You are on page 1of 11

Received: 23 December 2021 Revised: 6 July 2022 Accepted: 28 September 2022 IET Control Theory & Applications

DOI: 10.1049/cth2.12367

ORIGINAL RESEARCH

Local robust stability on compact set for nonlinear systems with


continuous time controller against to aperiodic sampling and
disturbance

Kazuki Umemoto1 Takahiro Endo2 Fumitoshi Matsuno2

1
Department of Mechanical Engineering, Nagaoka Abstract
University of Technology, Niigata, Japan
In this study, the stability of a sampled-value control system with time- and state-dependent
2
Department of Mechanical Engineering and disturbance and aperiodic sampling is investigated. To expand the application class of
Science, Graduate School of Engineering, Kyoto
University, Kyoto, Japan
the sampled-value controller, the local stability on a compact set is considered under the
assumption of local Lipschitz continuity on the compact set. Because the assumption is
Correspondence weaker than the global Lipschitz continuity, the proposed stability analysis is applicable to
Kazuki Umemoto, Department of Mechanical a wider class of controlled systems. Global stability under the assumption of global Lip-
Engineering, Nagaoka University of Technology,
schitz continuity is also considered. The effectiveness of the derived stability condition is
Niigata 940-2188, Japan.
Email: umemoto@mech.nagaokaut.ac.jp verified by conducting numerical simulations with some control parameters variations such
as the disturbance magnitude and sample interval.

1 INTRODUCTION cases, it is not possible to accurately know the sample interval


in advance. In recent years, attempts have been made to solve
Because a feedback control system using computers cannot real- these problems by applying control with arbitrary sample inter-
ize continuous feedback, this study investigated a sampled-value vals that are smaller than the maximum sampling interval (MSI).
feedback control system. There are two well-known approaches More information can be found in a previously published review
in the field of sampled-value control: the state space approach paper [15, 16] and its provided list of references.
[1–3] and the function space approach by the lifting operation On aperiodic sampled-value control, Fujioka derived a linear
proposed by Yamamoto [4]. matrix inequality (LMI) stability condition for a linear sampled-
Globally Lipschitz nonlinear systems have often been consid- value control system [17, 18]. Oishi and Fujioka derived an
ered in sampled-value control [5–8]. The assumption of global LMI stability condition for uncertain linear systems [19]. For
Lipschitz continuity restricts their applications compared with linear control systems with dead time, Fridman et al. derived
local Lipschitz continuity. For example, serial-link robot manip- a stability condition using a new Lyapunov function [20]. Su
ulators are not globally Lipschitz but locally Lipschitz. On et al. proposed a stability analysis method based on convex
sampled-value control for systems with disturbance, there are approximation for linear sampled-value control systems [21].
two major control approaches; sliding mode control [9–11] and Although these methods are intended for linear systems, non-
disturbance observer-based control [12, 13]. Additionally, con- linear systems have also been investigated to expand the class of
trol error evaluation method based on input-to-state stability controlled systems. Stability analysis methods based on approx-
was proposed [14]. imation have been proposed for nonlinear systems [22, 23].
In many sampled-data control approaches, the sample inter- Because approximations cause analysis errors, approaches with-
val must be constant or known, even if it is aperiodic. Therefore, out approximations have also been investigated. Bernuau et al.
these approaches cannot be applied to network systems. Addi- derived the global and local stability conditions for homoge-
tionally, in embedded systems, it is assumed that periodic neous systems and discussed the stabilization region [24]. Du
samples cannot be maintained to realize many functions, such et al. reported the asymptotic stability conditions for uncer-
as communication with other devices, including sensors. In such tain homogeneous systems [25]. However, they only considered

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2022 The Authors. IET Control Theory & Applications published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.

