Professional Documents
Culture Documents
Automatica
journal homepage: www.elsevier.com/locate/automatica
Technical communique
shows that the same accuracy for the sliding mode motion can Under this reaching law, the sliding mode state will enter the
be obtained as that of the equivalent-control-based DSMC, who vicinity of sliding mode surface s(k) = 0 with thickness ∆ and stay
provides a higher precision than that of the traditional reaching- there forever. That is, |s(k)| ≤ ∆ for all k > k∗ with a constant k∗ .
law-based DSMC. Finally, an example is provided to demonstrate Clearly, the size of ∆ is important, which determines the accuracy
the potential of the proposed method. of sliding mode motion and eventually the steady-state of system
state. Solving Eq. (4) based on (2) and (3) leads to
2. System description and some existing results u(k) = −(C Γ )−1 [C Φ x(k) − (1 − qh)s(k)
+ ε h · sgn(s(k)) + Cd(k)]. (5)
2.1. System description
Due to the unavailable information d(k), as that in Su et al. (2000),
As in Wang et al. (2009) and Su et al. (2000), consider the the disturbance can be estimated by using the so called delay
following single-input continuous-time system with matched estimate method, i.e.,
disturbances:
d̂(k) = x(k) − Φ x(k − 1) − Γ u(k − 1) = d(k − 1), (6)
ẋ = Ax + Bu u + Bd f , (1)
which yields the final realizable controller
where x ∈ Rn , u ∈ R, and f ∈ R are the state, input, and u(k) = −(C Γ )−1 [C Φ x(k) − (1 − qh)s(k)
disturbance, and A, Bu , Bd are constant matrices of appropriate
dimensions. The matching condition implies that rank[Bu , Bd ] = + ε h · sgn(s(k)) + C d̂(k)]. (7)
rank[Bu ]. The disturbance f is assumed to be smooth and bounded.
Under this controller, it can be concluded that
Assume the control law u is digitally implemented through
a zero-order-holder (ZOH), i.e., u(t ) = u(k) over the time in- s(k + 1) = (1 − qh)s(k) − ε h · sgn(s(k)) + C δ(k). (8)
terval [kh, (k + 1)h), where h is the sampling period. Then, the
By Lemma A.1 and Assumption 2.1, it can be concluded that the
continuous-time system (1) under the discrete-time control law
sliding mode state s(k) will enter the following region in a finite
u(k) can be written in a discrete form as follows:
number of steps and stay there forever
x(k + 1) = Φ x(k) + Γ u(k) + d(k), (2)
{s(k) : |s(k)| ≤ ε h + δ ∗ = O(h)}. (9)
where Φ = eAh , Γ = 0 eAτ dτ Bu , d(k) = 0 eAτ Bd f ((k + 1)h −
h h
τ )dτ . By Abidi et al. (2007), the system matrices of discrete-time Remark 2.2. The main problem of this method is to use the
system (2) have the following properties. switching function sgn(s(k)), which leads to the well-known
A2 h2
chattering problem in practice. And from the control law (7), it can
Property 2.1. (1) Φ = I + Ah + 2!
+ · · · = I + O(h); be found that the switching gain, i.e., the amplitude of chattering,
ABu h2 is (C Γ )−1 ε h. Since C Γ = O(h), then (C Γ )−1 ε h is in the order of
(2) Γ = Bu h + + · · · = O(h);
2!
(3) d(k) = O(h); O(1).
(4) The change rate of the disturbance δ(k) = d(k) − d(k − 1) =
O(h2 ). 2.2.2. Discrete-time SMC based on equivalent control
To avoid the chattering problem, another method called
2.2. Problems of the existing discrete-time SMC equivalent control is proposed in Su et al. (2000) to design a
chatter-free DSMC law. Directly solving s(k + 1) = 0, i.e., Cx(k +
To design a discrete-time SMC law, as that in Su et al. (2000), a 1) = 0 leads to
discrete-time switching function is usually chosen as:
u(k) = −(C Γ )−1 [C Φ x(k) + Cd(k)]. (10)
s(k) = Cx(k), (3)
As that in (7), the disturbance d(k) can be substituted by the
where C ∈ R 1×n
is a constant matrix and is chosen such that C Γ is estimate value d̂(k) defined in (6), which yields the final realizable
invertible and the matrix (Φ − Γ (C Γ )−1 C Φ ) has 1 zero pole and controller
(n − 1) poles inside the unit disk in the complex z-plane. Based on u(k) = −(C Γ )−1 [C Φ x(k) + C d̂(k)]. (11)
matrix C and Property 2.1(4), the following assumption is imposed.
