IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO.

5, MAY 2013 1269

Sliding Mode and Active Disturbance Rejection Control where is the state, is the control input, is a constant
to Stabilization of One-Dimensional Anti-Stable Wave number. The unknown disturbance is supposed to be bounded mea-
Equations Subject to Disturbance in Boundary Input surable, that is, for some and all . The
system represents an anti-stable distributed parameter physical system
Bao-Zhu Guo and Feng-Fei Jin ([7]).
We proceed as follows. The SMC for disturbance rejection is pre-
sented in Section II. The ADRC for disturbance attenuation is presented
in Section III. Section IV presents some numerical simulations for both
Abstract—In this technical note, we are concerned with the boundary control methods.
stabilization of a one-dimensional anti-stable wave equation subject to
boundary disturbance. We propose two strategies, namely, sliding mode II. SLIDING MODE CONTROLLER
control (SMC) and the active disturbance rejection control (ADRC). The
reaching condition, and the existence and uniqueness of the solution for
all states in the state space in SMC are established. The continuity and A. Design of Sliding Surface
monotonicity of the sliding surface are proved. Considering the SMC usu- Following [13], we introduce a transformation:
ally requires the large control gain and may exhibit chattering behavior,
we then develop an ADRC to attenuate the disturbance. We show that this
strategy works well when the derivative of the disturbance is also bounded.
Simulation examples are presented for both control strategies.
(2)
Index Terms—Boundary control, disturbance rejection, sliding mode
control (SMC), wave equation.
This transformation brings system (1) into the following system:

I. INTRODUCTION (3)

In the last few years, the backstepping method has been introduced where is the design parameter. The transformation (2) is
to the boundary stabilization of some PDEs ([6], [13]). This powerful invertible, that is
method can also be used to deal with the stabilization of wave equa-
tions with uncertainties in either boundary input or in observation ([4],
[5]). Owing to its good performance in disturbance rejection and in-
sensibility to uncertainties, the sliding mode control (SMC) has also (4)
been applied to some PDEs, see [1], [2], [8], [9], [11], [12]. Other ap-
proaches that are proposed to deal with the disturbance include the Let us consider systems (1) and (3) in the state space
active disturbance rejection control (ADRC) ([3]), and the adaptive with inner product given by
control method for systems with unknown parameters ([6], [7]). In
the ADRC approach, the disturbance is first estimated in terms of the
output; and then canceled by its estimates. This is the way used in [4],
[5], [7]. Compared with the SMC, the ADRC has not been used in dis- (5)
tributed parameter systems.
Motivated mainly by [1], [11], we are concerned with the stabiliza- In this section, we consider as a real function space. In Section III,
tion of the following PDEs: is considered as the complex space.
Define the energy of system (3):
. Then
(1)

It is seen that in order to make non-increasing on the sliding sur-

Manuscript received February 01, 2012; revised August 17, 2012; accepted face for system (3), which is a closed subspace of , it is natural
September 03, 2012. Date of publication September 13, 2012; date of current to choose (so ), i.e.
version April 18, 2013. This work was supported by the National Natural
Science Foundation of China, the National Basic Research Program of China (6)
(2011CB808002), and the National Research Foundation of South Africa.
Recommended by Associate Editor X. Chen.
B.-Z. Guo is with the Academy of Mathematics and Systems Science, In this way, on , and on , system (3)
Academia Sinica, Beijing 100190, China and also with the School of Com- becomes
putational and Applied Mathematics, University of the Witwatersrand, South
Africa (e-mail: [email protected]).
F.-F. Jin is with the School of Computational and Applied Mathematics, (7)
University of the Witwatersrand, Wits 2050, Johannesburg, South Africa and
also with , the School of Mathematical Sciences, Qingdao University, Qingdao
266071, China (e-mail: [email protected]).
Color versions of one or more of the figures in this technical note are available It is well-known that for any initial value ,
online at http://ieeexplore.ieee.org. there exists a unique -semigroup solution
Digital Object Identifier 10.1109/TAC.2012.2218669 to

0018-9286/$31.00 © 2012 IEEE

1270 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO. 5, MAY 2013

(7), where the norm in is the induced norm of . Moreover, Then, satisfies formally that
system (7) is exponentially stable in , that is, there exist ,
independent of initial value such that

