Professional Documents
Culture Documents
Abstract
Although the simple adaptive control (SAC) is widely studied both in theory and application in flexible space structure control
and other control problems, it is restricted by the almost strictly positive real (ASPR) conditions. In most practical control prob-
lems, the ASPR conditions are not satisfied. Therefore, based on the SAC theory, this paper proposes a backstepping simple
adaptive control algorithm which suits the system with arbitrary relative degree with no need of parallel feedforward compensa-
tor. The proposed control algorithm consists of decomposition of the arbitrary relative degree system into a known subsystem
and an unknown ASPR subsystem which are connected in cascade, design of constant output feedback controller for the known
subsystem, and implementation of backstepping method and SAC of the unknown ASPR subsystem. Inheriting the characteris-
tics of the SAC, this method can be adaptive online for the parameter uncertainties. Then, the application of the proposed con-
troller to large flexible space structure with collocated sensors and actuators is studied, and the simulation results validate the
proposed controller. It is a new strategy to apply the classical SAC to high relative degree plants.
Keywords: simple adaptive control; almost strictly positive real; backstepping; flexible structure; intermediate control law
1. Introduction 1
turbines [14-16].
In order to apply a simple adaptive controller, the
The simple adaptive control (SAC) methodology system should comply with some requirements. One of
was first introduced by Sobel, et al. in 1979 [1] and fur- the requirements is that the plant should be almost
ther developed by Bar-Kana, et al. [2]. It has also been strictly positive real (ASPR) [8], that is, there exists a
developed by Balas, et al. to infinite-dimensional sys- constant output feedback so that the resulting
tems [3], discrete systems [4] and systems with unknown closed-loop system is strictly positive real. It has been
delays and persistent disturbances [5]. It gives a shown that plants with minimum-phase transfer func-
low-order controller with no need of the order knowl- tions of relative degree 1 in the single-input sin-
edge of the controlled plant, so it can stabilize gle-output (SISO) case, or of relative degree m in the
large-scale system with small number of adjustable multiple-input multiple-output (MIMO) multivariable
parameters. The simplicity and robustness of SAC have case, are ASPR [13]. However, most practical plants are
led to successful implementations in such diverse ap- not ASPR. Taking the flexible space structure as an
plications as flexible space structures [2,6-8], flight con- example, it has a relative degree of 2, so it is not an
trol [9-10], power systems [11], robotics [12-13] and wind ASPR system. In order to apply the SAC to the flexible
space structure system, Mehiel and Balas [17-18], Mufti [7]
and Bar-Kana, et al. [2,8,19] have done a lot of jobs to
*Corresponding author. Tel.: +86-10-82339275.
E-mail address: starsjxu@yahoo.com.cn
solve the ASPR problem. But those solutions are not
Foundation item: National Natural Science Foundation of China suitable for many other control problems, especially
(10902003) the system with higher relative degree. One needs a
1000-9361/$ - see front matter © 2012 Elsevier Ltd. All rights reserved.
design method which can expand the SAC to high rela-
doi: 10.1016/S1000-9361(11)60401-9 tive degree systems.
No.3 LIU Min et al. / Chinese Journal of Aeronautics 25(2012) 446-452 · 447 ·
It has been subsequently shown how various forms ° x p2 (t ) Ap2 xp2 (t ) Bp2 yp1 (t )
of parallel feedforward compensators (PFCs) [10,13,20] Sp2 : ® (2)
can be used to satisfy the ASPR conditions. However, °̄ yp2 (t ) C p2 xp2 (t )
with the addition of the feedforward compensator, the where the subsystem Sp1 is an unknown nonlinear sys-
output tracking error of the controlled system is no tem with relative degree mp1. The plant state xp1(t) is a
longer guaranteed to approach zero asymptotically, but np1×1 vector, up1(t) the mp1×1 control input, yp1(t) the
remains bounded. Besides, the design of an appropriate mp1×1 system output vector, and Ap1, Bp1 and Cp1 are
PFC accurate parameters of the original plant. Some-
unknown matrices while the plant {Ap1, Bp1, Cp1} is
times this may be unfeasible, because usually the SAC
ASPR. We assume throughout this paper that f (xp1, t)
method is adopted in plants with parameter uncertain-
is Lipchitz continuous. The subsystem Sp2 is a known
ties.
