Output Feedback Stabilization for a 1-D Conservative Wave Equation with General Corrupted Boundary Observation ()
1. Introduction
The stabilization of strings and flexible beams is always an important research direction in recent decades, see [1] [2] [3] [4], to name just a few. When actuators and sensors are collocated, system can be stabilized by utilizing passive principle [5] [6]. Compared with the collocated case, the non-collocated stabilization problem is more difficult because the passivity principle can not be used. However, non-collocated case is more widely used than collocated case in engineering (see, e.g., [7] [8]). With the proposal of the backstepping approach, this method has been extensively used in stabilization problem for parabolic equations [9] [10] [11], first-order hyperbolic equations [12] [13] [14], wave equations [15] [16] [17] and other partial differential equations [18] [19] [20]. In [21], in order to stabilize an unstable wave equation, using the backstepping method, not only the collocated Dirichlet boundary control but also the non-collocated Neumann boundary control is considered. The strong stabilization of unstable wave equation by using non-collocated boundary displacement can be found in [22]. Then in [23], the stabilization of unstable wave equation with Neumann boundary control can be achieved by using only collocated boundary displacement. Good progress has been made in [24], where the finite-time stabilization of 1-D wave equation by using only non-collocated boundary displacement is considered.
When there exist external disturbances or unknown internal nonlinear uncertainties, there are some methods to achieve stabilization. Sliding mode control is applied in [25] to stabilize a 1-D wave equation with nonlinear van der Pol type boundary condition that covers the anti-stable boundary, and subject to boundary control matched disturbance on the other side. Adaptive method is used in [26] to study the stabilization problem of an unstable wave equation, in which the boundary observation is suffered with a harmonic disturbance. However, the above methods are not applicable when considering the stabilization of a 1-D wave equation with corrupted boundary observation by general disturbance. ADRC plays an important role in solving the stabilization problem with general corrupted boundary disturbance. In [27], ADRC is the first time adopted to set up an ordinary differential equation disturbance estimator to estimate the disturbance, in which the designed disturbance estimator is only dependent on the output of the original system.
In this paper, we consider the stabilization of the following 1-D conservative wave equation
(1.1)
where and henceforth
or
is the derivative of y with respect to s and
or
the derivative with respect to t.
is the boundary input,
is the boundary output,
and
are initial values and suppose that
is a differentiable external disturbance. The major concern for this kind of output is that the velocity is relatively difficult to measure. If there is no disturbance in the velocity measurement, it is easy to see that system (1.1) can be exponentially stabilized directly by an output feedback controller.
The purpose of this paper is to use ADRC approach to stabilize (1.1) through the output of (1.1). Our method is more general than [28], where the stabilization of (1.1) with an infinite dimensional exosystem periodic disturbance is studied. Accurately speaking, in [28], the boundary velocity measurement with the disturbance has the following form
(1.2)
in which
denote Fourier coefficients and period
. In [28],
is an output of an exosystem, and then the stabilization of the system coupled by the original system and the exosystem is considered. However, this method can only be used to solve the periodic disturbance which can be written as an output of external system. It may not be suitable for more general periodic disturbance. In addition, when
, the method is not applicable.
The organization in this paper is as follows. In Section 2, a disturbance estimator is designed to online estimate disturbance, then we verify the convergence of the error system. In Section 3, an observer-based law is designed and the closed-loop system is verified to be asymptotically stable. In Section 4, some numerical simulations are provided.
2. Estimator and Observer Design
In this section, we design a disturbance estimator to estimate the disturbance
, then establish an observer in terms of the designed disturbance estimator. Suppose
and
(2.1)
in which
, and for any
satisfies
(2.2)
Same as the disturbance estimator in [27], we design the disturbance estimator as
(2.3)
where
is the initial value of estimator,
is an approximation of
. Then, the observer of (1.1) is designed according to disturbance estimator (2.3) as follows
(2.4)
where
is a constant. We consider the observer (2.3) and (2.4) in the space
,
. Let
(2.5)
It is easy to see that the error system is governed by
(2.6)
It can be known that the ODE-part of (2.6) is independent of its PDE-part and its well-posedness and convergence has been proved in [27]. Therefore, for brevity, the proof of the well-posedness and convergence of the ODE-part is omitted in this paper, we only need to consider the well-posedness and convergence of the PDE-part. We consider the PDE-part of (2.6) in the space
with the normal inner product.
Define an operator
as
(2.7)
As we all know,
generates an exponentially stable
-semigroup
on
. The dual operator is
(2.8)
Take the inner product of
with the PDE-part of (2.6) to obtain
(2.9)
in which
being Dirac distribution and
(2.10)
Hence, the PDE-part of (2.6) is equivalent to
(2.11)
or
(2.12)
where
.