(8)

Transforming by (2) into the original system (1), that is

(16)

The closed-loop system of system (3) under the state feedback con-
(9) troller (15) is

we get the sliding surface for system (1) as

(17)
Note that (16) is just the well-known “reaching condition” for system
(10) (3) but we do not know if makes sense for the initial value in the
state space in present. This issue is not discussed in [1]. It would be
studied in Section III.
on which the original system (1) becomes

C. Solution of Closed-Loop System

(11)
In this section, we investigate the well-posedness of the solution to
(14). Since (14) and (17) are equivalent, and (17) takes a simpler form,
which is exponentially stable by (8) and the equivalence between (7) we study the solution of (17) outside the sliding surface.
and (11). Define an operator as follows:

B. State Feedback Controller

(18)
To motivate the control design, we differentiate (9) formally with
respect to to obtain
A direct computation shows that the adjoint operator of is given by

(19)

(12)
Take the inner product on both sides of (17) with
This suggests us to design the boundary controller to get

(13)

where and is defined by (9). Note that controller (13)

deals with the worst case of disturbance by the high gain . Under the
feedback (13), the closed-loop system of (1) reads where is the dual of with the pivot space . Then
system (17) can be written as

(14)

(20)

By the transformation (4), the corresponding controller for system

(3) is and is the Dirac distribution.
Proposition 1: Suppose that is bounded measurable in time. Then
for any , there exists a , depending
on initial value, such that (17) admits a unique solution
and for all . Moreover,
(15)
is continuous and monotone in .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO. 5, MAY 2013 1271

Proof: We first suppose that . In this case, it follows III. ACTIVE DISTURBANCE REJECTION CONTROL
from (17) that in the beginning of
It is well-known that the so-called chattering behavior is associated
with the SMC, due to discontinuity of control. In this section, we shall
(21) use a direct approach to attenuate rather than to reject the disturbance.
This is the key to the ADRC method in finite-dimensional systems
([3]). The idea is first to estimate the disturbance and then to cancel
Let be defined by (18). For any , it has the estimate in the feedback-loop. Unlike the SMC which usually uses
high gain control, the control effort in ADRC is found to be moderate.
In estimating the disturbance, we assume that the derivative of the
disturbance is bounded: for some and all .
Again, by equivalence, we discuss (3) only for it has a simpler form.
So is dissipative. Since by argument below, , the resolvent
The objective now is to design a continuous controller which can
set of , it follows that generates a -semigroup of contractions
stabilize system (3) in the presence of disturbance. In view of (15), this
on by the Lumer-Phillips theorem ([10, Theorem 4.3, p.14]).
controller is designed as follows:
Consider the observation problem of dual system of (20)

(22)
(27)

Then we can easily show that where , also continuous, is to be designed in what follows. Under
control (27), the closed-loop of system (3) becomes

(23) (28)

A straightforward computation shows that is bounded Introduce a variable . Then the boundary condition at
on . This together with (23) shows that is admissible for the in (28) gives that
-semigroup generated by ([14, Theorem 4.4.3, p.127]).
Therefore, system (20) admits a unique solution which satisfies, for (29)
all , that
It is seen that (29) is an ODEs with state and control . Then we
are able to design an extended state observer to estimate both and
as follows ([3]):

(30)

(24) where is the tuning small parameter. The errors ,

satisfy

Set in the first identity of (24) to obtain

(31)
(25)
which can be rewritten as
This shows that is continuous in the interval where . More-
over, (16) holds true. Therefore, there exists a such that
is monotone in and for all . In particular, if
, then . This completes the proof.
Returning to the original system (14) by the inverse transformation
(4), we obtain the first main result of this technical note.
Theorem 1: Suppose that is bounded measurable in time. Then (32)
for any , there exists some , depending
on initial value, such that (14) admits a unique solution A straightforward computation shows that the eigenvalues of the matrix
and for all . Moreover, is contin- are
uous and monotone in , where is the sliding surface
of system (14) determined by
(33)

(26) and

Any solution of (14) in the state space will reach the sliding surface
in finite time and remains on afterwards.
1272 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO. 5, MAY 2013