IET Control Theory Appl. 2023;17:133–143. wileyonlinelibrary.com/iet-cth 133


134 UMEMOTO ET AL.

homogeneous sampled-value systems. Mazenc et al. derived the stability analysis proposed in this paper can be applied to many
MSI by applying the Lyapunov function for linear systems [20] nonlinear systems. In this paper, although controller design for
to nonlinear systems [26]. Li et al. reported that the stability con- disturbance rejection is not discussed, high-gain feedback and
dition of a nonlinear sampled-valued control system with input H∞ control can be applied to the stability analysis method.
constraints can be reduced to the LMI condition [27]. Omran The proposed stability analysis was applied to a two-link
et al. proposed a new stability condition and reported that the manipulator, and its effectiveness was verified by numerical sim-
stability condition can be reduced to the LMI condition in lin- ulations.
ear time-varying systems with polytopic uncertainties [28]. Mao
et al. [29] proposed adaptive control for systems with uncer-
tainty. In [29], the uncertainty is assumed to be bounded by a 2 PROBLEM SETTING AND MAIN
norm of system states. In [28, 29], uncertainties have been con- RESULTS
sidered, but disturbances have not. In contrast with [28, 29],
for periodic sampling system, Chu et al. [30] analyzed the sta- The following nonlinear system with uncertainty and distur-
bility of a single-input/output sampled-value output feedback bance is considered:
system with global Lipschitz uncertainty and state-dependent
disturbance. Li et al. [31] extended the previous results [30] and ẋ = f (x) + Δ f (x) + (g(x) + Δg(x))u(x) + d (t , x), (1)
performed stability analysis by considering the input coefficient
uncertainty and Lipschitz uncertainty for the drift term. Nonlin- where x ∈ ℝn , u ∈ ℝm and d (t , x) ∈ ℝn denote the state,
ear systems satisfying the global Lipschitz continuity are limited. control input, and time-varying disturbance, respectively; Δ f
Additionally, state- and time-dependent disturbances have not and Δg are uncertainties; f (x), Δ f (x) ∈ ℝn and g(x), Δg(x) ∈
been considered. The definition of the disturbance in [30, 31] ℝn×m hold; n and m are natural numbers.
does not even include constant disturbances because the dis- This study considers the case wherein the control input u is
turbance in [30, 31] is assumed to be bounded by a norm of discretely implemented by samples {ti }, (ti+1 − ti ) ∈ (0, h) and
system states. zero-order holds; h is positive and i is a non-negative integer.
Time-dependent disturbances can cause steady-state errors Specifically, the controlled system (1) can be described using
in feedback control systems and, generally, static feedback con- piece-wise constant input, as follows:
trollers cannot guarantee asymptotic stability. Owing to their
importance in practical situations, the time-dependent distur- ẋ = f (x) + Δ f (x) + (g(x) + Δg(x))u(x(ti )) + d (t , x),
bance has been extensively investigated for continuous-time and (2)
periodic sampled-value feedback systems [9–14], but the prob- for t ∈ [ti , ti+1 ),
lem remains unresolved for aperiodic sampled-data feedback
systems. In [30, 31], the disturbance was assumed to be bounded where ti → ∞ when i → ∞, and the following assumptions
by a norm of system states, time-dependent disturbance can- hold:
not be considered by the approach in [30, 31]. Additionally,
the assumption of the global Lipschitz continuity restricts the Assumption 1. There exists a positive definite function V (x)
class of controlled systems. Specifically, a system that cannot such that the following conditions hold on a compact set Ω:
be globally stabilized, because it is not globally Lipschitz con-
tinuous, may be locally stabilized on a compact set. Therefore, 1. There exist positive constants 𝛼1 and 𝛼2 such that 𝛼1 ‖x‖2 ≤
stability analysis for locally Lipschitz systems with time- and V (x) ≤ 𝛼2 ‖x‖2 .
𝛿V
state-dependent disturbance is a significant issue to be solved. 2. There exists a positive constant c1 such that ( f (x) +
𝛿x
This study considers a novel problem wherein a control
g(x)u(x)) ≤ −c1 ‖x‖2 .
system has time- and state-dependent disturbance and local Lip- 𝛿V
schitz uncertainty. The local Lipschitz continuity on a compact 3. There exists a positive constant c2 such that ‖ ‖ ≤ c2 ‖x‖.
𝛿x
set is a weaker condition compared with global Lipschitz con-
tinuity, and the applicable class can be expanded. However, it Assumption 2. f (0) = 0, Δ f (0) = 0, and u(0) = 0 are sat-
is necessary to carefully consider whether the state trajectory isfied. f , Δ f and u are locally Lipschitz continuous on a
is maintained in a considered compact set. The contribution compact set Ω. Specifically, there exists a positive constant
of this study is summarized by the two following points: (1) c f such that ‖ f (x1 ) − f (x2 )‖ ≤ c f ‖x1 − x2 ‖ ∀x1 , x2 ∈ Ω.
Consideration of time- and state-dependent disturbance. Previ- There exists a positive constant cΔ f such that ‖Δ f (x1 ) −
ous studies have derived a state-independent MSI condition [26, Δ f (x2 )‖ ≤ cΔ f ‖x1 − x2 ‖ ∀x1 , x2 ∈ Ω. There exists a positive
28–31], but this condition is not independent in systems with constant cu such that ‖u(x1 ) − u(x2 )‖ ≤ cu ‖x1 − x2 ‖ ∀x1 , x2 ∈
time- and state-dependent disturbance. In this study, a state- Ω. Additionally, the positive constants 𝜌d , cg , and cΔg exist such
dependent MSI condition was derived. (2) Expansion of a class that ‖d (t , x)‖ < 𝜌d , ‖g(x)‖ ≤ cg , and ‖Δg(x)‖ ≤ cΔg , ∀x ∈
of controlled systems by assuming local Lipschitz continuity on Ω, respectively.
compact sets. Previously reported analysis methods [26, 28–31]
cannot be applied to complex nonlinear systems, including Remark 2.1. If the control input u(x) can be continuously imple-
robot manipulators. By assuming local Lipschitz continuity, the mented for (1), from Assumption 1, the equilibrium point of
UMEMOTO ET AL. 135

the control system x = 0 is locally asymptotically stable for ( 1∕2 )


