m ˆ t  ,
T
1 1 1 1 1  (28)

CR
where c1  0 , m1  0 .
Substituting (27) and (28) into (26) yields
 1 
V1  t     1
  z12  ,t  d   1
  z2  1
, t d  s1q2  t  q1  t 
0 0

 c1q12  t   m1
Step 2: Consider the Lyapunov function as
1 2
2
1 t  1 t  
1  t   C D0 , t V0 .
1 ˆ
1

US (29)
AN
1  1 
V2  t      z22  , t  d    z 22  , t d  V1  t  ,
2 2 0 2
(30)
2 0 2

sin  , sin  .
where 2 0, 2
  2
2
2
  2
2
M

By calculating the fractional derivative of q2  t  , one has

C D0,t1 q2  t    t   t   xˆ , xˆ   t   u
T
2
T
2 2 1 2 2 q / 2  k2 l 2 C D0,t1 v2 t  . (31)
ED

To avoid the repeated derivative for (27), a fractional-order filter is employed to solve this issue. Then 2f of
a fractional-order filter is able to replace C D0,t 1
v2  t  as
r2C D0,t 1
2f  2f  v2 , 2f 0  v2  0 , (32)
PT

where r2 is a time constant.

It gets C D0,t 1 2 t   C D0,t ˆ2 t  . Then the corresponding frequency distributed models satisfy
1

 z2  , t 
   z2  , t   2T  t   2T  t     xˆ , xˆ    k2 l   t   uq /
CE

,

2 1 2 2f 2 2 2
 t
q  t   
 2 0 2   z2  , t  d ;
 (33)
 z 2  , t    z
 , t  C D0,t1 ˆ2  t  ,
AC

 t 2

 
 2 t  
 0 2   z 2  , t  d .
From (30) and (33), the derivative of V2  t  is deduced

9
 ACCEPTED MANUSCRIPT

1 1 
V2  t     ˆ2  t  T
2  x 2  x 1 2f  q2  t   u q / 2  s1q1  t   k 2l 2  q2  t 
2 2 
2  2 
1
 i
  zi2  ,t  d    i
  z2  i
, t d  c1q12  t   m1 1  t  ˆ1  t  (34)
0 0
i 1 i 1 i
2
1 1 2
 2  t   C D0,t ˆ2  t  
1 T
2 2
1
 x  2  x  q2  t   
i  t   C D0 , t V0 .
2 i 1 2

Choose the control input of the quadrature-axis and corresponding adaptation law as
  1 ˆ t  
uq  2    c2   q2  t   2 2  x  2  x   1  2 f  s1q1  t   k 2 l 2  ,
T
(35)

T
  2 2 
 
1 ˆ
t   
 x   x  q t   m t  ,
ˆ 

IP
T
D C 0, t 2 2 2 2 2 2 2 (36)
where c2  0 , m2  0 .

CR
Substituting (35) into (34) results in
2  2  2
1
V2  t     i
  zi2  ,t  d    i
  z2  i
, t d   ci qi2  t 
0 0
i 1 i 1 i i 1
2 2
(37)
1
  mi i  t  ˆi  t     t  C D0,t V0 .
US
2 1
i
i 1 i 1 2

Step 3: Similar to the procedure of Step 1, define q3  t   xˆ3  t  and choose the Lyapunov function as
1  1 
V3  t     z32  ,t  d    z2  , t d  V2  t  ,
2 0 
AN
(38)
3
2 3 0 3 3

sin  , sin  .
where 3 0, 3
  3
3
3
  3
3

Its time derivative of q3  t  is given by

M

C D0,t1 q3  t    t   t   xˆ   t   u
T
3
T
3 3 3 3 d / 3  k3l 3 . (39)
Then, the related frequency distributed models are
ED

 z3  , t 

t
  z3  , t   3T  t   3T  t     xˆ    t   k l
3 3 3 3 3  ud / 3 ,

q  t   
 3 0 3   z3  , t  d ;
 (40)
PT

 z 3  , t    z
  , t  C D0,t1 ˆ3  t  ,
t 3

 
 3 t     z 3  ,t d .
 