Since Proposition 2: Suppose that both and are uniformly bounded

in time. Then for any , there exists a unique
solution to (40). Moreover, can
reach arbitrary vicinity of zero as , in (36).
Proof: We consider only the case of . The case of
the first term above can be arbitrarily small as by the exponen- can be treated similarly. Introduce an auxiliary system as follows:
tial stability of , and the second term can also be arbitrarily small
as due to boundedness of and the expression of . As (41)
a result, can be arbitrarily small as .
Since is an approximation of , we can design the controller for
system (28) by cancelation/feedback as which can be rewritten as an evolution equation in

(34) (42)

under which the overall closed-loop system of (28), (29), and (30) be- where
comes

(43)
(35) It is a trivial exercise to show that is skew self-adjoint:
. Compute the eigenvalues , and the eigenfunctions
corresponding to of , to get

(44)
Using the error dynamics (31), we see that (35) is equivalent to

A direct computation shows that . Actually, we have

(36)

It is seen that in (36), the variable is independent of the “

part”, and can be made as small as desired as , . Thus,
we need to consider only the “ part” that is rewritten as
(45)
(37)
So is compact on by the Sobolev embedding theorem.
It follows from a general result in functional analysis that the eigen-
System (37) will be considered in the state Hilbert space functions form
also with the norm defined in (5). Define system an orthonormal basis for . Decompose as
operator of (37) by

(38)
By the norm defined in (5), , and
. Define a map
A direct computation gives

Then is an isometric from to . It is further seen that

, , where
(39)
Similar to (20), system (37) can be rewritten as an evolution equation
in as (46)

(40) Since form an orthonormal

basis for , it is equivalent to saying that
where is defined in (20). form an orthonormal basis for .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO. 5, MAY 2013 1273

Now we go back to system (40). When , the eigenvalues

and the eigenfunctions corresponding to
of are found to be

(53)
(47)
Since generates an exponential stable -semigroup and is ad-
missible to , for any initial value , it
follows that there exists a unique solution to (40) provided that
It is obvious that . Let
, which can be written as ([15])
for . Then it is found that

(54)
(48)
This is the first part of the theorem. Now we show the second part.
Hence, form a Riesz basis for For any given , by assumption, we may assume that
. Hence form a Riesz for all , for some and . We can
basis for . Since write (54) as
is -linearly independent and is quadratically close to
, it follows from the classical
Bari’s theorem that
form a Riesz basis for . This implies that generates a -semi-
group on , and that the spectrum-determined growth condition
is true for . Moreover, since ,
is exponentially stable. (55)
Next, we show that is admissible for , or equivalently, is
admissible for . To this end, we find the dual system of (40) to be
Since the admissibility of implies that

(49)
(56)

for some constant that is independent of , and since is expo-

Define functions nential stable, it follows from Proposition 2.5 of [15] that

(50) (57)

and where is a constant that is independent of , and

(51) (58)

Similar to (23), we can show from (50) and (51) that Suppose that for some , . By (55), (56),
and (57), we have

(59)
(52)
As , the first two terms of (59) tend to zero. The result is then
proved by the arbitrariness of .
is then admissible for if we can show that is bounded Remark 1: When , instead of (41), we choose the corre-
in . This is trivial since sponding auxiliary system as follows:

(60)
1274 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 58, NO. 5, MAY 2013

Fig. 1. Displacements for disturbance with unbounded derivative (a) SMC (b) Fig. 3. Displacements for disturbance with bounded derivative (a) SMC (b)
ADRC. ADRC.