v1
‖Δ f ‖ = 0, ‖Δg‖ = 0. When the controlled system is feed- q3 (t ) ∶= c2 (q1 (t ) + 1) + 𝜌d q2 (t )
1∕2
back linearizable, the conditions for V in Assumption 1 can 𝛼1
be satisfied for a quadratic Lyapunov function. If the sys- ( 1∕2 )
tem is not feedback linearizable, it is difficult to find the v1
constants in Assumption 1. However, there are many feed- × (cg cu q1 (t ) + cΔg cu ) + 𝜌d (cg cu q2 (t ) + 1) .
1∕2
𝛼1
back linearizable systems because Assumption 1 relates to the
continuous-time control system without the disturbance and (6)
the uncertainty. The proposed analysis has wide application
targets. Note that h(v1 ) > 0 exists if (3) holds.
An assumption similar to Assumption 2 has been consid-
ered by previous studies [26, 28–31]. Notably, the disturbance The stability conditions for the system with the time-
considered in [30, 31] is assumed to be bounded by the and state-dependent disturbance in Theorem 2.1 depend
state norms. Hence, the class of the disturbance in [30, 31] on the state. In contrast, when there is no disturbance
is similar to uncertainty Δ f (x) in this article. In this arti- (d (t , x) = 0), Corollary 2.1 is obtained as a state-independent
cle, we consider not only Δ f but also bounded disturbance result.
d (t , x). Additionally, in previous studies, global properties were
assumed. In this study, however, local properties are assumed. Corollary 2.1. Let x = 0 be an equilibrium point of the sampled-
Owing to the assumption of local properties, the restrictions value control system (2). Assumptions 1, 2, and 𝜌d = 0 hold for Ω ∶=
on the class of the functions f , g of the controlled system are {x | V (x) ≤ v0 }. The initial state x(0) is assumed to be inside Ω. If
relaxed.
𝛼1 (c1 − c2 cΔ f ) − 𝛼2 c2 cΔg cu > 0, (7)
This study analyzes the local robustness of a sample-and-hold
implementation of a nonlinear continuous-time controller by holds, and the sampling time is smaller than h1 , which is defined as
investigating the condition on the upper limit of the sampling follows:
time h for the initial value x(0) under assumptions 1 and 2. The
following Theorem 2.1 is obtained as the main result of this h1 ∶= sup t such that V1 (t ) < 1, (8)
t >0
study:
then, ‖x(t )‖ → 0 for t → ∞, where
Theorem 2.1. Let x = 0 be an equilibrium point of the sampled-value
control system (2). Assumptions 1 and 2 hold for Ω ∶= {x | V (x) ≤ c1 −c2 cΔ f ( t c1 −c2 cΔ f )
− t s
v0 } and a positive constant v0 . The initial state x(0) is assumed to be V1 (t ) ∶= e 𝛼2 q4 (s)e 𝛼2 ds + 1
inside Ω. If there exists a positive constant v1 such that the following ∫0
relationship holds:
q4 (t ) ∶= c2 (q1 (t ) + 1)(cg cu q1 (t ) + cΔg cu ). (9)
𝛼1 𝛼2
v0 ≥ v 1 > 𝜌 > 0, (3) h1 > 0 exists if (7) holds.
𝛼1 (c1 − c2 cΔ f ) − 𝛼2 c2 cΔg cu d

and the sampling time is smaller than h(v1 ), which is defined as follows: Additionally, the local stability criteria of Theorem 2.1 can be
easily extended to the following global stability criteria:
h(v1 ) ∶= sup t such that V0 (v1 , t ) < 1, (4)
t >0 Corollary 2.2. Let x = 0 be an equilibrium point of the sampled-
value control system (2). Assumptions 1 and 2 hold for Ω ∶= ℝn . If
then, the state of the controlled system is ultimately bounded as follows: there exists a positive constant v1 such that the following relationship
holds:
‖x(t )‖2 < v1 ∕𝛼1 , t → ∞, (5)
𝛼1 𝛼2
v1 > 𝜌 > 0, (10)
𝛼1 (c1 − c2 cΔ f ) − 𝛼2 c2 cΔg cu d
where
c1 −c2 cΔ f ( t c −c2 cΔ f ) and the sampling time is smaller than h(v1 ) defined by (4), then the state
− t q3 (s) 1 s
V0 (v1 , t ) ∶= e 𝛼2 e 𝛼2 ds + 1 , of the controlled system is ultimately bounded as (5), where V0 is defined
∫0 v1
by (6).
c f + cΔ f + (cg + cΔg )cu
q1 (t ) ∶= (e(c f +cΔ f )t − 1), Remark 2.2. The stability condition is significantly changed
c f + cΔ f
depending on whether or not a disturbance exists. The
1 condition in Corollary 2.1 is state-independent except for
q2 (t ) ∶= (e(c f +cΔ f )t − 1),
c f + cΔ f Assumption 2. However, the conditions (3) and (4) in
136 UMEMOTO ET AL.

From (2), (12), Assumptions 1, and 2,

ż ≤ ‖ f (x) + Δ f (x) + (g(x) + Δg(x))u(x(ti )) + d (t , x)‖

≤ (c f + cΔ f )(z + ‖x(ti )‖) + (cg + cΔg )cu ‖x(ti )‖ + 𝜌d

c f + cΔ f + (cg + cΔg )cu


≤ (c f + cΔ f )z + V (x(ti ))1∕2 + 𝜌d ,
1∕2
FIGURE 1 Function estimates in 1st and 2nd steps: (a) time response of 𝛼1
state; (b) time response of V
(13)

