You are on page 1of 19

Advances in Robotics Research, Vol. 2, No.

1 (2015) 1-19
DOI: http://dx.doi.org/10.12989/arr.2015.2.1.001 1

Theoretical and experimental study of switched nonlinear


cascade systems: common Lyapunov function approach
Hossein Chehardoli∗
Department of Mechanical Engineering, K. N. Toosi University of Technology, Tehran, Iran

(Received May 5, 2015, Revised December 14, 2015, Accepted December 31, 2015)

Abstract. This paper describes a procedure to solve the global stabilization and tracking problems of the
switched nonlinear systems presented in the parametric strict feedback form. In previous studies,
simultaneous domination condition (SDC) was introduced for these systems. If this condition is satisfied for
all subsystems by means of virtual control laws, a common Lyapunov function (CLF) can be constructed by
backstepping approach. As a major drawback of the SDC, this condition cannot be satisfied for complicated
switched nonlinear systems. In this paper a new approach based on multiple Lyapunov function (MLF) is
introduced which solves the tracking and regulation problems for switched nonlinear systems in strict
feedback form without any restriction on the SDC condition. For rendering the merit of the proposed MLF-
based backstepping control design, this method is applied practically to a variable structure mechatronic test-
bed. According to the simulation and experimental studies, all simple and complicated switched nonlinear
systems in strict feedback form can be controlled by the proposed approach.

Keywords: backstepping; strict feedback; simultaneous domination condition; common Lyapunov


function; multiple Lyapunov function; dwell time

1. Introduction

A hybrid dynamical system is a dynamical system whose behavior depends on couplings


between discrete and continuance dynamical systems. There are many examples of these systems
such as, room temperature control by thermostat (Van der Schaft and Schumacher 1999), bouncing
ball, gear shift control, computer-controlled system (Lygeros et al. 2012), multi agent systems
(Emmanuel et al. 2014), social networks (Usman and Jadbabaie 2014), etc. Many systems in
various practical applications such as robotic, aerial traffic control, power electronic systems,
chemical and other systems can be modeled as hybrid systems (Yazdi et al. 2009, Tomlin et al.
1998, Ferrari-Trecate et al. 2004, Tan et al. 2004, Yazdi and Jahed-Motlagh 2009, Xu 2014).
Recent progresses in technology have caused a significant interest in the study of hybrid
dynamical systems (Ames et al. 2014, Casau et al. 2013, 2014, Cairano et al. 2014).
Switched systems constitute an important class of hybrid dynamical systems. These systems
consist of a family of subsystems and a rule specifying the switching sequence between them (Hao
et al. 2014, Liberzon 2003). Switched systems have received increasing attention in recent years

*Corresponding author, E-mail: h.chehardoli@mail.kntu.ac.ir

Copyright © 2015 Techno-Press, Ltd.


http://www.techno-press.org/?journal=arr&subpage=7 ISSN: 2287-4976 (Print), 2287-4984 (Online)
2 Hossein Chehardoli

and many studies have been carried out on them, such as stability analysis and stabilization
problem (Shaker and How 2010, Lin and Antsaklis 2009, Cairano et al. 2014, Zhai and Yang
2012), control and tracking problem (Niu and Zhao 2013b, Chiang and Fu 2010), optimal
performance and optimal control (Heydari and Balakrishnan 2014, Zhai et al. 2005), time delay
analyses (Alwan and Liu 2008, Dong et al. 2011), stochastic phenomena (Andrew et al. 2014), etc.
The stability analysis of switched systems is conducted either by application of the common
Lyapunov function (CLF) for arbitrary switching (Buyukoroglu 2012, Liberzon 2003, Wu 2008,
Ma and Zhao 2010, Qingyu et al. 2014) or by application of the multiple Lyapunov function
(MLF) for constrained switching (Niu and Zhao 2013b, Hespanha 2004, Long and Zhao 2012). If
all subsystems of a switched system share a CLF, then it is stable for arbitrary switching (Wu
2009, Niu and Zhao 2013a, Liberzon 2003). On the other hand, a switched system under
constrained switching is stable if all subsystems are stable and the switching signals are relatively
slow such that the switching periods are larger than the dwell time (Liberzon 2003, Pettersson
2006).
Since the complexity of nonlinear systems’ analysis is relatively higher than the linear systems,
stability analysis of switched nonlinear systems includes more difficulties relative to the switched
linear systems. So, fewer results and conclusions can be found for switched nonlinear systems than
switched linear systems in the previous studies. Most of the proposed studies are carried out on
stabilization, control, tracking, optimal performance and other analyses of switched linear systems
(Song and Liu 2014, Qiugang et al. 2013, Chiang and Fu 2012, Grace et al. 2013).
The idea of simultaneous domination condition (SDC) at first was presented by Wu for
switched nonlinear systems described by parametric strict feedback form (Wu 2009). According to
this condition, if all subsystems of a switched nonlinear system in strict feedback form are
simultaneously dominatable by intermediate virtual control laws, a CLF can be constructed
straightforwardly by resorting to backstepping approach (Niu and Zhao 2013b, Maa and Zhao
2010, Wu 2009). At each step of the backstepping approach, a virtual control law is obtained
which makes all prior states stable (Khalil 1996). Since intermediate control laws should stabilize
all subsystems of a switched system, finding them for switched systems is more difficult relative to
non-switched systems. It will be shown later that finding these intermediate control laws which
make all subsystems simultaneously dominatable might be very difficult or even impossible for
complicated switched nonlinear systems. Although some of the researchers in previous studies
have used the SDC idea, but they did not present any procedure to find virtual control laws
satisfying the SDC (Niu and Zhao 2013a, Maa and Zhao 2010, Wu 2009). While the SDC idea
was extensively applied by researchers to solve different problems arisen of nonlinear switched
systems, but its intrinsic restriction for complicated switched nonlinear systems have not been
discussed (Hapanah et al. 2010, Niu et al. 2012, Niu and Zhao 2013b, Niu and Zhao 2013a, Wu
2009).
Since satisfying SDC in complicated switched nonlinear systems is impossible, this study is
composed so as to solve this problem analytically. In this study, a new method is devised in order
to cover the aforementioned problem. In addition, this method will be generalized to the trajectory
tracking problem. In fact, the most contribution of this paper is designing a controller based on
MLF without requiring to spend time between two switching sequences (dwell time). In this
approach it is not necessary for subsystems to be simultaneously dominatable and this is a useful
result of our approach.
The rest of this paper is organized as follows: in section 2, the proposed switched nonlinear
system is introduced and simultaneous domination condition will be presented briefly. In the
Theoretical and experimental study of switched nonlinear cascade systems:… 3