smoothly, but system (62) is oscillatory around the equilibrium before

. It shows that ADRC is not adequate to deal with the disturbance
with unbounded derivative. The corresponding control and sliding sur-
face are plotted in Fig. 2 in this case.
If the disturbance is described as . Then both and
are uniformly bounded. Take the steps of space and time by 0.005 and
0.001, respectively, which are larger than that in Fig. 1. We obtain the
displacements of the system (14) and (62), which are shown in Fig. 3(a)
and (b), respectively.
We point out that a chattering behavior is observed in Fig. 3(a) (see
Fig. 2. SMC for disturbance with unbounded derivative (a) Sliding surface (b) also the sliding surface in Fig. 2(a)), although it is convergent. On the
Controller. other hand, the displacement in Fig. 3(b) is quite smooth. The results
shows that the ADRC yields more satisfactory performance than the
Going back to the original system, we have the second main result SMC in dealing with the disturbance with bounded derivative.
of this technical note.
Theorem 2: Suppose that and are uniformly bounded measurable REFERENCES
and . Let [1] M. B. Cheng, V. Radisavljevic, and W. C. Su, “Sliding mode
boundary control of a parabolic PDE system with parameter variations
and boundary uncertainties,” Automatica, vol. 47, no. 2, pp. 381–387,
2011.
[2] S. Drakunov, E. Barbieri, and D. A. Silver, “Sliding mode control of a
heat equation with application to arc welding,” in Proc. IEEE Int. Conf.
(61) Control Appl., 1996, pp. 668–672.
[3] B. Z. Guo and Z. L. Zhao, “On the convergence of extended state ob-
server for nonlinear systems with uncertainty,” Syst. Control Lett., vol.
Then the closed-loop system of (1) described by 60, no. 6, pp. 420–430, 2011.
[4] W. Guo, B. Z. Guo, and Z. C. Shao, “Parameter estimation and sta-
bilization for a wave equation with boundary output harmonic distur-
bance and non-collocated control,” Int. J. Robust Nonlin. Control, vol.
21, no. 11, pp. 1297–1321, 2011.
[5] W. Guo and B. Z. Guo, “Stabilization and regulator design for a one-
dimensional unstable wave equation with input harmonic disturbance,”
Int. J. Robust Nonlin. Control, to be published.
[6] M. Krstic and A. Smyshlyaev, Boundary Control of PDEs: A Course
on Backstepping Designs. Philadelphia, PA: SIAM, 2008.
(62) [7] M. Krstic, “Adaptive control of an anti-stable wave PDE,” Dynamics
of Continuous, Discrete and Impulsive Systems Series A: Mathematical
admits a unique solution , and can Analysis, vol. 17, no. 6, pp. 853–882, 2010.
reach arbitrary vicinity of zero as , in (62). [8] Y. V. Orlov and V. I. Utkin, “Use of sliding modes in distributed
system control problems,” Automat. Remote Control, vol. 43, no. 9,
IV. NUMERICAL SIMULATION pp. 1127–1135, 1982.
[9] Y. V. Orlov and V. I. Utkin, “Sliding mode control in infinite-dimen-
In this section, we give some simulation results to illustrate the ef- sional systmes,” Automatica, vol. 23, no. 6, pp. 753–757, 1987.
fects of both the SMC and ADRC. [10] A. Pazy, Semigroups of Linear Operators and Applications to Partial
Differential Equations. New York: Springer-Verlag, 1983.
Consider systems (14) with the equivalent control when [11] A. Pisano, Y. Orlov, and E. Usai, “Tracking control of the uncertain
, and (62). Let the parameters be , , heat and wave equation via power-fractional and sliding-mode tech-
, , , and the disturbance . The niques,” SIAM J. Control Optim., vol. 49, no. 2, pp. 363–382, 2011.
initial conditions are [12] V. I. Utkin, “Sliding mode control: Mathematical tools, design and
applications,” in Nonlinear and Optimal Control Theory, ser. Lecture
Notes in Math.. Berlin, Germany: Springer-Verlag, 2008, vol. 1932,
(63) pp. 289–347.
[13] A. Smyshlyaev and M. Krstic, “Boundary control of an anti-stable
Note that is bounded but is unbounded. wave equation with anti-damping on uncontrolled boundary,” Systems
We apply the finite difference method to compute the displacement. Control Lett., vol. 58, no. 8, pp. 617–623, 2009.
[14] M. Tucsnak and G. Weiss, Observation and Control for Operator Semi-
Fig. 1(a) and (b) show the displacements of system (14) and (62) re-
groups. Basel, Switzerland: Birkhäuser, 2000.
spectively. Here the steps of space and time are taken as 0.001 and [15] G. Weiss, “Admissibility of unbounded control operators,” SIAM J.
0.0005, respectively. It is seen from Fig. 1 that system (14) converges Control Optim., vol. 27, no. 3, pp. 527–545, 1989.