Theorem 2.1 depend on v1 , which is related to the state. can be obtained. Based on the comparative theorem, the solu-
Therefore, if there exists a time- and state-dependent distur- tion of z holds z ≤ z using the solution of the following
bance, the state-independent stability conditions reported in differential equation:
the literature [26, 28–31] are not valid.
c f + cΔ f + (cg + cΔg )cu
The local stability analysis can be easily extended to the global ż = (c f + cΔ f )z + V (x(ti ))1∕2 + 𝜌d .
one. However, extension to local analysis of global analysis can- 1∕2
𝛼1
not be directly obtained, because inclusion of state trajectory (14)
must be additionally considered.
The solution of (14) at interval [ti , ti+1 ) for z(ti ) = 0 can be
Proof of Theorem 2.1. Theorem 2.1 is proven in three steps. In expressed as follows:
the 1st step, the time evolution of the state is estimated from
the initial value of an interval. In the 2nd step, the time evo- V (x(ti ))1∕2
lution of V (t ) is estimated using the result of the 1st step. z(t ) = p1 (t ) + 𝜌d p2 (t ), (15)
1∕2
Figure 1 shows the estimation of the time evolution of x and 𝛼1
V in the 1st and 2nd steps. Notably, in the 1st and 2nd steps, it
is assumed that the state exists in the compact set Ω defined in c f + cΔ f + (cg + cΔg )cu
p1 (t ) ∶= (e(c f +cΔ f )(t −ti ) − 1) (16)
Assumption 2. c f + cΔ f
In the 3rd step, the condition for V (ti+1 ) < V (ti ) is derived
using the result of the 1st and 2nd steps. Because this condition 1
p2 (t ) ∶= (e(c f +cΔ f )(t −ti ) − 1). (17)
depends on the initial value of V (t ) in the interval, we discuss c f + cΔ f
the change of V (t ) with respect to the change of the initial
value in the interval. Finally, we estimate the steady state using
the estimation of V (t ), and discuss whether the state maintains See Appendix A for the detailed derivation of (15).
the compact set Ω. Figure 2 shows the property of V (t ) with (2nd step, time evolution of positive definite func-
respect to the change of the initial value in the interval discussed tion V ) Next, we consider the time evolution in [ti , ti+1 )
in the 3rd step. of the positive definite function V (x) satisfying Assump-
(1st step, evolution of state errors from sample start time). tion 1. The time derivative of V (t ) satisfies the relation-
We discuss the estimate of the time evolution of the errors ship (18) using Assumption 1, 2, (2), and ‖x(t )‖ ≤ z +
between x(t ) and x(ti ) on [ti , ti+1 ). z(t ) is defined as an absolute ‖x(ti )‖,
̇ ) as follows:
integral of x(t
𝛿V ( )
t V̇ = f (x) + Δ f (x) + (g(x) + Δg(x))u(x(ti )) + d (t , x)
𝛿x
z ∶= ‖x(s)‖ds.
̇ (11)
∫t 𝛿V
i = ( f (x(t )) + g(x(t ))u(x(t )))
𝛿x
From the definition, z(t ) holds the following relation: 𝛿V ( ( ) )
+ Δ f (x) + g(x(t )) u(x(ti )) − u(x(t )) + Δg(x)u(x(ti )) + d (t , x)
𝛿x
t
( )
x(t ) − x(ti ) = ̇
x(s)ds ≤ − c1 ‖x(t )‖2 +c2 ‖x(t )‖ cΔ f ‖x(t )‖ + cg cu ‖x(ti ) − x(t )‖ + cΔg cu ‖x(ti )‖ + 𝜌d
∫t
i
c1 − c2 cΔ f ( )
‖ t ‖ t ≤− V + c2 (z(t ) + ‖x(ti )‖) cg cu z(t ) + cΔg cu ‖x(ti )‖ + 𝜌d
‖ ‖ 𝛼2
‖x(t ) − x(ti )‖ = ‖ ̇
x(s)ds ‖≤ ‖x(s)‖ds
̇ =z
‖∫t ‖ ∫t ( )( )
‖ i ‖ i c1 − c2 cΔ f V (x(ti ))1∕2 V (x(ti ))1∕2
≤− V +c2 z(t )+ cg cu z(t )+cΔg cu +𝜌d
𝛼2 𝛼1
1∕2 1∕2
𝛼1
‖x(t )‖ − ‖x(ti )‖ ≤ ‖x(t ) − x(ti )‖ ≤ z
c1 − c2 cΔ f
‖x(t )‖ ≤ z + ‖x(ti )‖. (12) ≤− V + p3 (t ), (18)
𝛼2
UMEMOTO ET AL. 137