CE

0 3

Using (40), it can obtain the derivative of V3  t 

 1 
V3  t     3
  z32  ,t  d   3
  z2  3
, t d
0 0
AC

1 1  1
 q3  t   ˆ3  t  3T  x  3  x   1  q3  t   k3l 3  ud / 3  2 3
2
t  (41)
 2 2 
1
2  t   C D0, t 3  t   3 2  x  2  x  q3  t    V2  t  .
 1 ˆ 
 T

3
Through transformation, the control input of the direct-axis with an update law is obtained as
  1 1 
ud  3    c3   q3  t   ˆ3  t  3T  x  3  x   1  k3l 3  , (42)
  2 2 

10
 ACCEPTED MANUSCRIPT

C D0,t1 ˆ3  t   3  2q  t 
3 3  xˆ3   m3 ˆ3  t   , (43)
where c3  0 , m3  0 .
3 3 3
1 1
There exists  mi i  t  ˆi  t    mi i2 t    mi i2 t  by the completion of squares. Simplifying (41)
i 1 2 2
i 1 i 1
yields
3  3 
1
V3  t     i
  zi2  ,t  d    i
 z 2  i
,t  d
0 0
i 1 i 1 i
(44)
3
1 3 1 3 3
1 2
  c q  t    mi i  t    mi i  t    i  t  C D0,t V0 .
2 2 2 1

T
i i
i 1 2 i 1 2 i 1 i 1 2
Theorem 1: Aimed at the anti-oscillation and chaos control issues for the fractional-order brushless DC

IP
motor system (3) with unknown dynamics, immeasurable states and chaotic oscillation, an adaptive echo
state network control scheme which is composed of controllers ((35), (42)) with parameters adaptive laws
((36), (43)), a fractional-order filter (32) and high-gain observers (14) is proposed. This scheme can

CR
guarantee that all signals in the closed-loop fractional-order brushless DC motor system are bounded, and the
related chaotic oscillation is completely suppressed.
Proof. Define the overall Lyapunov function
3 3
1  1  1 1
V t       zi2  , t  d   0 i  z 2i  , t  d  T P  32 . (45)
i 1 2

Its derivative is given as

V  t    
2
0


i
i 1 2 i

  zi2  , t  d   
2
1 
2

3
US
 z 2i  , t  d   ci qi2  t 
2
AN
i 0 i 0
i 1 i 1 i i 1
3 3 3
1 1 1
  mi  t    mi t     t   cl   k3l  1
2 2 2 2 2
i i i 3
2 i 1 2 i 1 i 1 2
(46)
1 21  l 1 1 1
t  t   
2 2
 l P     2 2
M

3 3
2 2  2 4 4 2
3
1 l 
  ci qi2  t    mmin    t   cl   k3l  1
2
 ,
2 2
3
i 1 2 2 
ED

1 21 2 1 2 1 3 1 1
where  l P      mi i2  t   3
2
 , mmin  mi , i  1,2,3 .
2  2  2 2 i 1 2 4
 l
Let  min 2q \  P  , 2ci , mi , i  1, 2, 3, mmin   , then (46) becomes
PT

min
 
V t    V t   , (47)
which implies that
CE

0  V t    V (0)e t . (48)

It can be concluded that from (48), signals like xi , xˆi , qi , ˆi  t  and i are bounded. The proof of Theorem 1
AC

is finished.
Remark 1: It should be mentioned that the traditional backstepping refers to “explosion of complexity”
caused by the repeated differentiation of virtual control function. To deal with this problem, a
fractional-order low-pass filter is employed by selecting a reasonable constant.
Remark 2: There are three fractional update laws such as (28), (36) and (43). The right side of all update laws
has two terms. The first one is used to eliminate the term from the derivative of fractional Lyapunov function.
The second one is designed to guarantee the parameter boundedness.
Remark 3: It should be pointed out that a transform related the number of weight vector for echo state
networks is presented. This means the proposed scheme just needs a weight of echo state networks. Then it
benefits fast solving and lower the level of difficulty in controller design.