continuance of this section, by an example it will be shown that for complicated switched
nonlinear systems, backstepping approach can not dominate all subsystems simultaneously. In
section 3, a method based on backstepping, MLF and dwell time approaches is presented to
stabilize and control the switched nonlinear systems in strict feedback form. In section 4, a
numerical simulation and an experimental study will be presented to show the effectiveness of this
approach for stabilization and tracking problems. Finally, the paper will be concluded in section 5.

2. Preliminaries and problem statement

Consider the following generic representation of a single-input single-output (SISO) switched


nonlinear system
x& = fσ ( t ) (x) + gσ ( t ) (x)u , σ (t ) ∈ {1, 2,..., m} (1)

where fσ (t ) (x) ∈ ℜ n and gσ (t ) (x) ∈ ℜ n× q are the vectors of switched functions, x ∈ ℜn is the system
state vector, u ∈ ℜq is the vector of system input, m is the number of subsystems and σ(t) is a time
dependent switching signal which is not observable.
A smooth positive definite function V (x) : ℜ n → ℜ+ , is called a CLF for all subsystems in Eq.
(1) if for all switching signals σ(t), the following inequality holds its validity (Liberzon 2003)
∂V
fσ (t ) (x) + gσ (t ) (x)u  < 0 (2)
∂x 
If a switched system has a CLF, it is asymptotically stable under arbitrary switching (Liberzon
2003)
Consider the following representation of the switched nonlinear strict feedback system (Wu
2009)
 x&1 = x2 + η1,σ (t ) ( x1 )

 M

 xi = xi +1 + ηi,σ (t ) (xi )
& ; y = x1 (3)

 M
 x&n = fσ ( t ) (x n ) + gσ ( t ) (x n )u

where, x1, x2,…, xn are system state variables and y is the system output. ηi, σ(t)(xi), fσ(t)(x) and g
σ(t)(x)=[gσ,1, gσ,2, …, gσ,q] are switching functions, u=[u1,u2,…,uq] is the vector of control input and
xi=[x1,x2,…,xi]. It is assumed that the origin is an equilibrium point of this system implying
consequently ηi,k (0)=0 and f σ(t) (0)=0. Besides, it is assumed that gσ(t),i(0)≠0, i=1,…,q. In the rest of
the paper the switching signal σ(t) will be shown by σ.
As a review of SDC, consider the backstepping control design for the switched system (3). At
the first step of this approach, the first equation of (3) is considered
x&1 = x2 + η1,σ (t ) ( x1 ) (4)

In this step, the CLF and its time derivative along x&1 is
4 Hossein Chehardoli

1 2
V1 = x1 ⇒ V&1 = x1 [ x 2 + η1,σ ( x1 )] (5)
2
According to the backstepping approach, the first intermediate virtual control law is supposed
as x2. All of the first-order subsystems (4) are simultaneously dominatable if there exists a
differentiable feedback law x2= α1 such that V& <0 for all σ∈{1,…,m}. In the procedure described
1
in(Wu 2009), at each step it is necessary to find an intermediate control law so as to make all
subsystems simultaneously dominatable.
This method for finding a CLF (which is based on backstepping approach) has two main basic
limitations: 1- if all subsystems are simultaneously dominatable, this approach only can stabilize
the proposed nonlinear system and cannot solve the tracking problem. 2- in the case of
complicated switched nonlinear systems, finding an appropriate virtual control law satisfying the
SDC for all subsystems is a hard or even impossible task. To illustrate this problem, two different
switched nonlinear systems are presented as follow.
System1
 x&1 = x2 + η1,σ ( x1 )
 ; σ ∈ {1, 2,3}
 x&2 = u (6)
η1,1 = x13 ,η1,2 = 2 x13 ,η1,3 = 3x13

This switched system has three subsystems with very similar structural nonlinearity (all are
cube). It can be shown that the consequence of choosing x2= α1=−4x13 is
 & σ =1
V1 = x1 ( x2 + x1 ) = −3 x1 < 0
3 4

 σ =2
x2 = α1 = −4 x13 ⇒ V&1 = x1 ( x2 + 2 x13 ) = −2 x14 < 0 (7)
 σ =3
V&1 = x1 ( x2 + 3x13 ) = − x14 < 0