FIGURE 2 Estimation of functions in 3rd step: (a) V ; (b) V n

where For V (x(ti )) = v1 , the upper limit of the sample interval


( ) (MSI) h for satisfying V n (V (x(ti )), ti ) > V n (V (x(ti )), ti+1 )
V (x(ti ))1∕2 can be derived using (19) and (23), as follows:
p3 (t ) ∶= c2 (p1 (t ) + 1) + 𝜌d p2 (t )
1∕2
𝛼1
( ) tsup ∶= sup t such that V n (v1 , t ) < 1,
t >ti
V (x(ti ))1∕2
× (cg cu p1 (t ) + cΔg cu ) + 𝜌d (cg cu p2 (t ) + 1) .
𝛼1
1∕2 h(v1 ) ∶= tsup − ti . (24)
(19) If (20) holds, then, h(v1 ) > 0 exists because it is guaranteed that
Because p1 (ti ) = p2 (ti ) = 0, if V n will decrease at the sample start time ti , and the function
V n (V (x(ti )), t ) belongs to class C 1 .
c 1 − c 2 cΔ f V (x(ti ))
− V (x(ti )) + c2 cΔg cu + 𝜌d Let us consider the variation of MSI, V , and V n for the
𝛼2 𝛼1 variation in V (x(ti )). Notably, p3 (t ) is a function that mono-
, (20) tonically increases with the increase of V (x(ti )) in any t ∈
𝛼1 (c1 − c2 cΔ f ) − 𝛼2 c2 cΔg cu
= 𝜌d − V (x(ti )) < 0 [ti , ti+1 ). Hence, V (V (x(ti )), t ) is a strict monotonically
𝛼1 𝛼2
increasing function for any t ∈ [ti , ti+1 ) increase in V (x(ti ))
is satisfied, then, V will decrease at t = ti from (18) and (19). for any t ∈ [ti , ti+1 ). In brief, the following relationship
Based on the comparative theorem, V satisfies V ≤ V using holds:
the solution V of the following differential equation: | |
V (V (x(ti )), t )| < V (V (x(ti )), t )| . (25)
|V (x(ti ))<v1 |V (x(ti ))=v1
c1 − c2 cΔ f
V̇ = − V + p3 (t ), (21)
𝛼2 Although p3 (t ) is a monotonic increase with respect to V (x(ti )),
p (t )
in contrast, 3 is a broad monotonic decrease with respect
for an initial value V (ti ) = V (x(ti )), where c1 − c2 cΔ f > 0 from V (x(ti ))
to V (x(ti )). In brief, the following relationship holds:
(3). The solution of (21) in interval [ti , ti+1 ) is expressed as
follows: | |
V n (V (x(ti )), t )| ≤ V n (V (x(ti )), t )| .
c1 −c2 cΔ f
( t c1 −c2 cΔ f
) |V (x(ti ))>v1 |V (x(ti ))=v1
− (t −ti ) (s−ti )
V (V (x(ti )), t ) = e 𝛼2 p3 (s)e 𝛼2 ds + V (x(ti )) . (26)
∫t
i

(22) Because V n increases with the decrease of V (x(ti )), for


V (x(ti )) < v1 , it is not guaranteed that the decrement of
Additionally, we consider the normalization V n ∶= V V n (V (x(ti )), ti+1 ) from V n (ti ) at the start of the next sam-
(V (x(ti )), t )∕V (x(ti )) of V using V (x(ti )). From (22), ple. Specifically, V n (V (x(ti )), ti+1 ) < V n (V (x(ti )), ti ) may
we can obtain the following relationship:
not hold. Moreover, the decrease of V n at t = ti is also not
( ) guaranteed. This means that the independence of the conven-
c1 −c2 cΔ f
− (t −ti )
t
p3 (s) c1 −c𝛼2 cΔ f (s−ti )
V n (V (x(ti )), t ) = e 𝛼2 e 2 ds + 1 . tional MSI condition on the state is not maintained, owing to
∫t V (x(ti ))
i the presence of disturbance.
(23) Although we considered the relations (25) and (26) for t ∈
[ti , ti+1 ), the relations are valid for t ∈ [ti , ti+1 ] including t =
V n is used to show boundedness of the state variables in steady ti+1 by the continuity of the system state. Below, we consider
state. The relation of V and V n with respect to the change of the estimate of V (x(ti+1 )) in t ∈ [ti , ti+1 ]. The estimate of
the initial value of a sample interval are shown in Figure 2. V (x(ti+1 )) is considered in three cases when the system is con-
(3rd step, derivation of MSI, variation of MSI, V , and V n trolled with a sample interval smaller than h(v1 ) obtained from
for variation of V (x(ti )) region containing state trajectory) (24) for V (x(ti )) = v1 .
138 UMEMOTO ET AL.