11
 ACCEPTED MANUSCRIPT

4.Simulation results
To demonstrate the feasibility and effectiveness of the anti-oscillation and chaos control scheme for the
fractional-order brushless DC motor system, simulation results are provided here. The designed controllers
are switched by using the stiff/ TR-BDF2 solver at 20s .The parameters of high-gain observers are selected as
k1  30 , k2  10 , k3  30 , l  2 .

25
0.5
20
0
15 -0.5
-1

T
10
-1.5
5
x ,hat(x )

-2
1

IP
0 19 20 21 22
1

-5

CR
-10 x1

-15 hat(x1)

-20
-25
0 10 20

US 30
t/s
Fig.5. Time courses for system states ( x1,xˆ1 ) with  1.6 and

5
40 50 60

1  0.95 .
AN
4 x2

3 hat(x2)

2
M
x2,hat(x2)

0
0.2
-1
ED

0.1
-2 0
-0.1
-3
20 21 22
-4
PT

0 10 20 30 40 50 60
t/s
Fig.6. Time courses for system states ( x2 , xˆ2 ) with  1.6 and 1  0.95 .
The time constant r2 related to the fractional-order filter is set to 0.02. The parameters of the proposed
CE

controllers in this study are chosen as c1  8 , c2  8 , c3  10 ,  2 , 2  2 , 3  2 , m1  1 , m2  1 ,

1

m3  1 . In addition, the center i is uniformly distributed in the field of [-5,5] and the width i is equal to 2 in
echo state networks. Finally, all initial values of states variables ( xi , i  1, 2,3 ) are set to zero.
AC

Figs.5-7 show time courses of x1  xˆ1 , x2  xˆ2 , x3  xˆ3 under the condition of  1.6 and 1  0.95 . The
high-gain observers can make observation states xˆi approximate real states xi , i  1, 2,3 with tiny observation
errors no matter what the motion of the fractional-order brushless DC motor system is chaotic or normal. Fig.
8 depicts the observation error e2 with different values of parameter . Before the designed controllers are
applied, parameter variation can produce the fluctuation of performance curve. When the fractional-order
brushless DC motor system is under control at 20s, four performance curves are in mutual coincidence. It can
be seen that observation states and real states are bounded, and chaotic oscillation of system is completely
suppressed after control.

12
 ACCEPTED MANUSCRIPT

2
x3
1
hat(x3)

x ,hat(x )
-1 -0.2
3 -0.4
-2
3

-0.6
-0.8
-3

T
-1
-1.2
-4
18 20 22 24 26

IP
-5
0 10 20 30 40 50 60
t/s

CR
Fig.7. Time courses for system states ( x3 , xˆ3 ) with  1.6 and 1  0.95 .

1
=1.6
=2.1

US
0.5
=2.36
=3.0
0
AN
2

-0.5 0.04
e

0.02
-1
0

-1.5 -0.02
M

20 20.5 21 21.5

-2
0 10 20 30 40 50 60
t/s
Fig.8. e2 for different in the case of  0.95 .
ED

Fig.9 presents the control input uq of the fractional-order brushless DC motor system for different . The
value of uq is equal to zero without control. Twenty seconds later the control input uq goes to zero at a very
rapid rate after small fluctuation. It can be seen that the proposed scheme has a rapid convergence.
PT