continuous linear time invariant system with relative
In this paper, based on backstepping control algo-
degree nmp1, and the subsystem Sp2 can be stabilized
rithm [21], a nonlinear backstepping simple adaptive
by an output feedback controller. xp2(t) is the np2×1
controller is derived for plants with arbitrary relative
degree. One of the advantages is that the proposed plant state vector, yp1(t) the mp1×1 control vector of
control algorithm is free of PFCs. The basic design subsystem Sp2, yp2(t) the mp1×1 output vector, and Ap2,
procedure consists of three steps. Firstly, the arbitrary Bp2 and Cp2 are known matrices with appropriate di-
relative degree unknown plant is decomposed into a mensions.
known subsystem and an unknown nonlinear subsys- The objective of control in this paper is to design the
tem with the linear part being ASPR system and the control input up1(t) to cause the output yp2(t) of the
nonlinear part being Lipchitz continuous. Then, the subsystem Sp2 to track the output of the continuous
output of the unknown subsystem is selected as the nonlinear time invariant reference model asymptoti-
virtual control vector and a constant feedback control cally which can also be decomposed into the two sub-
is designed to stabilize the known subsystem. Finally, systems connected in cascade as follows:
via adopting backstepping control method, the un- x m1 (t ) Am1 xm1 (t ) Bm1um1 (t ) g ( xm1 , t )
known subsystem is considered and the backstepping Sm1 : ®
simple adaptive controller is derived. The stability of ¯ ym1 (t ) C m1 xm1 (t )
the proposed controller is investigated by Lyapunov
(3)
stable theory and positive real theory.
To validate the efficiency of the controller, its appli- ° x m2 (t ) Ap2 xm2 (t ) Bp2 ym1 (t )
cation to large flexible space structure is studied by Sm2 : ® (4)
decomposing the large flexible space structure into the °̄ ym2 (t ) C p2 xm2 (t )
kinematics subsystem and the dynamics subsystem. where the reference subsystem Sm1 state xm1(t) is an
The ASPR property of the dynamics subsystem is con- nm1×1 vector with mp1×1 output ym1(t), and g(xm1,t) the
firmed, and then the backstepping simple adaptive nonlinear part of the Sm1. The subsystem Sm1 is de-
controller for flexible structures with collocated sen- signed to meet some desired performance properties
sors and actuators is given. Finally, a numerical exam- and has the same number of output as the subsystem
ple is presented. Sp1, but xm1(t) and xp1(t) need not to have the same di-
mensions, and it will be permissible to have
2. Problem Statement
np2 = dim( xm1 ) dim( xp1 ) np1 (5)
In many practical engineering problems, the con- where the operator “dim” denotes the dimension of the
trolled system can be described by subsystems in series. vector. um1(t) is mp1×1 control vector of reference
The most common case is that the controlled system is model and can be presented as the output of an com-
formed by a dynamics subsystem which describes the mand generating system of the form
evolution of the velocities as time progresses and a
known kinematics subsystem which describes the dis- vm (t ) Av vm (t )
® (6)
placement responses, and these subsystems are con- ¯um1 (t ) C v vm (t )
nected by velocities. For example, the flexible space
structures can be described by a dynamics subsystem where vm(t) is the command state vector. The matrices
and a simple known kinematics subsystem in cascade. Av and Cv are unknown, and only measurements of the
Even in a general case, one can use the method de- input um1(t) are permitted.
scribed in Ref. [22] to decompose the controlled sys- Since subsystem Sp2 is a known system, the subsys-
tem into subsystems. So we assume that the continuous tem Sm2 is selected the same as Sp2. xm2(t) is the np2×1
nonlinear time invariant controlled system with relative plant state vector, ym1(t) the mp1×1 control vector of
degree n can be decomposed into the two subsystems subsystem Sp2, and ym2(t) the mp1×1 output vector.