So, V&1 < 0, σ∈{1,2,3}. Therefore, all of the first-order subsystems of (7) are simultaneously
dominatable. Because the SDC is satisfied for this system, the stabilization problem can be solved
and α1 stabilizes all subsystems. But as mentioned previously, the tracking problem cannot be
considered. For the second example consider the following switched system
 x&1 = x 2 + η1,σ ( x1 ), σ ∈ {1,2,3},

 x& 2 = u (8)
η1,1 = x13 ,η1, 2 = x12 sin( x1 ),η1,3 = x13 + x1 sin( x1 )
System2
It is obvious that finding analytically a suitable α1 satisfying SDC for all subsystems of (8) is
impossible. Note that α1 should satisfies all the following inequalities
So, against (6), SDC is not satisfied for (8).
The above two examples show that the SCD can be satisfied in only simple nonlinear systems
(however by this method, the proposed switched system might be only stabilized). In the
complicated switched nonlinear systems the SDC is not satisfied. So, it is not possible to control or
Theoretical and experimental study of switched nonlinear cascade systems:… 5

 & σ =1 ?

V 1 = x1 (α 1 + x1
3
) < 0
 & σ =2 ?
x2 = α1 ⇒ V1 = x1 (α1 + 2 x13 ) < 0 (9)
 σ =3 ?
V&1 = x1 (α1 + 3x13 ) < 0

even stabilize these systems by SDC approach. In the next section a method based on MLF is
presented which obviates this defect.

3. Stability analysis and control design based on state transformation approach

In this section by application of the backstepping approach, a MLF for the switched system (3)
is introduced. Then it will be proven that with zero amount of dwell time the switched nonlinear
system (3) is globally asymptotically stable under arbitrary switching.
By the following coordinate transformation
 p1 = x1

 p 2 = x&1 = x 2 + η1,σ
 ∂η1,σ
 p3 = &x&1 = x3 + η 2,σ + ( x 2 + η1,σ ) (10)
 ∂x1
M

 p n = x1( n −1)

system (3) is converted to


 p&1 = p2
 p& = p
 2

3
; y = p1 (11)
 M
 p& n = hσ (xn ) + gσ (x n )u

which hσ(xn) is a complicated switched function of x, ηi,σ(xi) and fi,σ(x).


In the following subsections, the stability and tracking problems are investigated.

3.1 Stabilization problem

In order to design backstepping controller, the procedure of (Khalil 1996) is followed.


Step 1: By considering the definition z1=p1, the first equation of (11) can be written as z&1 = p 2 . At
this step, the MLF and its time derivative along z&1 is
1 2
V1 = z1 ⇒ V&1 = z1 p 2 (12)
2
Choosing the first virtual control law as α1=p2=−c1z1, will result in V&1 = −c1 z1 , c1 > 0 .
2
6 Hossein Chehardoli

Step 2: By setting the definition z2=p2−α1, the first two equations of (11) can be written as
z&1 = z 2 + α 1
∂α 1 (13)
z& 2 = p3 − p2
∂p1
The MLF for this step and its time derivative along (13) are obtained as
1 2 1 2 1 2 ∂α
V2 = V1 + z 2 = z1 + z 2 ⇒ V&2 = z1 ( z 2 + α 1 ) + z 2 [ p3 − 1 p 2 ] (14)
2 2 2 ∂p1
Choosing
∂α1
p3 = α2 = −c2 z2 + p2 − z1 , c2 > 0 (15)
∂p1

leads to V&2 = −c1 z1 − c2 z 2 .


2 2

Step r: Similarly, by setting zˏ=pˏ− αˏ-1, the first r equations of (11) can be written in terms of new
variables z1,z2,…,zˏ as
 z&1 = z 2 + α 1

 z& 2 = z 3 + α 2 − ∂α 1 p 2
 ∂p1
 (16)
 M
 r −1 ∂α
r −1
 z& r = p r +1 − ∑ pi +1
 i =1 ∂p i

For this step, the MLF and the virtual control law are
1 r 2 r −1
∂α r −1
Vr = ∑ zi ,
2 i =1
pr +1 = α r = −cr zr − zr −1 + ∑
i =1 ∂pi
pi +1 , cr > 0 (17)

Step n: Finally, by the variable change zn=pn−αn-1 the system (11) is converted to


 z&1 = z2 + α1
M

 r −1
∂α r −1
 z&r = pr +1 − ∑ pi +1 (18)
 i =1 ∂pi
M

 n −1
∂α n −1
z
 n& = hσ ( x ) + g σ ( x )u − ∑i =1 ∂pi
pi +1

1 n 2
The MLF for (18) is Vσ = ∑ zi . Time derivative of Vσ along (18) and defining the control law as
2 i =1

gTσ (x) n −1
∂α
u=− [c n z n + hσ (x) − ∑ n −1 pi +1 + zn −1 ], c n > 0 (19)
i =1 ∂pi
T
gσ (x)gσ (x)
Theoretical and experimental study of switched nonlinear cascade systems:… 7

n
will lead to V& = − ∑ ci z i2 . It is obvious that control input (19) stabilizes each subsystem
i =1
individually. But this is not sufficient for stability of the system in switching instants. It is a well-
known fact that when all subsystems of (3) are asymptotically stable, this system is asymptotically
stable if the dwell time τd is sufficiently large (Liberzon 2003). The required lower bound of τd for
a switched system with two subsystems can be calculated as (Liberzon 2003)
1 bb
τd > ln( 1 2 ) (20)
2(λ1 + λ2 ) a1a2

which a1,a2,b1 and b2 satisfy the following math expressions


2 2
aσ z ≤ V (z ) ≤ bσ z
(21)
V& ≤ −cσ z , σ ∈ {1, 2}
2

c c
where z=[z1,z2,…,zn], ||z|| is the Euclidean norm of z, λ1 = 2b and λ2 = 2b . It can be shown
1 2