It can be concluded that the parameter perturbation has almost no effect on the anti-oscillation and chaos
control of the fractional-order brushless DC motor system. Fig.10 shows the curve of q2 for different
fractional-orders with  1.6 . One sees that different fractional-orders produce a great change of q2 in front
CE

of twenty seconds. After 20s, q2 converges to zero, and three curves coincide each other. The asymptotical
convergence can be guaranteed. Despite the fractional-order brushless DC motor system has the character of
unknown dynamics, immeasurable states and chaotic oscillation, the purpose for anti-oscillation and chaos
control is achieved here.
AC

13
 ACCEPTED MANUSCRIPT

400
=1.6
200
=2.1
0 =2.36
=3.0
-200

uq -400
0
-600

-800 -20

-1000

T
-40
-1200
19.5 20 20.5 21 21.5

IP
-1400
0 10 20 30 40 50 60
t/s
Fig.9. uq for different in the case of  0.95 .

CR
1

15

10 =0.95

5
US =0.9
=0.8
2

0
q

AN
0.5
-5
0

-10 -0.5
M

20 21 22
-15
0 10 20 30 40 50 60
t/s
Fig.10. q2 for different fractional-orders in the case of  1.6 .
ED

5 5
=1.6 =1.6
4.5 4.5
=2.1 =2.1
4 =2.36 4 =2.36
PT

3.5 =3.0 3.5 =3.0

hat(f (·))

hat(f (·))

3 3 0.58
1

0.7 0.56
2.5 2.5
0.6 0.54
CE

2 2
0.5 0.52
1.5 1.5 0.5
0.4
1 1
18 20 22 20.5 21 21.5
0.5 0.5
0 10 20 30 40 50 60 0 10 20 30 40 50 60
AC

t/s t/s
Fig.11 the approximation ability of echo state networks.
Fig.11 shows the approximation ability of the echo state network for different . If the proposed scheme
gets involved in the fractional-order brushless DC motor system, the estimate values of unknown nonlinear
functions fˆi  , i  1,3 approximates the true values of f i   , i  1, 3 in a short time.
To show the advantage of the proposed scheme, a comparison is given under the condition of ignoring the
immeasurable states. According to the Ref. [37], the adaptive fuzzy backstepping controller (AFBC) for the
fractional-order brushless DC motor system is established as
 
v 2 f  t   k11q1 f  t   k12 sign q1 f  t   1 f  t  1 f  x  ,
T
(49)

14
 ACCEPTED MANUSCRIPT

C D0,t1 1f t   1f q1 f  t  1f  x   m1 f 1f t  , (50)
uqf  k21q2 f  t   k22 sign  q2 f  t    T
2f t  2f  x  q1 f t  , (51)

C D0,t  t   2 f q2 f  t  2 f  x   m2 f 2 f  t  ,
1
2f (52)
udf  k31q3 f  t   k32 sign  q3 f t    3Tf  t  3 f  x  , (53)

C D0,t1 ˆ3 t   3f q3 f t  3f  x  m3 f 3f t  , (54)
where kij  0, i  1  3, j  1  2 , mif  0, i  1  3 , if  0, i  1  3 , q1 f  t   x1  t  , q2 f  t   x2  t   v2 f t  ,
q3 f  t   x3  t  , if  x  , i  1  3 represent for the fuzzy basis function, if  t  , i  1  3 denote the vector

T
 t    t  , t  ,  t 
T
wherein if if 1 if 2 , ifN .

IP
Firstly, it is well know that the fuzzy logic system is one of universal approximators for unknown smooth
function. Compared with if  t  in the Ref. [37], the proposed scheme just needs a weight of the echo state

CR
network by using formula transformation. So it can benefit fast solving and reduce the computation load.
Secondly, all virtual control items in the controller design are approximated by the fuzzy logic system in the
Ref. [37]. This method is very effective under the theory and MATLAB environment. However, it gives rise
to enhance complexity and difficulty in the process of dSPACE real-time simulation system and physical
controller because of ignoring direct and useful data. Finally, as can be seen in the Fig. 11, the performance of
the proposed scheme is superior to the AFBC‟s because of high convergence rate and small amplitude.