Under this discrete-time controller, the dynamical behavior of
Assumption 2.1. There is a constant δ ∗ such that |C δ(k)| ≤ δ ∗ , sliding mode state is given by
k = 0, 1, 2, . . . , and the constant δ ∗ is in the order of O(h2 ).
s(k + 1) = Cx(k + 1) = C Φ x(k) + C Γ u(k) + Cd(k)
Remark 2.1. This Assumption is satisfied if the change rate of the = C [d(k) − d̂(k)] = C δ(k). (12)
disturbance f (i.e., f˙ (t )) for the system (1) is bounded, whose
proof can be found in Lemma 1 in Abidi et al. (2007). In practice, By Assumption 2.1, s(k) is bounded by
many kinds of disturbances satisfy this property, such as the load |s(k)| ≤ δ ∗ = O(h2 ). (13)
disturbances for permanent magnet synchronous motor (Liu & Li,
2012), etc.
Remark 2.3. Compared with the reaching-law-based DSMC, the
2.2.1. Discrete-time SMC based on reaching process sliding mode state s(k) under the equivalent-control-based DSMC
has a higher accuracy. However, the magnitude of equivalent-
Similar to the continuous-time SMC, a method called reaching-
control-based DSMC law u(k) in (11) is in the order of O(h−1 )
law-based DSMC approach is firstly introduced in Gao et al. (1995).
The reaching law for sliding mode is designed as: because (C Γ )−1 = O(h−1 ), C Φ = O(1), and C d̂(k) = Cd(k − 1) =
O(h) from Property 2.1(3). Clearly, if h is sufficiently small, the
s(k + 1) = s(k) − qh · s(k) − ε h · sgn(s(k)), equivalent-control-based DSMC needs an over-large control input
with ε > 0, q > 0, 0 < 1 − qh < 1. (4) which may be undesirable due to the control saturation constraint.
H. Du et al. / Automatica 68 (2016) 87–91 89
u(k) =−(C Γ )−1 [C Φ x(k) − s(k) + q1 h · s(k) (1 − q1 h)s1−α (k) > ψ 1−α (α)q2 h. (22)
Proof. By (2) and (3), we obtain s2 (2) ≤ s2 (1) − µ2 ≤ s2 (0) − 2µ2 , . . . , s2 (m) ≤ s2 (0) − mµ2 .
(26)
s(k + 1) = C [Φ x(k) + Γ u(k) + d(k)]. (17)
By directly solving s (0) − mµ = ρ , it can be obtained that the
2 2 2
Under the proposed DSMC law (14), we obtain real number solution is m∗ defined in (16). That is to say that after
K ∗ = [m∗ ] + 1 steps, |s(K ∗ )| ≤ ρ.
s(k + 1) = s(k) − q1 hs(k) − q2 hsigα (s(k)) + C δ(k). (18)
Next, we will prove the fact that if s(k) ∈ Ω , then s(k + 1) ∈ Ω .
Next, we will show that the sliding mode state s(k) will enter the 1/α 1
1−α
δ∗
region Ω in a finite number of steps. Since there is a fractional For the sake of brevity, denote Φ = max q2 h
, 1−q2qh1 h ,
power term, the Lyapunov analysis method is employed. Construct which means that ρ = ψ(α)Φ . Then we have the following
a Lyapunov function V (k) = s2 (k). From (18), it can be obtained proposition (see Appendix for proof).
that
Proposition 1. δ ∗ ≤ q2 h · Φ α ≤ (1 − q1 h)Φ .
△V (k) = V (k + 1) − V (k)
Since −ρ ≤ s(k) ≤ ρ , we can assume that
= − q1 h · s(k) + q2 h · sigα (s(k)) − C δ(k)
s(k) = θ · ρ = θ ψ(α)Φ , −1 ≤ θ ≤ 1. (27)
× 2s(k) − q1 h · s(k) − q2 h · sigα (s(k)) + C δ(k) . (19) By (18), it can be obtained that
Now, we consider the two cases when s(k) ̸∈ Ω . s(k + 1) = (1 − q1 h)θ ρ − q2 hsigα (θ ρ) + C δ(k)
1/α 1
1−α ≤ (1 − q1 h)ψ(α)θ Φ − q2 hsigα (ψ(α)θ )Φ α + δ ∗ . (28)
δ∗
Case 1: s(k) > ρ = ψ(α) · max , q2 h
q2 h 1−q1 h
. Using Proposition 1, we get when ψ(α)θ ≥ 0, then
In this case, sigα (s(k)) = sα (k). Since
s(k + 1) ≤(1 − q1 h)ψ(α)θ Φ − (ψ(α)θ )α δ ∗ + δ ∗ .
