Professional Documents
Culture Documents
Articulo - Sliding Mode Stability
Articulo - Sliding Mode Stability
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 1
Abstract—This article deals with robust fixed-time stability towards the equilibrium which is a desirable property
and stabilization. First, new global robust fixed-time stability for engineering applications. In particular, fixed-time sta-
results are proposed for scalar systems by using constant and bilization using a SMC strategy with time-independent
variable exponent coefficients. Then, they are applied to global
robust fixed-time stabilization of a class of uncertain nonlinear controllers has been proposed in [15], [16], [17] also by
second-order systems by using sliding mode control. All the solving a singularity problem. The singularity problem
results are illustrated in simulation. comes from the fact that the simplest finite-time and fixed-
Index Terms—Fixed-time stability, sliding mode control, time sliding variables are non differentiable. It results in
robust control. more complex feedback controls to implement. Finally,
the notion of predefined/prescribed-time SMC has been
I. Introduction introduced in [18], [19], [20] by using time-dependent
Sliding mode control (SMC) has been developed by controllers.
Utkin in [1] and then by many authors, see [2] and the
references therein for more details. The aim of SMC is to In this article, new global robust fixed-time stability
enforce a dynamical system to reach a manifold called results are provided for scalar systems by using constant
“sliding surface” defined by a function called “sliding and state-dependent variable exponent coefficients. State-
variable” with an appropriate controller ensuring that a dependent variable exponent coefficients have already
constraint on the sliding variable is satisfied. After the been used in the context of homogeneous self-triggered
constraint is checked, the system trajectories “slide” on the control in [21]. They have also been used for defining the
sliding surface towards the desired equilibrium. The main controllers in [22] for finite-time SMC. But to the best of
advantage of SMC lies in the simplicity of its feedback the authors’ knowledge, it has never been used for fixed-
control strategy after choosing the sliding variable, its time stability. By employing the SMC strategy, global
robustness when using discontinuous controllers and the robust asymptotic stabilization of the global x−system of
finite-time convergence of the closed-loop system trajec- the state variable x with robust fixed-time stabilization
tories to reach the sliding surface. Moreover, it has been of the s−system of the sliding variable s is obtained by
refined over time, for instance with the integral SMC [3]. using constant exponent coefficients in the sliding variable
Finite-time stability has been developed by Bhat and and the controllers for a class of uncertain nonlinear
Bernstein in [4] and then applied for finite-time stabiliza- second-order systems. Moreover, global robust fixed-time
tion, for instance in [5], [6]. It ensures that dynamical stabilization of the global x−system is obtained by using
systems reach their equilibrium in a finite-time called state-dependent variable exponent coefficient in the sliding
settling-time depending on the initial conditions. Finite- variable and the controllers. The new sliding mode con-
time stabilization using a SMC strategy has been devel- trollers are time-independent, non singular, robust with
oped in [7], [8] where a singularity problem is solved. respect to bounded disturbances and easy to implement.
The fixed-time stability was introduced by Polyakov in So using a variable exponent coefficient allows to obtain
[9] and then developed by many authors, for instance robust fixed-time SMC of the global x−system contrary
in [10], [11], [12], [13], [14]. In addition to finite-time to the constant exponent coefficient strategy. Actually,
stability, fixed-time stability ensures that the settling- it is not easy to obtain fixed-time stabilization of the
time does not depend on the initial conditions. Fixed- global x−system when dealing with SMC because some
time stabilization provides a predefined convergence time singularities appear when using the simplest fixed-time
sliding variable, see [16] and [17]. With the use of a
E. Moulay is with XLIM (UMR CNRS 7252), Université de variable exponent coefficient in the sliding variable and
Poitiers, 11 bd Marie et Pierre Curie, 86073 Poitiers Cedex 9, France
(e-mail: emmanuel.moulay@univ-poitiers.fr). the controllers, we obtain a simple solution because the
V. Léchappé is with Ampère (UMR CNRS 5005), INSA Lyon, controllers have no singularity.
20 Avenue Albert Einstein, 69100 Villeurbanne, France (e-mail:
vincent.lechappe@insa-lyon.fr).
E. Bernuau is with GENIAL (UMR INRA 1145), AgroParisTech, The paper is organized as follows. After some prelim-
1 Avenue des Olympiades, 91744 Massy Cedex, France (e-mail: inaries in Section II, the main results on robust fixed-
emmanuel.bernuau@agroparistech.fr). time stability are given in Section III. The application to
F. Plestan is with LS2N (UMR CNRS 6004), Ecole Centrale de
Nantes, 1 rue de la Noë, 44321 Nantes Cedex 3, France (e-mail: SMC is developed in Section IV. Finally, a conclusion is
franck.plestan@ec-nantes.fr). addressed in Section V.