0.6
US 0.4
the proposed scheme
AN
the proposed scheme 0.35
0.5 AFBC AFBC
0.3
0.4
0.25
0.18
0.3 0.2 0
q2

q3

0.17
M

0.2 0.15
-0.02
0.16 0.1
0.1
0.15 0.05 -0.04
0.5 1 1.5 2
0
0 0.5 1 0
ED

-0.1 -0.05
0 5 10 15 0 5 10 15
t/s t/s

40 60
the proposed scheme the proposed scheme
PT

20 AFBC 40 AFBC

0
20
-20 4
CE
ud

uq

2 0
-40
0 0
-20
-60
-2 -5
-80 -40
-4 -10
AC

0 0.2 0.4 0.6 0.8

-100 0 0.5 1 1.5 2
0 5 10 15 -60
0 5 10 15
t/s t/s

Fig.11 the performance comparisons between the proposed scheme and AFBC.

5.Conclusion
An anti-oscillation and chaos control scheme for the fractional-order brushless DC motor system is
reported in the presence of unknown dynamics, immeasurable states and chaotic oscillation. Considering the
fractional-order of the rotor angular speed, quadrature-axis and direct-axis currents, the math model of the
fractional-order brushless DC motor system is established. The presented high-gain observers can deal with
the issues like immeasurable states and obtain the information of system states controlled by the cost and

15
 ACCEPTED MANUSCRIPT

characteristics of physical sensors. An echo state network with a weight can compensate uncertainties
existing in the dynamic system with arbitrary precision. A fractional-order filter better solves the problem of
repeated derivative associated with backstepping. By applying the fractional-order Lyapunov stability
criterion into controller design, the anti-oscillation and chaos control scheme fusing with the high-gain
observer, echo state network and filter are presented in the recursive steps of backstepping. The requirements
of precise model and exact measurement of system parameters which are necessary in previous literature can
be ignored in our scheme. Finally, the favorable anti-oscillation and chaos control performances are validated
based on the comprehensive simulation results.
Acknowledgments
This work is funded by the Major Research Plan of National Natural Science Foundation of China (No.

T
91746116), National Natural Science Foundation of China (Nos. 51505170, 51475097 and 51505045),
Research Project of Introduction of Talents of Guizhou University (No. [2017]27), Major Project of Basic

IP
Research of Guizhou Province (No. [2014]2001), Key Scientific Research Program of Guizhou Province
(No. [2017]3001) and High-level Innovative Talent Program of Guizhou Province (No. [2015]4011).

CR
References
[1] E.A. Boroujeni, H.R. Momeni, Non-fragile nonlinear fractional order observer design for a class of nonlinear fractional
order systems, Signal Process., 92 (10) (2012) 2365–2370.
[2] Y. Sun, Y. Wang, X. Wu, Y. Hu, Robust extended fractional Kalman filter for nonlinear fractional system with missing

US
measurements, J. Frankl. Inst., 355 (1) (2018) 361-380.
[3] S. Das, V.K. Yadav, Chaos control and function projective synchronization of fractional-order systems through the
backstepping method, Theor. Math. Phys., 189 (1) (2016) 1430-1439.
[4] S. Luo, S. Li, F. Tajaddodianfar, Adaptive chaos control of the fractional-order arch MEMS resonator, Nonlinear Dyn., 91
(1) (2018) 539-547.
[5] H. Liu, S. Li, G. Li, H. Wang, Adaptive controller design for a class of uncertain fractional-order nonlinear systems: an
AN
adaptive fuzzy approach, Int. J. Fuzzy Syst., (9) (2017) 1-14.
[6] S. Luo, S. Li, F. Tajaddodianfar, J. Hu, Adaptive synchronization of the fractional-order chaotic arch
micro-electro-mechanical system via Chebyshev neural network, IEEE Sens. J., 18 (9) (2018) 3524-3532.
[7] I. Petras, Fractional-order memristor-based Chua's circuit, IEEE Trans. Circuits Syst. II, 57 (12) (2010) 975-979.
[8] D. Sheng, Y. Wei, S. Cheng, J. Shuai, Adaptive backstepping control for fractional order systems with input saturation, J.
M

Frankl. Inst., 354 (5) (2017) 2245-2268.