1 2

(Appendix 1) that for a switched system with m subsystems, the lower limit of τd can be calculated
by
1 b b ...b
τd > ln( 1 2 m ) (22)
2(λ1 + λ2 + .. + λm ) a1a2 ...am

in which, the parameters λi,ai and bi,(i=3,…,m) are defined similar to λi,ai and bi,(i =1,2) (Appendix
1 n 2
1) The MLF σ 2 ∑ zi satisfies the following inequality
V =
i =1

1 2 1 2
( −ε) z ≤ Vσ (z ) ≤ ( + ε ) z (23)
2 2
1 1
in which, ε is an infinitesimal positive value. Hence, for all subsystems: bσ = + ε and aσ = − ε .
2 2
With these values of bσ and aσ, the lower bound of τd is calculated as
1
( + ε )m
1 b1b2 ...bm 1 1
τ d > lim ln( )= lim ln 2 = ln(1) = 0 (24)
ε → 0 2(λ + λ + .. + λ ) 2(λ1 + λ2 + .. + λm ) ( − ε ) m 2(λ1 + λ2 + .. + λm )
a1a2 ...am ε → 0 1
1 2 m
2
By accumulation of the aforementioned results, the following theorem is achieved.
Theorem 1: The switched nonlinear system (3) with control input (19) is globally asymptotically
stable under arbitrary switching.
Proof
n
Time derivative of the MLF Vσ along (18) and applying (19) lead to V& = − ∑ ci z i2 . So, all
i =1
subsystems are globally asymptotically stable without considering the switching process. For
stability proof in switching instants, the time elapsed between two switching operations should be
greater than the lower limit of dwell time which was calculated zero in (24). It is obvious that
8 Hossein Chehardoli

every infinitesimal positive value as the dwell time, satisfies the expression τd > 0. So, it is
concluded that the switching system (3) is globally asymptotically stable under the control input
(19) and under arbitrary switching.

3.2. Tracking problem

In this part by applying simple variable changes, the problem of tracking a desired trajectory is
solved for system (3). To this objective, the variable x1 which is considered as the output of (3), is
supposed to track a desired trajectory xd.
Step 1: By defining z1=p1−xd, the first equation of (11) is converted to z&1 = p2 − x& d . In this step,
the MLF and the virtual control law are obtained as
1 2
V1 = z1 ⇒ V&1 = z1 ( p 2 − x& d ) ⇒ p 2 = α 1 = −c1 z1 + x& d (25)
2
Step r: Consider zˏ=pˏ− αˏ-1, which αˏ-1 is the virtual control law of step (r-1). The first r equations
of (11) in terms of new variables zi (i=1,…,r), are written as
 z&1 = z 2 + α 1 − x& d

 z& 2 = z 3 + α 2 − ∂α 1 p 2 − &x&d
 ∂p1
 (26)
 M
 r −1 ∂α
 z& r = x r +1 − ∑ r −1 pi +1 − x d
(r )
 i =1 ∂p i

Hence, the MLF and the virtual control law for this step are
1 r 2 r −1
∂α r −1
Vr = ∑ zi ,
2 i =1
pr +1 = α r = −cr zr − zr −1 + ∑
i =1 ∂pi
pi +1 + xd( r ) (27)

Step n: Similar to the previous steps, by defining zn=pn− αn-1 and by using definitions of zi, i =
1,…,n-1, the system (11) is converted to


 z&1 = z2 + α1 − x&d
 M

 r −1
∂α r −1
 z&r = pr +1 − ∑ pi +1 − xd( r ) (28)
 i =1 ∂pi
 M

 n −1
∂α n −1
z
 n& = hσ ( x ) + g σ ( x )u − ∑ ∂
pi +1 − xd(n)
 i =1 p i

n
1
By applying the backstepping design, the MLF is generated as Vσ = 2 ∑ zi . By time
2

i =1

differentiation of Vσ along (28) and choosing the following control law


Theoretical and experimental study of switched nonlinear cascade systems:… 9

gTσ (x) n −1
∂α n −1
u=− T
gσ (x)gσ (x)
[c z
n n + hσ ( x ) − ∑
i =1 ∂pi
pi +1 + zn −1 − xd(n) ], c n > 0 (29)

n
the time derivative of Lyapunov function will be V& = − ∑ ci z i2 . So, the output of (3) will track the
i =1
desired signal xd asymptotically. The results are presented as the following theorem.
Theorem 2: The output of the switched nonlinear system (3) tracks the desired trajectory xd by
applying control input (29), under arbitrary switching.

Proof
n
Time derivative of Vσ along (28) and using control input (29) result in V& = − ∑ ci z i2 . So, each
i =1
subsystem is globally asymptotically stable. Since the lower limit of dwell time is zero (similar to
(24)), the stability in switching instants is assured. So, the output of (3) will track the desired
trajectory asymptotically.

Remark
As an important point, it should be noted that if a switched system has a MLF with the zero
lower limit of dwell time, then it is stable under arbitrary switching not constrained switching.
Consequently, the necessity for finding a CLF (which is difficult in the most cases) is vanished. In
other words, a MLF with zero lower limit of dwell time ensures the stability under arbitrary
switching.

4. Simulation and experimental studies

In this section, in order to show the effectiveness of the proposed technique, a numerical
example about stabilization problem and an experimental example for tracking problem are
presented.