1/α
s(k) > ρ = ψ(α) δ ∗ /(q2 h) , (20) If ψ(α)θ ≥ 1, then s(k + 1) ≤ (1 − q1 h)ψ(α)θ Φ ≤ ψ(α)Φ = ρ .
If 0 ≤ ψ(α)θ < 1, then s(k + 1) ≤ [1 + ψ(α)θ − (ψ(α)θ )α ](1 −
q1 h)Φ ≤ (1 − q1 h)Φ < ψ(α)Φ = ρ. When ψ(α)θ ≤ 0, it follows
from Proposition 1 that
1 For the sake of statement, the notation [m∗ ] denotes the maximal integer
bounded below the real number m∗ . s(k + 1) ≤−(1 − q1 h)|ψ(α)θ |Φ + |ψ(α)θ |α (1 − q1 h)Φ + δ ∗ .
90 H. Du et al. / Automatica 68 (2016) 87–91
Remark 3.2. Note that the boundary for the sliding mode state s(k)
is determined by the value ρ . Since δ ∗ = O(h2 ) by Assumption 2.1,
it follows from (15) that
1/α 1
1−α
ρ = ψ(α) · max O(h) , O(h) . (30)
To obtain the smallest boundary for the final sliding mode motion,
it is better to choose α = 1/2 such that 1/α = 1/(1 − α) = 2,
which leads to ρ = O(h2 ). That is to say that sliding mode motion
can reach the same level of accuracy as that of the equivalent-
control-based DSMC. In addition, for the proposed new DSMC law
(14), since C Φ x(k) − s(k) = C (Φ − I )x(k), it can be found that the Fig. 2. Response curves of the closed-loop system under the reaching-law-based
magnitude of u(k) will be in the order of O(1), which avoids the DSMC with parameters q1 = ε = 5.
need of over-large control effort if h is sufficiently small. Since that
there is no switching term, there is no chattering phenomenon in is x(0) = [−1 0]T . The sampling period is selected as: h = 1
the control signal. ms. The matrix C is chosen as: C = [0.5 0.5]. For comparison,
three kinds of DSMC methods are employed, i.e., reaching-law-
Remark 3.3. Note that the steady-state for system’s state is not based DSMC, equivalent-control-based DSMC, and non-smooth-
discussed in this paper. Actually, by substituting the three kinds control-based DSMC. The response curves are given in Figs. 1–3. It
of DSMC laws, i.e., (7), (11) and (14), into the discrete-time system is shown that the non-smooth-control-based DSMC considers both
(2), it can be shown that the nominal part of closed-loop system the performance requirement and the control signal quality.
i.e.,
x(k + 1) = (Φ − Γ (C Γ )−1 C Φ )x(k), 4. Conclusions
is the same. The difference lies in the part from the sliding A new chattering-free DSMC method based on non-smooth
mode state. Therefore different steady-state of sliding mode state control has been presented. It has shown that there is no chattering
directly affects the final steady-state for system’s state, which is phenomenon and no need for an overly large control effort in
shown in the following example. the proposed DSMC. By synthesizing the benefit of non-smooth
control, the high precision for sliding mode motion can be kept
Example. Consider the high-precision position control for a under the proposed DSMC, who provides a new choice for the
piezomotor-driven linear stage system (Abidi, Xu, & She, 2009), design of DSMC law.
whose dynamical equation is:
Appendix
kf kv
ẋ1 = x2 , ẋ2 = u− x2 + f , (31)
m m This Appendix collects some useful lemmas. The first lemma is
where x1 is the linear displacement, x2 is the linear velocity, u is the a simple extension of Lemma 2 of Yu et al. (2008), whose proof is
voltage input, and f is an unknown disturbance. In the simulation, similar and is omitted here.
the parameters are chosen as: m = 1, kv = 144, kf = 6, and the Lemma A.1. For the scalar dynamical system z (k + 1) = a · z (k) +
unknown disturbance is given by: f (t ) = 1 for 0 ≤ t ≤ 10 and g (k)−ε · sgn(z (k)) with 0 ≤ a ≤ 1, if |g (k)| < γ , 0 ≤ γ < ε , there
f (t ) = 1 + 2.2 sin(0.5π t ) for 10 ≤ t ≤ 20. The initial condition is a finite number K ∗ > 0 such that |z (k)| ≤ ε + γ < 2ε , ∀k ≥ K ∗ .