Case 1. For V (x(ti )) = v1 ≤ v0 , using (24), V n = Therefore, if V (x(ti )) ≥ v1 holds, then, V decreases with each
V (V (x(ti )), t )∕V (x(ti )), and V (x(ti+1 )) ≤ V (V (x(ti )), sample, and after satisfying V (x(ti+ j )) < v1 , the relationship
ti+1 ), the following relationships are obtained: V (x(ti+ j +k )) < v1 is maintained for any positive integer k from
Case 3. According to Assumption 1, we can obtain ‖x(t )‖2 ≤
V n (V (x(ti )), ti+1 ) <1 V (x(t ))∕𝛼1 < v1 ∕𝛼1 for t → ∞.
The conditions of the theorem are obtained by setting ti = 0
V n (V (x(ti )), ti+1 )V (x(ti )) <V (x(ti )) and V (x(ti )) = v1 in (20) and (24). Consequently, Theorem 2.1
is proven. □
V (x(ti+1 )) ≤ V (V (x(ti )), ti+1 ) <V (x(ti ))
Corollary 2.1. Because Corollary 2.1 can be proven by
V (x(ti+1 )) <V (x(ti )). (27) the procedure included in the proof of Theorem 2.1, the
main differences are discussed. For 𝜌d = 0, (20) in the proof
Because V (t ) ≤ V (x(ti )) holds for any t ∈ [ti , ti+1 ], and the of Theorem 2.1 is equivalent to (7) for any V (x(ti )) >
state maintains x(t ) ∈ Ω, the Lipschitz continuity and norm 0. Additionally, from (19), p3 (s)∕V (x(ti )) in (23) becomes
boundedness in Assumption 2 can be used at all times within independent of V (x(ti )). Therefore, (8) can be derived
the interval. Thus, the results of the 1st and 2nd steps are from (24).
valid.
Corollary 2.2. By defining the compact set Ω as ℝn , Corol-
Case 2. For v0 ≥ V (x(ti )) > v1 , the following relationship holds lary 2.1 can be directly proven based on the similar procedure
from (24) and (26): included in the proof of Theorem 2.1.
| |
V n (V (x(ti )), ti+1 )| ≤ V n (V (x(ti )), ti+1 )| < 1. Remark 2.3. In contrast to periodic sampled-value control,
|V (x(ti ))>v1 |V (x(ti ))=v1
(28) we need to consider stability for any sample intervals less
than MSI in aperiodic sampled-value control. Additionally,
Based on a procedure similar to that in Case 1, V (x(ti+1 )) < in this study, by considering local stability, the class of sys-
V (x(ti )) for V (x(ti )) > v1 can also be obtained. The state tems satisfying Assumptions 1 and 2 is expanded compared
maintains x(t ) ∈ Ω; thus, in this case, the conditions in to global stability. Stability analysis methods with the wide
Assumption 2 are also valid at all times within the interval. applications including nonlinear mechanical systems have been
developed.
Case 3. For V (x(ti )) < v1 , the decrement of V (x(t )) in one In the proof of Theorem 2.1, the 1st and 2nd steps are
sample interval cannot be guaranteed. Hence, the upper bound similar to the approaches discussed in [30, 31]. However,
of V (x(t )) in t ∈ [ti , ti+1 ] is estimated. For any t ∈ [ti , ti+1 ], the the discussion of state dependence in the 3rd step is novel,
following relationship holds from V (x(t )) ≤ V (V (x(ti )), t ), owing to the consideration of the local stability and time- and
state-dependent disturbance.
V n = V (V (ti ), t )∕V (ti ), (24), and (25):
In the 3rd step, we proved decrement of V for V (x(ti+1 )) >
| | v1 when the sample interval is less than MSI h(v1 ) derived
V (x(t ))| ≤ V (V (x(ti )), t )| from (24). For V (x(ti+1 )) < v1 , we cannot prove decrement
|V (x(ti ))<v1 |V (x(ti ))<v1
(29) of V by existence of the disturbance. Therefore, by consider-
|
< V (V (x(ti )), t )| ≤ v1 . ing boundedness of V , we proved V (t ) ≤ v1 , ∀t ∈ [ti , ti+1 ) for
|V (x(ti ))=v1
V (x(ti+1 )) < v1 . Therefore, we prove the ultimate boundedness
For any t ∈ [ti , ti+1 ], V (x(t )) < v1 ≤ v0 holds from (29). The of the state trajectory in the three cases. Additionally, because
state maintains x(t ) ∈ Ω; thus, the conditions in Assumption 2 local Lipschitz continuity is assumed, we investigated whether
are valid at all times within the interval. the state trajectory is maintained in the assumed local region,
From Cases 1 and 2, if the sample interval is in the with consideration to the state increment.
open interval (0, h(v1 )) and v0 ≥ V (x(ti )) ≥ v1 holds, there
exists a positive constant 𝜖 satisfying V n (V (x(ti )), ti+1 ) <
3 NUMERICAL EXAMPLE
𝜖 < 1.
For a positive integer j satisfying V (x(ti+ j −1 )) ≥ v1 ,
This section presents the numerical simulation conducted to
we can obtain the following relationships from V n = verify the effectiveness of the proposed stability analysis. First,
V (V (x(ti )), t )∕V (x(ti )): the motion equation of a robot manipulator is presented in
Section 3.1. As discussed in Section 3.2, we confirms whether
V n (V (x(ti )), ti+1 ) <𝜖 the stability conditions in the theorem hold or not. The
numerical simulation results are presented in Section 3.3. Two
V (V (x(ti )), ti+1 ) <𝜖V (x(ti )) numerical simulations, wherein the parameters do not sat-
isfy the stability condition in Theorem 2.1, are presented in
V (x(ti+ j )) ≤ V (V (x(ti+ j −1 )), ti+ j ) <𝜖 j V (x(ti )). (30) Section 3.4.
UMEMOTO ET AL. 139

3.1 Controlled system definition and 3.2 Confirmation of stability conditions and
controller design derivation of MSI
Let us consider the following equation of a two-link manipula- To confirm the stability conditions of Theorem 2.1, the system
tor: is represented in the same form as (1). By defining the state
vector as x = [qT , q̇ T ]T , (31) can be rewritten as follows:
M q̈ = u + h + w, (31)
ẋ = Ax + BM −1 (u + h + w), (36)
where q ∶= [𝜃1 , 𝜃2 ]T ∈ ℝ2 denotes a joint angle; u ∈ where
ℝ2 denotes a joint torque as the control input; w ∶= [ ] [ ]
O2×2 I2 O2×2
kw [sin t , cos t ]T ∈ ℝ2 denotes the disturbance torque; M and A ∶= , B ∶= , (37)
O2×2 O2×2 I2
h are defined as (32) and (33).