4.1 Stabilization problem (Simulation study)

Consider the following switched nonlinear system with three subsystems


 x&1 = x2 + η1,σ ( x1 )

 x&2 = x3 + η2,σ (x 2 ) , σ ∈ {1, 2,3} (30)
&
 x3 = fσ (x) + gσ (x)u
where
η1,1 = x1 , η1,2 = x13 , η1,3 = x12 sin( x1 )

η 2,1 = sin( x2 ), η 2,2 = x1 x2 , η 2,3 = x1 x2
2

 (31)
 f1 = x1 , f 2 = x1 x2 , f 3 = x2 + x3
2 2


 g1 = 2 + x1 , g 2 = 5, g 3 = 7 + x2
2 4
10 Hossein Chehardoli

According to (29), the control input of this system is


1 2
∂α
u=− [c3 z 3 + hσ (x) − ∑ 2 pi +1 + z2 − xd(3) ] (32)
gσ (x) i =1 ∂pi

where
∂α1
z 3 = p3 − α 2 , α 2 = −c 2 z 2 + p2 − z1 , z 2 = p2 − α1 , α1 = −c1z1 , z1 = p1 − xd , xd = 0, c1 , c 2 , c3 > 0
∂p1
∂η1,σ ∂η 2,σ ∂η 2,σ ∂η1,σ
hσ (x) = fσ (x) + ( x3 + η 2,σ )( + ) + ( x2 + η1,σ )[ +( )2 ] + ( x2 + η1,σ ) 2
∂x1,σ ∂x2,σ ∂x1 ∂x1

According to the theorem (1), system (30) should withstand any switching signal. Because this
theorem indicates that the proposed switched system is stable with the zero lower limit of dwell
time. So, a switching signal with very high frequencies is applied to system (30) which is depicted
in Fig. 1. Fig. 2 shows the time variation of states x1, x2 and x3. According to this figure, system
(30) is asymptotically stable with the control input (32). Fig. 3 shows the time variation of the
control input (32). According to this figure, because of the high frequent switching behavior of the
system (30), the control input (32) is jagged.

3.5

3
Switching signal

2.5

1.5

0.5
0 1 2 3 4 5 6 7 8 9 10
Time (second)

Fig. 1 Switching signal

4
x1
x2
2 x3
States of system

-2

-4

-6

-8

-10

-12
0 1 2 3 4 5 6 7 8 9 10
Time (second)

Fig. 2 State responses of system (30)


Theoretical and experimental study of switched nonlinear cascade systems:… 11

40

20

Control input
-20

-40 10

-60 0

-80 -10

0.4 0.6 0.8 1


-100
0 1 2 3 4 5 6 7 8 9 10
Time (second)

Fig. 3 Control input for the system (30)

4.2 Tracking problem (Experimental study)

In order to evaluate practically the proposed MLF-based switched nonlinear algorithm, a


variable structure test-bed consisting of mechanical and electrical structures is designed (Fig. 4).

4.2.1 Mechanism structure


The proposed structure consists of two shape memory alloy wires adjoining differentially.
There is a 5K Ω rotational potentiometer in the center of mechanism to measure the angular
position of system. Two load cells are used to measure SMA wire forces. Each load cell consists
of a wetston bridge with 0.9K Ω resistance in unloading condition. Two indicator clocks are used
to show the linear displacement of the center of mechanism. Fig. 5 shows the electrical structure of
this mechanism. As this figure shows, there is an electrical diode in every wire path to rectify the
electric current of wire.
The microcontroller (MC) of the mechanism is of the Cortex M3. The most important duty of
this MC is controlling the angular position of the mechanism. The STM-32 MC processes and
implements the command algorithms developed for controlling angle of the mechanism. Other
important characteristics of this MC are maximum processing clock-rate of 53 MHz and 256
Kbytes flash memory.
The governing motion equation of the this mechanism is
θ&&& = hk u + g k θ& + d k , k ∈ {1, 2, 3} (33)
Where, θ is the angular position of the mechanism. u = i 2 , i is the current of SMA wire, k is
the number of active subsystem and d k is the disturbance signal. For the above mechanism, we
have three subsystems by the following switching functions
Arα ΩξT ,1 + θt Ar 2 D kr 2 Ar ΩξT ,1 + θ t
k = 1: h1 = , g1 = −( + ), d1 = − β (T − T∞ ) (34)
I 1 − Ωξσ ,1 Il0 1 − Ωξσ ,1 I I 1 − Ωξσ ,1

Arα ΩξT ,2 + θ t Ar 2 D kr 2 Ar ΩξT ,2 + θ t


k = 2 : h2 = , g 2 = −( + ), d 2 = − β (T − T∞ ) (35)
I 1 − Ωξσ ,2 Il0 1 − Ωξσ ,2 I I 1 − Ωξσ ,2
12 Hossein Chehardoli

Fig. 4 Experimental setup of shape memory alloy actuator

Fig. 5 Electrical structure of the SMA actuators consisting of loadcells, diodes, amplifiers,
potentiometer and a PC

Arα Ar 2 kr 2 Ar
k = 3 : h3 = θ t , g3 = −( D+ ), d3 = − θt β (T − T∞ ) (36)
I Il0 I I

For keeping the paper commensurate, only beneficial concepts are discussed. Readers can refer
to (M. Moalm and V. Tbrzi 2009) for more and comprehensive details. It should be noted that the
switching controller was not designed for SMA actuators in the previous studies such as (M.
Moalm, V. Tbrzi 2009).
The equation (33) can be represented in state space form as following
 x&1 = x2

 x&2 = x3 ; k ∈ {1, 2,3} ,[ x1 , x2 , x3 ] = [θ ,θ&, θ&&] (37)
 x& = h u + g x + d
 3 k k 2 k