connected in cascade as follows:
° x p1 (t ) Ap1 xp1 (t ) Bp1up1 (t ) f ( xp1 , t ) 3. Backstepping Adaptive Controller
Sp1 : ® (1)
°̄ yp1 (t ) C p1 xp1 (t ) According to the backstepping control algorithm,
· 448 · LIU Min et al. / Chinese Journal of Aeronautics 25(2012) 446-452 No.3
V1 ep2
T
(t )( Ac2
T
P PAc2 )ep2 (t ) (13) Define the output error of subsystem Sp1 as
e (t ) yp1 (t ) p1 (t ) (20)
Since Sp2 can be made stable by output feedback,
there is a positive definite matrix Q which satisfies When yp1(t) differs from p1(t) at t=, asymptotic
tracking is achievable provided stabilizing output
Ac2
T
P PAc2 Q (14)
feedback is included in the control law. The control law
Substituting Eq. (14) into Eq. (13) results in
up1 (t ) K *x xm1 (t ) K u* um1 (t ) K e*e (t ) (21)
V1 ep2
T
(t )Qep2 (t ) d0 (15)
results in yp1(t) ė p1(t) ė ym1(t) and e(t) ė0, where
It is obvious that V1 0 if and only if ep2(t)=0, K e* is a stabilizing output feedback gain, and K e* always
namely the set E={ep2| ep2(t)=0} is the largest invariant exists because the subsystem Sp1 satisfies the ASPR
set contained in V1 0 . Then according to the LaSalle conditions. When asymptotic tracking occurs, the cor-
invariance principle of differential equation, when responding state and control input are defined to be the
yp1(t)=p1(t), under the intermediate control law de- asymptotic stable state trajectories and control trajecto-
scribed in Eq. (10), the output of subsystem Sp2 tracks ries. These trajectories are denoted by xp1 *'
(t )
ym2(t) asymptotically, that is ep2(t)ė0 and e2(t)ė0 as
and up1
*'
(t ) respectively, and the superscript “ * ' ” denotes
tė.
Then, backing a step, considering the subsystem Sp1, the asymptotic tracking condition. So the next equa-
backstepping control algorithm is adopted to get the tions satisfy:
adaptive controller for the controlled system. When x *' (t ) Ap1 xp1
*'
(t ) Bp1up1
*'
(t ) f ( xp1
*'
, t)
*' ° p1
perfect output tacking occurs (i.e., when yp1(t) = ym1(t) Sp1 :® (22)
°̄ yp1 (t ) C p1 xp1
*' *'
for tt), the corresponding state and control trajecto- (t )
ries are defined to be the ideal state and ideal control
and the error
trajectories, respectively. These ideal trajectories are
denoted by xp1 *
(t ) and up1
*
(t ) , where the superscript e* (t ) yp1
*'
(t ) p1 (t ) (23)
“ȗ” denotes the ideal tracking condition. By definition, vanishes asymptotically. The state tracking error is
these ideal trajectories satisfy Eq. (1), therefore, defined as
No.3 LIU Min et al. / Chinese Journal of Aeronautics 25(2012) 446-452 · 449 ·
° yield
¯ K e (t ) e (t )e (t ) e , K e (0) Ke0
T
V e T (t )( AT P P A )e (t )
2 p1 c1 2 2 c1 p1
where , x, u and e are positive definite adaptive (t ) 2tr( K T 1 K )
parameter matrices that determine the rate of adapta- 2ep1
T
(t ) P2 Bp1 Kr
tion, K0, Kx0, Ku0 and Ke0 are initial values. 2ep1
T
(t ) P2 'f (38)
f ( xp1 , t ) f ( xp1
*'
, t) (30) 2ep1
T
(t ) P2 'f (40)
f (t ) ym (t ) k ( (t ) m (t ))
°
°ef (t ) y (t ) f (t )
°
®u(t ) K f (t )rf (t ) K fx (t ) xm (t ) (60)
° K fu (t )um (t ) K fe (t )ef (t )
°
° K (t ) e (t )r T (t )
¯ f f f
trol including the Symposium. 1979; 347-351. tive realness of second order and almost second order
[2] Bar-Kana I, Kaufman H, Balas M. Model reference systems. AIAA-2005-5941, 2005.
adaptive control of large structural systems. AIAA [18] Mehiel E A, Balas M J. Determining strictly positive
Journal of Guidance 1983; 6(2): 112-118. realness from system modal characteristics. Journal of
[3] Balas M J. Finite-dimensional direct adaptive control Guidance, Control, and Dynamics 2007; 30(5): 1539-
for discrete-time infinite-dimensional linear systems. 1542.