H. Du et al. / Automatica 68 (2016) 87–91 91
References
Abidi, K., Xu, J., & She, J. (2009). A discrete-time terminal sliding-mode control
approach applied to a motion control problem. IEEE Transactions on Industrial
Electronics, 56(9), 3619–3627.
Abidi, K., Xu, J., & Yu, X. (2007). On the discrete-time integral sliding-mode control.
IEEE Transactions on Automatic Control, 52(4), 709–715.
Bhat, S. P., & Bernstein, D. S. (2000). Finite-time stability of continuous autonomous
systems. SIAM Journal on Control and Optimization, 38(3), 751–766.
Drazenovic, B. (1969). The invariance conditions in variable structure systems.
Automatica, 5(3), 287–295.
Galias, Z., & Yu, X. (2007). Euler’s discretization of single input sliding-mode control
systems. IEEE Transactions on Automatic Control, 52(9), 1726–1730.
Gao, W., Wang, Y., & Homaifa, A. (1995). Discrete-time variable structure control
systems. IEEE Transactions on Industrial Electronics, 42(2), 117–122.
Li, S., Du, H., & Yu, X. (2014). Discrete-time terminal sliding mode control systems
based on Euler’s discretization. IEEE Transactions on Automatic Control, 59(2),
546–552.
Liu, H., & Li, S. (2012). Speed control for pmsm servo system using predictive
functional control and extended state observer. IEEE Transactions on Industrial
Electronics, 59(2), 1171–1183.
Perruquetti, W., & Barbot, J. P. (2002). Sliding mode control in engineering. New York:
Fig. 3. Response curves of the closed-loop system under the equivalent-control- Marcel Dekker.
based DSMC. Qu, S., Xia, X., & Zhang, J. (2014). Dynamics of discrete-time sliding-mode-
control uncertain systems with a disturbance compensator. IEEE Transactions
α 1 on Industrial Electronics, 61(7), 3502–3560.
Lemma A.2 (Li et al., 2014). If function ψ(α) = 1 + α 1−α − α 1−α Su, W. C., Drakunov, S. V., & Ozguner, U. (2000). An O(T 2 ) boundary layer in sliding
and 0 < α < 1, then 1 < ψ(α) < 2 and for any x ∈ [0, 1] mode for sampled-data systems. IEEE Transactions on Automatic Control, 45(3),
482–485.
xψ(α) − xα ψ(α)α + ψ(α) − 1 ≥ 0. Utkin, V. I. (1992). Sliding modes in control optimization. Berlin: Springer-Verlag.
Utkin, V. (1994). Sliding mode control in discrete-time and difference systems. In
A. S. I. Zinober (Ed.), Variable structure and Lyapunov control: Vol. 193
Proof of Proposition 1. Two cases will be considered. (pp. 87–107). New York: Springer-Verlag.
1/α 1
1−α Wang, B., Yu, X., & Chen, G. (2009). ZOH discretization effect on single-input
First, if δ ∗ /q2 h ≤ q2 h/(1 − q1 h) = Φ , then sliding mode control systems with matched uncertainties. Automatica, 45(1),
118–125.
Xia, X., & Zinober, A. S. I. (2006). Delta-modulated feedback in discretization of
q h 1−α α sliding mode control. Automatica, 42(5), 771–776.
= q2 h · Φ α = (1 − q1 h)Φ .
2
δ ∗ ≤ q2 h (A.1) Yu, X., Wang, B., Galias, Z., & Chen, G. (2008). Discretization effect on equivalent
1 − q1 h control-based multi-input sliding-mode control systems. IEEE Transactions on
Automatic Control, 53(6), 1563–1569.
∗ 1/α 1
1−α Yu, X., Wang, B., & Li, X. (2012). Computer-controlled variable structure
Second, if Φ = qδ h , then
q2 h
2
≥ 1−q1 h systems: the state-of-the-art. IEEE Transactions on Industrial Informatics, 8(2),
197–205.
Yu, X., & Xu, J. X. (2002). Lecture notes in control and information sciences: Vol. 274.
δ ∗ = q2 h · Φ α , and (δ ∗ )1−α ≥ q2 h/(1 − q1 h)α , (A.2) Variable structure systems: Towards the 21st century. Berlin: Springer-Verlag.
Yu, S., Yu, X., Shirinzadeh, B., & Man, Z. (2005). Continuous finite-time
1/α control for robotic manipulators with terminal sliding mode. Automatica, 41,
which leads to δ ∗ ≤ (1 − q1 h) qδ h
∗
2
= (1 − q1 h)Φ . 1957–1964.