[ ]
J1 + l12 m1 + l22 m2 + 4l12 m2 + 4l1 l2 m2 cos 𝜃2 l2 m2 (l2 + 2l1 cos 𝜃2 )
M ∶= , (32)
l2 m2 (l2 + 2l1 cos 𝜃2 ) J2 + m2 l22
[ ]
2l1 l2 m2 sin 𝜃2 (2𝜃̇ 1 𝜃̇ 2 + 𝜃̇ 22 ) + l1 m1 g sin 𝜃1 + 2l1 m2 g sin 𝜃1 + l2 m2 g sin(𝜃1 + 𝜃2 )
h ∶= . (33)
l2 m2 g sin(𝜃1 + 𝜃2 ) + 2l1 l2 m2 𝜃̇ 1 𝜃̇ 2 sin 𝜃2

The initial state is defined as q = [𝜋∕6, −𝜋∕6]T and q̇ = O2×2 denotes a zero 2 × 2 matrix, and I2 denotes a 2 × 2
[0, 0]T ; kw = 2.0 × 10−2 and g = 9.8 are used. The nomi- identity matrix. By comparing (1), we obtain the following
nal values of the physical parameters are ln1 = 0.5, ln2 = 0.4, relationships:
mn1 = 0.5, mn2 = 0.4, and Jn1 = 0.3, Jn2 = 0.2, and the actual
values of the physical parameters l1 , l2 have 20% uncertainty f = Ax + BMn−1 hn
compared with the nominal values ln1 , ln2 . It is assumed that Δ f = B(M −1 h − Mn−1 hn )
m1 , m2 , J1 , and J2 have errors similar to L1 and L2 . In later sim-
ulations, [l1 , l2 ]T = 0.8[ln1 , ln2 ]T , [m1 , m2 ]T = 1.2[mn1 , mn2 ]T , g = BMn−1
and [J1 , J2 ]T = 1.2[Jn1 , Jn2 ]T , were used, but we did not use Δg = B(M −1 − Mn−1 )
them in the control design. The control input is designed using
the nominal values, as follows: d = BM −1 w. (38)

Next, we derive the parameters in Assumption 1. We define


u ∶= −Mn (Kx + hn ). (34) V = x T Px using the solution of the Lyapunov equation (A −
BK )P + P (A − BK )T = Q from f + gu = (A − BK )x. Using
This controller consists of feedback linearization and linear P and Q, 𝛼1 = 𝜆min (P ), 𝛼2 = 𝜆max (P ), c1 = 𝜆min (Q), and c2 =
feedback control, and Mn and hn are obtained by replacing 2‖P‖ are defined, where 𝜆max and 𝜆min denote the maxi-
the parameters in M and h with the nominal parameters. The mum and minimum eigenvalues, respectively. Using Q = 1.0 ×
feedback gain K is designed as follows: 10−4 I4 , we can obtain the following solution of the Lyapunov
[ ]
5 0 2 0
K ∶= . (35)
0 5 0 2

Remark 3.1. (31) contains the product and squares of the system
state, and the controlled system is locally Lipschitz continuous
but not globally Lipschitz continuous. Additionally, the time-
and state-dependent disturbance w is considered.
Previously proposed approaches [26, 28–31] cannot be
applied to complex nonlinear systems, including robot manip-
ulators, and numerical examples for virtual systems have been
presented. The analysis presented in this paper has the advan-
tage of being applicable to complex nonlinear systems, such as FIGURE 3 Time response of V0 . The time when V0 is V0 = 1 for t > 0
robot manipulators. denotes the MSI
140 UMEMOTO ET AL.

FIGURE 4 Simulation results. (a) shows the time response of the state, and (b) shows the time response of V . In (b), the decrement of V is guaranteed for
V > v1 . Thus, it can be confirmed that V decreased for V > v1 . For V < v1 , it is guaranteed that V does not exceed v1 , although V may increase

FIGURE 5 Time response of input. To confirm the zero-order hold and random sampling, an enlarged view is shown in (b), and the sample interval for each
time is shown in (c)

To obtain the values of c f , the GlobalSearch function in the


MATLAB software was used. Thus, c f = 6.9273 and cu =
14.8266 were derived. Similarly, for the norm maximum, cg =
max ‖g‖ = 4.6818 was obtained. Let us determine the cΔ f and
cΔg including uncertainties. In addition to the region Ω of the
state, cΔ f can be obtained as follows:

‖Δ f (x1 ) − Δ f (x2 )‖
cΔ f = xmax ,
,x , 1 2 ‖x1 − x2 ‖
l1 , l2 ,
m1 , m2 ,
J1 , J2

FIGURE 6 Comparison between ‖x‖2 and v1 ∕𝛼1 ⎧x1 , x2 ∈Ω (41)



⎪l1 ∈ [0.8ln1 , 1.2ln1 ]
such that ⎨ ,
equation: ⎪ ⋮
⎪J ∈ [0.8Jn2 , 1.2Jn2 ]
⎩2
⎡ 0.5 0 −0.5 0 ⎤
⎢ 0 0.5 0 −0.5⎥⎥
P = 1.0 × 10−4 ⎢ . (39) with consideration to the uncertainty of the physical param-
⎢−0.5 0 1.5 0 ⎥ eters. Using GlobalSearch of MATLAB, we obtained cΔ f =
⎢ ⎥
⎣ 0 −0.5 0 0.15 ⎦ 4.1078. By considering the uncertainty in a similar manner,
we obtained the norm maximum cΔg = max ‖Δg‖ = 1.3470.
Additionally, the parameters in Assumption 2 are derived. We From ‖B‖ = 1, the norm maximum of the disturbance satisfies
derive the values c f , cu and cg , which are determined by the nom- ‖d ‖ = ‖BM −1 kw [sin t , cos t ]T ‖ ≤ kw ‖M −1 ‖. Considering the
inal values. Based on the initial state and definition of V , we state region Ω and the uncertainty, we obtained 𝜌d = 0.35837
determine v0 as v0 = V (x(0)) = 2.74 × 10−5 . Considering the from max ‖M −1 ‖ = 5.9729.
region Ω for the determined v0 , the Lipschitz constant c f can be The right-hand side of (3) was calculated to obtain
𝛼1 𝛼2
determined as follows: 𝜌d = 2.1268 × 10−5 . As can be seen, (3)
𝛼1 (c1 −c2 cΔ f )−𝛼2 c2 cΔg cu
is satisfied for v1 = 2.2 × 10−5 . Next, we calculated (4) and
‖ f (x1 ) − f (x2 )‖ obtained MSI h(v1 ) = 2.571 × 10−3 s by performing a line
c f = max . (40)
x1 , x2 ∈Ω ‖x1 − x2 ‖ search for the curve shown in Figure 3. The time response of
UMEMOTO ET AL. 141