[9] N. Bigdeli, H.A. Ziazi, Finite-time fractional-order adaptive intelligent backstepping sliding mode control of uncertain
fractional-order chaotic systems, J. Frankl. Inst., 354 (1) (2017) 160–183.
[10] M.P. Aghababa, A switching fractional calculus-based controller for normal non-linear dynamical systems, Nonlinear
Dyn., 75 (3) (2014) 577-588.
ED

[11] G.C. Wu, D. Baleanu, H.P. Xie, Riesz Riemann-Liouville difference on discrete domains, Chaos, 26 (8) (2016) 084308.
[12] N. Özdemir, O.P. Agrawal, B.B. İskender, D. Karadeniz, Fractional optimal control of a 2-dimensional distributed system
using eigenfunctions, Nonlinear Dyn., 55 (3) (2008) 251-260.
[13] A.E. Matouk, A.A. Elsadany, E. Ahmed, H.N. Agiza, Dynamical behavior of fractional-order Hastings–Powell food chain
model and its discretization, Commun. Nonlinear Sci. Numer. Simul., 27 (1–3) (2015) 153-167.
[14] P. Zhou, R.J. Bai, J.M. Zheng, Stabilization of a fractional-order chaotic brushless DC motor via a single input, Nonlinear
PT

Dyn., 82 (1-2) (2015) 519-525.

[15] F. Zhang, D. Lin, M. Xiao, H. Li, Dynamical behaviors of the chaotic brushless DC motors model, Complexity, 21 (4)
(2016) 79-85.
[16] K. Rajagopal, S. Vaidhyanathan, A. Karthikeyan, P. Duraisamy, Dynamic analysis and chaos suppression in a fractional
CE

order brushless DC motor, Electr. Eng., 99 (2) (2017) 721-733.

[17] A.S. de Paula, M.A. Savi, A multiparameter chaos control method based on OGY approach, Chaos Solitons Fractals, 40
(3) (2009) 1376-1390.
[18] H. Gritli, S. Belghith, N. Khraief, OGY-based control of chaos in semi-passive dynamic walking of a torso-driven biped
robot, Nonlinear Dyn., 79 (2) (2015) 1363-1384.
[19] E. Nobakhti, A. KhakiSedigh, N. Vasegh, Control of multichaotic systems using the extended OGY method, Int. J.
AC

Bifurc. Chaos, 25 (08) (2015) 1550096.

[20] Z.-M. Ge, J.-W. Cheng, Chaos synchronization and parameter identification of three time scales brushless DC motor
system, Chaos Solitons Fractals, 24 (2) (2005) 597-616.
[21] Y. Uyaroglu, B. Cevher, Chaos control of single time-scale brushless DC motor with sliding mode control method, Turk.
J. Elect. Eng. Comput. Sci., 21 (3) (2013) 649-655.
[22] C.L. Li, W. Li, F.D. Li, Chaos induced in brushless DC motor via current time-delayed feedback, Optik, 125 (21) (2014)
6589-6593.
[23] R. Narmada, M. Arounassalame, Design and performance evaluation of fractional ORDER controller for brushless DC
motor, Int. J. Electr. Eng. Inform., 6 (3) (2014) 606-617.
[24] H. Jaeger, M. Lukosevicius, D. Popovici, U. Siewert, Optimization and applications of echo state networks with
leaky-integrator neurons, Neural Networks, 20 (3) (2007) 335-352.
[25] S.I. Han, J.M. Lee, Precise positioning of nonsmooth dynamic systems using fuzzy wavelet echo state networks and
dynamic surface sliding mode control, IEEE Trans.Ind. Electron., 60 (11) (2013) 5124-5136.