The disturbance signals d k are bounded. So, upper limit Dk is assumed for them |dk|<Dk,
k=1,2,3. Functions hk and gk change very slowly. So, they are considered as unknown parameters
Theoretical and experimental study of switched nonlinear cascade systems:… 13

and the robust adaptive backstepping approach is used to design controller. By following this
approach the CLF is constructed as
1 2 1 2 1 2 1 %2 1 %2 %
V= z1 + z2 + z3 + ξ1 + ξ 2 , ξ = ξ − ξˆ (38)
2 2 2hk 2γ 1 2γ 2

ξˆ is the estimation of ξ, ξ1=g/h, ξ2= 1/h and γ1, γ2 are positive gains. Control and adaptation laws
can be obtained by time derivation of (38) along (37) and simplifying results as
− z 2 − k 3 z 3 − ζˆ1 x 2 − ζˆ2 Fζk , k = 1,2
u=
imax, k =3
(39)
&ˆ &ˆ ∂α 2 ∂α 2
ζ = γ 1 z 3 x2 ; ζ 2 = γ 2 z 3 Fζk = sign( z 3 ) D k − ( z 2 − k1 z1 ) − ( z 3 − z1 − k 2 z 2 ) + &x&&d , k = 1,2
∂z1 ∂z 2
Fig. 6 shows the closed loop structure of SMA mechanism and controller. According to this
figure, when the switching signal is produced by system the adaptive backstepping controller
switches to new structure.
Fig. 7 shows the angular position tracking problem of the SMA mechanism. The switching
signal is depicted in Fig.8. Control input (39) is depicted in Fig. 9. According to Fig. 7, the
tracking accuracy is desirable in the presence of variable disturbance (dk) and switching events
(Fig. 8).

Disturbance
Switching Load
Signal

r(t) + e(t) Adaptive u(t) Switching y(t)


Backstepping Plant
- Controller

Output
Analyzer
Output
Measurement

Fig. 6 The closed-loop system of SMA mechanism

25
actual
20 desired

15

10
Angular Position (degree)

-5

-10

-15

-20

-25
0 5 10 15 20 25 30 35 40
Time (Second)

Fig. 7 Angular position tracking for test-bed mechanism


14 Hossein Chehardoli

3.5

Switching signal
2.5

1.5

1
0 5 10 15 20 25 30 35 40
Time (second)

Fig. 8 Switching signal

0.35

0.3

0.25
Control input (Amper)

0.2

0.15

0.1

0.05

0
0 5 10 15 20 25 30 35 40
Time (second)

Fig. 9 Control input u (Amper)

-0.0599 0.085

-0.0599 0.08

0.075
-0.06

0.07
Estimation of g/h

Estimation of 1/h

-0.06
0.065
-0.06
0.06
-0.06
0.055

-0.06
0.05

-0.0601 0.045

-0.0601 0.04
0 10 20 30 40 0 10 20 30 40
Time (Second) Time (Second)

Fig. 10 Estimation of unknown parameters ξ1=g/h and ξ2=1/h


Theoretical and experimental study of switched nonlinear cascade systems:… 15

Fig. 10 shows the estimations of unknown parameters ξ1 and ξ2 (according to adaptation


laws(39)), respectively.

5. Conclusions

In this paper an earlier method in the previous researches based on simultaneous domination
condition (SDC) for finding a common Lyapunov function for a family of switched nonlinear
systems was reviewed briefly. Then, by an example it was shown that SDC cannot be satisfied in
complicated switched nonlinear systems. To overcome this difficulty, a new approach based on
state transformation and backstepping was presented. It was shown that this new approach solves
the problems of stabilization and tracking control for all cascade switched nonlinear systems
without any restriction. To show the effectiveness of the proposed approach, a numerical
simulation and an experimental study were offered. According to obtained results, the new
approach shows merits for controlling and stabilizing the switched nonlinear systems in the
parametric strict feedback form.

References

Alwan, M.S. and Liu, X. (2008), “On stability of linear and weakly nonlinear switched systems with time
delay”, J. Math. Comput. Model., 48(7), 1-11.
Ames, A.D., Galloway, K., Sreenath, K. and Grizzle, J.W. (2014), “Rapidly exponentially stabilizing control
lyapunov functions and hybrid zero dynamics”, Automatic Control, 59(4), 876-891.
Büyükköroğlu, T. (2012), “Common diagonal Lyapunov function for third order linear switched system”, J.
Comput. Appl. Math., 236(15), 3647-3653.
Casau, P. Cabecinhas, D. and Silvestre, C. (2014), “Hybrid control strategy for the autonomous transition
flight of a fixed-wing aircraft”.
Casau, P., Cabecinhas, D. and Silvestre, C. (2013), “Hybrid control strategy for the autonomous transition
flight of a fixed-wing aircraft”, Proceedings of the IEEE Transactions on Control Systems Technology,
21(6), 2194-2211.
Chiang, M.L. and Fu, L.C. (2010), “Variable structure adaptive backstepping control for a class of unknown
switched linear systems”, Proceedings of the American Control Conference (ACC), Baltimore, USA.
Chiang, M.L. and Fu, L.C. (2012), “Average dwell time condition of unknown switched linear systems with
variable structure adaptive backstepping control”, Proceedings of the American Control Conference
(ACC), Montréal, Canada
Deaecto, G.S., Fioravanti, A.R. and Geromel, J.C. (2013), “Suboptimal switching control consistency
analysis for discrete-time switched linear systems”, Eur. J. Control, 19(3), 214-219.
Di Cairano, S., Heemels, W.M.H., Lazar, M. and Bemporad, A. (2014), “Stabilizing dynamic controllers for
hybrid systems: a hybrid control Lyapunov function approach”, Proceedings of the IEEE Transactions on
Automatic Control, 59(10), 2629-2643.
Dong, Y., Liu, J., Mei, S. and Li, M. (2011), “Stabilization for switched nonlinear time-delay systems”,
Nonlin. Analysis: Hybrid Syst., 5(1), 78-88.
Ferrari-Trecate, G., Gallestey, E., Letizia, P., Spedicato, M., Morari, M. and Antoine, M. (2004), “Modeling
and control of co-generation power plants: a hybrid system approach”, Proceedings of the IEEE
Transactions on Control System Technology, 12(5), 694-705.
Hapanah, A. Etasad, M., Hemyan, M.R. and Khtian, A. (2010), “Adaptive stabilization of uncertain
switched nonlinear systems in strict-feedback form”, Proc. Int. Mec. Eng. Vancouver, Canada.
16 Hossein Chehardoli