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 2
II. Preliminaries with α+12 > 1. By using Lemma 1, the result follows.
In the following, denote R+ the set of positive real Remark 1: On the one hand, the function x 7→
numbers and e the constant such that ln(e) = 1. Recall |x|γ sgn(x) with 0 < γ < 1 is not necessary to obtain the
some results on finite-time stability and fixed-time stabil- fixed-time stability while still used for instance in [14],
ity. Consider the following ordinary differential equation [9]. On the other hand, the sign function x 7→ sgn(x)
coupled with the function x 7→ |x|α sgn(x) where α > 1
ẋ(t) = f (x(t)), x(t) ∈ Rn (1) allows the fixed-time stability and is known to reject the
x(0) = x0 disturbances. This is the reason why robust fixed-time
stability is obtained. Moreover, if only the first term is
with f a continuous function such that f (0) = 0.
used, i.e. if k2 = k3 = k4 = 0, one only obtains robust
Definition 1: [6], [4] System (1) is globally finite-time
finite-time stability.
stable if it is Lyapunov stable and for all x0 ∈ Rn there
In the following, we compare in simulation the ro-
exists T (x0 ) > 0 dependent on the initial conditions such
bustness of the state x(t) of system (4) and the term
that limt→T (x0 ) ∥x(t)∥ = 0, i.e. ∥x(t)∥ ≡ 0 for all t ≥
u(t) = −k1 sgn(x) − k2 |x|α sgn(x) − k3 |x|γ sgn(x) − k4 x in
T (x0 ). The function T is called the settling-time.
several cases with the disturbance d(t) = sin(10t) leading
Definition 2: [9] System (1) is globally fixed-time stable
to δ = 1, the initial condition x(0) = 3, α = 1.5 and
if:
γ = 0.5. Here are the three cases:
(1) it is globally finite-time stable;
• Case 1. k1 = 2, k2 = 2, k3 = 0, k4 = 0;
(2) the settling-time function T is upper bounded by a
• Case 2. k1 = 2, k2 = 2, k3 = 2, k4 = 0;
constant T > 0, i.e. T (x0 ) ≤ T and does not depend
• Case 3. k1 = 2, k2 = 2, k3 = 2, k4 = 2;
on the initial conditions.
that leads to T (x0 ) ≤ 2s. Figure 1 shows that system (4)
Lemma 1: [9], [14] If there exists a continuously dif-
is robust with respect to the disturbance for all cases and
ferentiable positive definite radially unbounded function
Figure 2 shows the induced chattering for the steady state
V : Rn → R+ such that
x(t). The settling-time of system (4) in case 3 is strictly
V̇ (x) ≤ −aV (x)γ − bV (x)α (2) lower than the settling-time in cases 1 and 2 because the
time-derivative V̇ is rendered more negative. This explains
where x ∈ R , a > 0, b > 0 and 0 < γ < 1 < α, then
n
the interest of introducing additional terms in system (4)
system (1) is globally fixed-time stable and the settling-
while keeping the robust fixed-time stability.
time satisfies
1 1
T (x0 ) ≤ + . (3)
a(1 − γ) b(α − 1) 3
x(0) = x0 -20
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 3
2
-2
1.5
1
-3
2 2.1 2.2 2.3 2.4 2.5 0.5
time (s)
0
-0.5
Fig. 2. Zoom on the steady state x(t) versus time (sec) 0 0.5 1 1.5 2
time (s)
λx2 Fig. 3. Top. State x(t) versus time (sec) Bottom. u(t) versus time
Proof. ) that the function φ : x 7→ |x| 1+µx =
First note 2
( (sec)
λx2
exp 1+µx2 ln(|x|) is continuous at x = 0 with φ(0) = 1.
Therefore the right-hand side of (8) is locally bounded.
Consider the following quadratic Lyapunov function
10 -3
V (x) = x2 . (10) 3
It leads to 2
λx2
+1
V̇ (x) = −2k|x| 1+µx2 + 2dx (11)
1
2
Consider the case V (x) ≥ 1. We have 1+µx λx
2 + 1 ≥
λ
1+µ + 1 > 2. As |x| ≥ 1 and θ = λ
1+µ > 1 it leads to 0
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 4
0
0 0.5 1 1.5 2 2.5 3
IV. Application to sliding mode control time (s)
6
ẋ1 = x2
(19) 4
ẋ2 = f (x) + g(x)u + d 2
different cases presented in Subsection III-A. Since all the u(x) = − g(x) f (x) + k|s| 1+µ2 s2 sgn(s)
parameters are the same as in Subsection III-A, one still ( ) ]
λ1 x2
has T (s0 ) ≤ 2s. The time evolution of the sliding variable βλ1 |x1 |x2 2 ln |x1 | 1
+ 1 |x1 | 1+µ1 x1
2
+ (27)
s(t) and the norms of the state variable ∥x(t)∥ associated 1 + µ1 x21 1 + µ1 x21
to the closed-loop system (19)–(20)–(21) are shown on
λ2
Figure 5. with θ2 = λ2
> 1, k > δe 2e .