Journal of Mathematical Analysis and Application 1995; [19] Bar-Kana I, Fischl R, Kalata P. Direct position plus
196: 153-171. velocity feedback control of large flexible space struc-
[4] Balas M J, Fuentes R, Mehiel E. Adaptive control and tures. IEEE Transactions on Automatic Control 1991;
rejection of persistent disturbances: continuous and 36(10): 1186-1188.
discrete-time. AIAA-2003-5323, 2003. [20] Iwai Z, Mizumoto I, Deng M. A parallel feedforward
[5] Balas M J. Adaptive control of nonlinear systems with compensator virtually realizing almost strictly positive
unknown delays and persistent disturbances. AIAA- real plant. Proceedings of the 33rd IEEE Conference on
2010-7537, 2010. Decision and Control. 1994; 2827-2832.
[6] Bar-Kana I, Kaufman H. Some applications of direct [21] Zheng M J. Spacecraft attitude control system design
adaptive control to large structural systems. Journal of via backstepping control method. PhD thesis, School of
Guidance 1984; 7(6): 717-724. Aeronautics, Beihang University, 2007. [in Chinese]
[7] Mufti I H. Model reference adaptive control for large [22] Dosho T, Ohmori H, Sano A. New design method of
structural systems. Journal of Guidance 1987; 10(5): simple adaptive control for systems with arbitrary rela-
507-509. tive degree. Proceedings of the 36th IEEE Conference
[8] Barkana I, Ben-Asher J Z. A simple adaptive control on Decision and Control. 1997; 1914-1917.
application to large flexible structures. 15th IEEE [23] Barkana I. On gain condition and convergence of sim-
Mediterranean Electrotechnical Conference. 2010; 806- ple adaptive control. AIAA-2003-5327, 2003.
811. [24] Wie B, Byun K W. New generalized structural filtering
[9] Mooij E. Model reference adaptive guidance for re- concept for active vibration control synthesis. Journal
entry trajectory tracking. AIAA-2004-4775, 2004. of Guidance 1989; 12(2): 147-154.
[10] Barkana I. Classical and simple adaptive control for
nonminimum phase autopilot design. Journal of Guid-
Biographies:
ance, Control, and Dynamics 2005; 28(4): 631-638.
[11] Barkana I, Fischl R. A simple adaptive enhancer of
voltage stability for generator excitation control. Pro- LIU Min received B.S. degree from Beihang University in
ceedings of the American Control Conference. 1992; 2007 and is now a Ph.D. student in the School of Astronaut-
1705-1709. ics, Beihang University. His main research interests are
[12] Ulrich S, Sasiadek J Z, Barkana I. Modeling and direct flexible spacecraft attitude control and simple adaptive con-
adaptive control of a flexible-joint manipulator. Journal trol.
of Guidance, Control, and Dynamics 2012; 35(1): 25- E-mail: liumin@sa.buaa.edu.cn
39.
[13] Kaufman H, Barkana I, Sobel K. Direct adaptive con- XU Shijie is a professor and Ph.D. supervisor at the School
trol algorithms. 2nd ed. New York: Springer, 1998. of Astronautics, Beihang University. He received the Ph.D.
[14] Johnson K E, Fingersh L J, Pao L Y, et al. Adaptive degree from Henri Poincaré University in 1995. His current
torque control of variable speed wind turbines for in- research interests are guidance, navigation, dynamics and
creased region 2 energy capture. AIAA-2005-392, control of spacecraft.
2005. E-mail: starsjxu@yahoo.com.cn
[15] Balas M J, Li Q, Peterman R. Adaptive disturbance
tracking control for large horizontal axis wind turbines
in variable speed region II operation. AIAA-2010-248, HAN Chao is a professor and Ph.D. supervisor at the School
2010. of Astronautics, Beihang University. He received the Ph.D.
[16] Frost S A, Balas M J, Wright A D. Modified adaptive degree from the same university in 1989. His current research
control for region 3 operation in the presence of wind interests are guidance, navigation, dynamics and control of
turbine structural modes. AIAA-2010-249, 2010. spacecraft.
[17] Mehiel E A, Balas M J. Some conditions for strict posi- E-mail: hanchao@buaa.edu.cn