Hespanha, J.P. (2004), “Uniform stability of switched linear systems: extensions of LaSalle's invariance
principle”, Automatic Control, 49(4), 470-482.
Heydari, A. and Balakrishnan, S.N. (2014), “Optimal switching and control of nonlinear switching systems
using approximate dynamic programming”, Neur. Network. Learn. Syst., 25(6), 1106-1117.
Khalil, H.K. and Grizzle, J.W. (1996), Nonlinear systems. New Jersey, Prentice hall.
Khan, U.A. and Jadbabaie, A. (2014), “Collaborative scalar-gain estimators for potentially unstable social
dynamics with limited communication”, Automatica, 50(7), 1909-1914.
Krstic, M., Kokotovic, P.V. and Kanellakopoulos, I. (1995), Nonlinear and adaptive control design, John
Wiley & Sons.
Liberzon, D. (2003), Switching in systems and control, Boston, Birkhauser.
Lin, H. and Antsaklis, P.J. (2009), “Stability and stabilizability of switched linear systems: a survey of
recent results”, Proceedings of the IEEE Transactions on Automatic control, 54(2), 308-322.
Long, L. and Zhao, J. (2012), “Control of switched nonlinear systems in-normal form using multiple
Lyapunov functions”, Automatic Control, 57(5), 1285-1291.
Lu, Q., Zhang, L., Karimi, H.R. and Shi, Y. (2013), “Hα control for asynchronously switched linear
parameter-varying systems with mode-dependent average dwell time”, IET J. Control Theo. Appl., 7(5),
673-683.
Lygeros, J., Sastry, S. and Tomlin, C. (2012), Hybrid Systems: Foundations, advanced topics and
applications, Springer Verlag.
Ma, R. and Zhao, J. (2010), “Backstepping design for global stabilization of switched nonlinear systems in
lower triangular form under arbitrary switchings”, Automatica, 46(11), 1819-1823.
Moalem, M. and Tbrizi, V. (2009), “Tracking control of an antgonstic shape memory alloy actuator”,
Proceedings of the IEEE Transactions on Control Systems Technology,17(1), 184-190.
Niu, B. and Zhao, J. (2013), “Output tracking control for a class of switched non-linear systems with partial
state constraints”, IET Control Theor. Appl., 7(4), 623-631.
Niu, B. and Zhao, J. (2013), “Tracking control for output-constrained nonlinear switched systems with a
barrier Lyapunov function”, Int. J. Syst. Sci., 44(5), 978-985.
Niu, B. Georgi, D.M. and Zhao, J. (2012), “Tracking control for a class of switched nonlinear systems with
full state constraints”, Proceedings of the 31st Chinese Control Conference (CCC).
Nuño, E., Valle, D., Sarras, I. and Basañez, L. (2014), “Leader-follower and leaderless consensus in
networks of flexible-joint manipulators”, Eur. J. Control, 20(5), 249-258.
Pettersson, S. (2006), “Designing switched observers for switched systems using multiple Lyapunov
functions and dwell-times”, Analysis Des. Hybrid Syst., 2(1), 18-23.
Shaker, H.R. and How, J.P. (2010), “Stability analysis for class of switched nonlinear systems”, Institute of
Electrical and Electronics Engineers.
Song, H. and Liu, G.P. (2014), “Networked predictive control of linear switched systems”, Proceedings of
the 33rd Chinese Control Conference (CCC).
Su, Q., Long, L. and Zhao, J. (2014), “Stabilization of state‐constrained switched nonlinear systems in
p‐normal form”, Int. J. Robust Nonlin. Control, 24(10), 1550-1562.
Tan, J., Xi, N. and Wang, Y. (2004), “A singularity-free motion control algorithm for robot manipulators-a
hybrid system approach”, Automatica, 40(7), 1239-1245.
Teel, A.R., Subbaraman, A. and Sferlazza, A. (2014), “Stability analysis for stochastic hybrid systems: A
survey”, Automatica, 50(10), 2435-2456.
Tomlin, C., Pappas, G.J. and Sastry, S. (1998), “Conflict resolution for air traffic management: A study in
multiagent hybrid systems”, Proceedings of the IEEE Transactions on Automat. Control, 43(4), 509-521.
Van der Schaft, A. and Schumacher, H. (1999), An introduction to hybrid dynamical systems, London,
Springer-Verlag.
Wu, J.L. (2008), “Feedback stabilization for multiinput switched nonlinear systems: two subsystems case”,
Proceedings of the IEEE Transactions on Automatic Control, 53(4), 1037-1042.
Theoretical and experimental study of switched nonlinear cascade systems:… 17