Theorem 2.1. For arbitrary initial datum
, the PDE-part of (2.6) has a unique solution
, and for arbitrary
, there is a
depending on T only such that
Proof. By [29], it only needs to verify that
is admissible for
, which is equal to say (a)
is bounded from
to
, and (b) for arbitrary
, there is a
only depends on T, which makes
(2.13)
satisfies
where
A simple calculation obtains
(2.14)
Thus,
is bounded on
. Then, by differentiating
we get
integrating both sides yields
.
Theorem 2.2. Suppose that
and
. Then, the observers (2.3) and (2.4) are well-posed, i.e., for arbitrary initial date
, (2.3) and (2.4) have a unique solution
. Furthermore, if we also suppose that
and
satisfy (2.2) and (2.1), respectively, then the solution of (2.3) and (2.4) satisfy
(2.15)
Proof. Based on Theorem 2.1, the solution of (2.12) depends on the initial date and
. For arbitrary
, for some
, we can assume that
for all
. Thus, we can write the solution of (2.12) as
(2.16)
It follows from the admissibility of
and [29] that
(2.17)
in which L is a constant independent of
, and
Since
is exponential stable, there are two constants
and
, hold in
Thus, we have
(2.18)
Rewriting (2.16) as
(2.19)
according to (2.17) and (2.18), we obtain
(2.20)
Take
on both sides of (2.20) to get
(2.21)
Because
is arbitrarily selected, we get
(2.22)
Combining (2.5), we have
(2.23)
(2.15) then can be obtained by (2.23) and [27].
Remark 2.1. If we only consider using displacement to stabilize (1.1), i.e., use
only. Inspired by [24], let
(2.24)
thus,
is determined by
(2.25)
The observer of (2.25) is constructed as
(2.26)
Let
be the error. Therefore, the error
is determined by
(2.27)
As we all know, system (2.27) is well-posed and can be finite-time stable:
as
( [24]), i.e.,
as
. Therefore, if the controller is presented like
, then when
, we have
, which will use the velocity measurement
of system (2.25). With (2.24), we will use the velocity measurement
of system (1.1). Thus, the stabilization of (1.1) can’t use only the displacement measurement
.
3. Well-Posedness and Stability of Closed-Loop System
The closed-loop system consists of system (1.1), observer (2.3) and (2.4) in the state space
. Based on the observer we designed, we can apply the same controller as in [30]
(3.1)
where m is a normal constant. Therefore, the closed-loop system is
(3.2)
Theorem 3.1. Assume that
and
satisfy (2.2) and (2.1), respectively. For arbitrary initial date
, system (3.2) has a unique solution
and (3.2) is asymptotically stable, i.e.,
(3.3)
Proof. By the invertible transformation
(3.4)
then the PDE-part of (3.2) is equivalent to
(3.5)
then system (3.5) can be written as
(3.6)
where
(3.7)
and the operator
is defined by
(3.8)
As we all know,
generates an exponential stable
-semigroup
on
. Along the same line for (2.7) to (2.23), we can obtain that
in (3.9) is admissible for
and
(3.9)
Combining (3.4), we have
(3.10)
Then, the stability of the closed-loop system is obtained by Theorem 2.1, Theorem 2.2 and (3.10).
4. Simulation Results
In this section, some simulations are carried out for open-loop (1.1) and closed-loop (3.2). In (1.1) and (3.2), the initial date is selected as follows
(4.1)
For system (1.1) and system (3.2), we use the finite element method to calculate their solutions. The system (1.1) is conservative, which is presented in Figure 1. If we give a (3.1) controller at the
endpoint, the system will be asymptotically stable, it can be seen in Figure 2. We can see from Figure 3 that the designed observer is convergent. In Figure 4,
can approximate converge to
well, and when time tends to infinity, both
and
converge to 0, which indicates that the established disturbance estimator has satisfactory convergence.
(a) (b)
Figure 1. Displacement of open-loop system (1). (a) Displacement of
; (b) Displacement of
.
(a) (b)
Figure 2. The y-part displacement of closed-loop system (31). (a) Displacement of
; (b) Displacement of
.
(a) (b)
Figure 3. The
-part displacement of closed-loop system (31). (a) Displacement of
; (b) Displacement of
.
Figure 4. Trajectory of the ODE-part of (31).
5. Concluding Remarks
In this paper, the problem of stabilization for a 1-D conservative wave equation is studied. The difficulty in this paper is that the boundary velocity observation is affected by a general disturbance. The merit of our method lies in that the boundary velocity observation is subjected to a general disturbance, including constant disturbance and periodic disturbance as its special cases. If there is no collocated boundary displacement measurement, it seems that only using the corrupted collocated boundary velocity measurement cannot estimate the disturbance, which is a disadvantage of this paper. In future studies, we will extend this method to the Schrödinger equation and the Euler-Bernoulli equation.