1+µ2
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 5
ṡ = f (x) + g(x)u(x) 2
( ) λ1 x2
βλ1 |x1 |x2 2 ln |x1 | 1
+ 2 2 + 1 |x1 | 1+µ1 x21 + d
0
1 + µ 1 x1 1 + µ 1 x1
λ 2 s2 -2
0 0.5 1 1.5 2 2.5 3
= −k|s| 1+µ2 s2 sgn(s) + d (29) time (s)
By using Theorem 2, one deduces that system (29) starting Fig. 7. Sliding variable s(t) versus time for different initial conditions
at s(0) = s0 reaches the sliding surface {s = 0} in a fixed- x2 (0)
time satisfying T (s0 ) ≤ (k−δ)(θ
1
2 −1)
+ −λ12 . From (26),
ke 2e −δ
one has Remark 4: Fist of all, the controller (27) is not singular
λ1 x2
1
ẋ1 = −β|x1 | 1+µ1 x21 sgn(x1 ). (30) due to the fact that limx1 →0 |x1 | ln(|x1 |) = 0. Then, if a
more simple sliding variable
By using one more time Theorem 2, it is deduced that
x1 (t) starting at x1 (0) = x10 reaches the origin in a fixed- s(x) = x2 + β|x1 |λ|x1 | sgn(x1 ) (31)
time satisfying T (x10 ) ≤ β(θ11−1) + −λ 1
. Finally, the with β > 0, λ > 0 is chosen and if the controller reads as
[ ( )
1
βe 2e
closed-loop system (19)–(26)–(27) reaches the origin in a u(x) = − g −1 (x) f (x) + βλ x2 (ln |x1 | + 1) |x1 ||x1 |
fixed-time T (x0 ) = T (s0 ) + T (x10 ) that is bounded by ]
(28). + k|s||s| sgn(s) (32)
Consider the closed-loop system (19)–(26)–(27) with
with k > δ then
f = 0, g = 1, β = 0.2, λ1 = 2, λ2 = 4, µ1 = 0.1 µ2 = 1,
k = 10, x(0) = [0.1, x2 (0)]T and d(t) = sin(10t). In the ṡ = f (x) + g(x)u(x) (33)
( )
case, one gets T (x0 ) ≤ 13.7s. Figure 6 displays the time +βλ x2 (ln |x1 | + 1) |x1 ||x1 | + d (34)
evolution of the sliding variable s(t) and the state variable
|s|
x(t) = (x1 (t), x2 (t)). = −k|s| sgn(s) + d (35)
The fixed-time stabilization is obtained but the
controller u(x) is singular due to the fact that
4
limx1 →0 |x1 ||x1 | ln |x1 | = −∞.
3
In order to reduce the chattering induced by the use of
2
controller (27), consider the following controller
1 [
u(x) = − g −1 (x) f (x) + k1 sgn(s) + k2 |s|α sgn(s)
0
-1
0 0.5 1 1.5 2 2.5 3
time (s)
( ) ]
λ1 x2
βλ1 |x1 |x2 2 ln |x1 | 1
+ 1 |x1 |
3 1+µ1 x2
+ 1 (36)
2 1 + µ1 x21 1 + µ1 x21
1 with k1 > δ, k2 > 0, α > 1.
0 Proposition 3: The closed-loop system (19)–(26)–(36) is
-1
globally fixed-time stable and the settling-time satisfies
0 0.5 1 1.5 2 2.5 3
time (s) 1 1
T (x0 ) ≤ + (37)
k1 − δ k2 (α − 1)
Fig. 6. Top. Sliding variable s(t) versus time (sec) Bottom. State 1 1
variable x(t) = (x1 (t), x2 (t)) versus time (sec) + + . (38)
β(θ1 − 1) βe −λ 2e
1
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 6
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAC.2021.3069667, IEEE
Transactions on Automatic Control
SUBMITTED TO IEEE TRANSACTIONS ON AUTOMATIC CONTROL - MARCH 23, 2021 7
0018-9286 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: CINVESTAV. Downloaded on May 12,2021 at 22:19:38 UTC from IEEE Xplore. Restrictions apply.