16
 ACCEPTED MANUSCRIPT

[26] G.K. Venayagamoorthy, Online design of an echo state network based wide area monitor for a multimachine power
system, Neural Networks, 20 (3) (2007) 404-413.
[27] W. Xingyuan, M. Juan, Observer-based adaptive fuzzy synchronization for hyperchaotic systems, Chaos, 18 (3) (2008)
033102.
[28] H. Fakham, M. Djemai, K. Busawon, Design and practical implementation of a back-emf sliding-mode observer for a
brushless dc motor, IET Elec. Pow. Appl., 2 (6) (2008) 353-361.
[29] C.E. Ren, S. Tong, Y. Li, Fuzzy adaptive high-gain-based observer backstepping control for SISO nonlinear systems
with dynamical uncertainties, Nonlinear Dyn., 67 (2) (2011) 941-955.
[30] R. Cui, J. Guo, Z. Mao, Adaptive backstepping control of wheeled inverted pendulums models, Nonlinear Dyn., 79 (1)
(2015) 501-511.
[31] Y. Yang, D. Yue, Y. Xue, Decentralized adaptive neural output feedback control of a class of large-scale time-delay
systems with input saturation, J. Frankl. Inst., 352 (5) (2015) 2129-2151.
[32] F. Wang, Q. Zou, C. Hua, Q. Zong, Disturbance observer-based dynamic surface control design for a hypersonic

T
vehicle with input constraints and uncertainty, Proc. Inst. Mech. Eng. I-J. Syst. C, 230 (6) (2016) 522-536.
[33] S. Luo, Y. Song, Chaos analysis-based adaptive backstepping control of the microelectromechanical resonators with
constrained output and uncertain time delay, IEEE Trans.Ind. Electron., 63 (10) (2016) 6217-6225.

IP
[34] J. Fei, W. Yan, Adaptive control of MEMS gyroscope using global fast terminal sliding mode control and
fuzzy-neural-network, Nonlinear Dyn., 78 (1) (2014) 103-116.
[35] S.C. Tong, Y.M. Li, Observer-based adaptive fuzzy backstepping control of uncertain nonlinear pure-feedback systems,

CR
Sci. China Inf. Sci., 57 (1) (2014) 1-14.
[36] Y.D. Song, H. Zhou, X. Su, L. Wang, Pre-specified performance based model reduction for time-varying delay systems
in fuzzy framework, Inform. Sciences, 328 (C) (2016) 206-221.
[37] H. Liu, Y. Pan, S. Li, Y. Chen, Adaptive fuzzy backstepping control of fractional-order nonlinear systems, IEEE Trans.
Syst. Man Cybern. Syst., 47 (8) (2017) 2209-2217.
[38] S. Luo, J. Wang, S. Wu, K. Xiao, Chaos RBF dynamics surface control of brushless DC motor with time delay based on

system, Chaos Solitons Fractals, 22 (5) (2004) 1165-1182.US

tangent barrier Lyapunov function, Nonlinear Dyn., 78 (2) (2014) 1193-1204.
[39] Z.-M. Ge, J.-W. Cheng, Y.-S. Chen, Chaos anticontrol and synchronization of three time scales brushless DC motor

[40] A.M.A. El-Sayed, A. Elsonbaty, A.A. Elsadany, A.E. Matouk, Dynamical analysis and circuit simulation of a new
fractional-order hyperchaotic system and its discretization, Int. J. Bifurc. Chaos, 26 (13) (2016) 1650222.
AN
[41] J. Sabatier, O.P. Agrawal, J.T. Machado, Advances in fractional calculus, Springer, Netherlands, 2007.
M
ED
PT
CE
AC

17