Wu, J.L. (2009), “Stabilizing controllers design for switched nonlinear systems in strict-feedback form”,
Automatica, 45(4), 1092-1096.
Xu, Q. (2014), “Design and smooth position/force switching control of a miniature gripper for automated
microhandling”, Proceedings of the IEEE Transactions on Industrial Informatics, 10(2), 1023-1032.
Y. Hao, B. Jiang, C. Vincent (2014). Stabilization of Switched Nonlinear Systems with Unstable Modes,
springer Verlag.
Yang, H., Jiang, B. and Cocquempot, V. (2014), Stabilization of Switched Nonlinear Systems with Unstable
Modes, Springer.
Yazdi, M.B. and Jahed-Motlagh, M.R. (2009), “Stabilization of a CSTR with two arbitrarily switching
modes using modal state feedback linearization”, Chem. Eng. J., 155(3), 838-843.
Yazdi, M.B., Jahed-Motlagh, M.R., Attia, S.A. and Raisch, J. (2009), “Switched feedback equivalence of a
class of planar switched nonlinear systems”, 28th Chinese Control Conference.
Zhai, G., Lin, H., Kim, Y., Imae, J. and Kobayashi, T. (2005), “Gain analysis for switched systems with
continuous-time and discrete-time subsystems”, Int. J. Control, 78(15), 1198-1205.
Zhai, S. and Yang, X. S. (2012), “Stability analysis of feedback switched systems with state and switching
delays”, IMA J. Math. Control Inform., 30, 21-35.

CC
18 Hossein Chehardoli

Appendix

In this section the lower bound of dwell time ensuring the stability of switched systems under
constrained switching is obtained. Consider the following representation of switched systems
x& = fσ ( x, t , u ), σ ∈ {1, 2,..., m} (A-1)
where m is the number of subsystems.
Theorem (A-1) (Liberzon 2003): Let switched system (A-1) be a finite family of globally
asymptotically stable systems, and Vσ be a family of corresponding radially unbounded Lyapunov
functions. Consider the positive definite continuous functions Wσ with the property that for every
pair of switching times (ti, tj), i < j
Vσ ( x(t j )) − Vσ ( x(ti )) ≤ −Wσ ( x(t j )) (A-2)
Then the switched system (A-1) is globally asymptotically stable. Suppose there exist positive
values aσ, bσ and cσ satisfying following inequalities
2 2 d 2
aσ x ≤ Vσ (x) ≤ bσ x ; Vσ (x) ≤ −cσ x (A-3)
dt
Combining the above inequalities leads to
d
Vσ (x) ≤ −2λσ Vσ (x) (A-4)
dt
cp
in which, λσ = . This implies that
2bp

Vσ (x(t0 + τ d )) ≤ e−2λσ τ d Vσ (x(t 0 )) (A-5)

provided that t∈[t0, t0+ τd]. Suppose that ti, i∈{0,1,…,m-1} is a switching instant. In other words,
ti+1−ti ≥ τd, for example in t = t1 the first switching occurs and second subsystem is active.
For simplifying the next calculations, consider the case σ∈{1,2}. σ =1 when t0 ≤ t < t1 and σ =2
on t1 ≤ t < t2 (according to Fig. A-1).
In this case, from the first inequality in (A-3) and inequality (A-5) we have
b2 b
V2 (t1 ) ≤ V1 (t1 ) ≤ 2 e −2 λ1τ d V1 (t0 ) (A-6)
a1 a1
It can be easily shown that the following expression is hold between every two consecutive
subsystems

Fig. A-1 switching instants and dwell time


Theoretical and experimental study of switched nonlinear cascade systems:… 19

bi b
Vi (ti −1 ) ≤ Vi −1 (ti −1 ) ≤ i e−2λi−1τ d Vi −1 (ti − 2 ) (A-7)
ai −1 ai −1

So, for a general switched system with m subsystems the following math expressions are hold
b2 b
V2 (t1 ) ≤ V1 (t1 ) ≤ 2 e −2 λ1τ d V1 (t0 )
a1 a1
b3 b
V3 (t2 ) ≤ V2 (t2 ) ≤ 3 e−2λ2τ d V2 (t1 )
a2 a2
M M M (A-8)
bm b
Vm (tm −1 ) ≤ Vm −1 (tm −1 ) ≤ m e−2λm−1τ d Vm −1 (tm − 2 )
am −1 am −1
b1 b
V1 (tm ) ≤ Vm (tm ) ≤ 1 e −2 λmτ d Vm (tm −1 )
am am

combining the above inequalities leads to


b1b2 ...bm −2(λ1 + λ2 +...+ λm )τ d
V1 (tm ) ≤ e V1 (t0 ) (A-9)
a1a2 ...am

According to the theorem (A-1), the relation V1(tm) − V1(t0) < 0 will be satisfied if
b1b2 ...bm −2(λ1 + λ2 +...+ λm )τ d
e −1 < 0 (A-10)
a1a2 ...am

which can be equivalently written as


a1a2 ...am
−2(λ1 + λ2 + ... + λm )τ d < ln (A-11)
b1b2 ...bm

or finally as
1 b b ...b
τd > ln 1 2 m (A-12)
2(λ1 + λ2 + ... + λm ) a1a2 ...am

You might also like