FIGURE 7 Simulation results for disturbance gain kw = 6.0 × 10−1 . (a) shows the time response of the state, and (b) shows the time response of V

FIGURE 8 (a) shows the time response of the input, and (b) shows the comparison of ‖x‖2 and v1 ∕𝛼1 for the disturbance gain kw = 6.0 × 10−1

FIGURE 9 Simulation results for sample interval [2.0 × 10−1 , 5.0 × 10−1 ]s. (a) shows the time response of the state, and (b) shows the time response of V

V0 is shown in Figure 3. Based on this, the random values for each sample interval. The results are presented in Figures 4, 5,
the sampling time within the interval [1.0 × 10−3 , 2.5 × 10−3 ]s and 6. As shown in Figure 4, the system became stable and V
were used in the simulations discussed in Section 3.3. In Theo- decreased. Additionally, it can be seen that V did not asymptoti-
rem 2.1, there is no restriction on the lower limit of the sample cally close to 0, but remained at V < v1 after the decrease in V .
interval, but a lower limit was set to ensure feasibility. From Figure 5, it can be seen that random sampling was realized
and the input was zero-order in the sample intervals. In Figure 6,
it can be seen that (5) was satisfied in the steady state. How-
3.3 Simulation results ever, the estimation was also satisfied in the transient state and
converged to a value that was much smaller than the estimated
The simulation results were obtained by solving the differen- value. Hence, the estimation of the steady state property is con-
tial equations for constant input using ode45 in MATLAB for servative. These results support the decrement of V and the

FIGURE 10 (a) shows the time response of the input, and (b) shows the comparison of ‖x‖2 and v1 ∕𝛼1 for sample interval [2.0 × 10−1 , 5.0 × 10−1 ]s
142 UMEMOTO ET AL.

ultimate boundedness of the state, which is guaranteed by the In Assumption 1, we assume that the controlled system
theorem. Therefore, it is confirmed that the stability condition without the uncertainty and the disturbance is exponentially
in Theorem 2.1 is valid. stabilizable by a continuous controller. If the controlled sys-
tem cannot be feedback linearized, it is difficult to satisfy
Assumption 1. Relaxation of the assumption is one of our
3.4 Verification of effectiveness of the future work.
proposed stability conditions by comparative The controller discussed in this paper is a continuous static
simulations feedback controller, but approaches using discontinuous or
dynamic controllers (e.g. sliding mode control, disturbance
Theorem 2.1 is a conservative analysis based on the norm con- observer-based control) are known to increase the robust-
ditions. Therefore, this section discusses the conservativeness ness of control systems against disturbances. In future work,
and validity of the stability condition in Theorem 2.1. The stability analysis will be carried out for systems with robust
results presented in Section 3.3 are compared with the results controllers.
for which the stability condition is not satisfied by changing
the gain kw and sample interval of the disturbance from the AUTHOR CONTRIBUTIONS
condition discussed in Sections 3.1 and 3.2. Kazuki Umemoto: Conceptualization, formal analysis, investi-
The simulation results for the case wherein kw changed from gation, methodology, software, visualization, writing - original
kw = 6.0 × 10−2 to kw = 6.0 × 10−1 are presented in Figures 7 draft. Takahiro Endo: Conceptualization, validation, writing -
and 8. For the conditions discussed in Sections 3.1 and 3.2, the review and editing. Fumitoshi Matsuno: Project administration,
decrement of V is guaranteed for V ≤ v1 . However, in the case supervision, writing - review and editing.
of kw = 6.0 × 10−1 , which does not satisfy the stability condi-
tion, V increased at approximately t = 3s, as can be seen in CONFLICT OF INTEREST
Figure 7b. Figure 8b shows that the steady state error is larger The authors declare no conflicts of interest associated with
than that in Figure 6, but the condition for the steady-state error this manuscript.
‖x‖2 ≤ v1 ∕𝛼1 is satisfied. The results indicate that the stability
condition and steady state error estimation are conservative. DATA AVAILABILITY STATEMENT
Figures 9 and 10 present the results obtained when kw = The datasets generated during and/or analysed during the cur-
6.0 × 10−2 ; the sample interval was set from [1.0 × 10−3 , 2.5 × rent study are available from the corresponding author on
10−3 ]s to [2.0 × 10−1 , 5.0 × 10−1 ]s. Although V increased for reasonable request.
V > v1 , as shown in Figure 9, it can be seen from 10 that, as in
the case of kw = 6.0 × 10−1 , the estimation of the steady-state ORCID
error ‖x‖2 ≤ v1 ∕𝛼1 is satisfied. According to the results, when Kazuki Umemoto https://orcid.org/0000-0002-1148-4114
the sample interval was further extended to [3.0 × 10−1 , 7.5 ×
10−1 ]s, the control system became destabilized. Moreover, the REFERENCES
results reveal that the stability condition is conservative and that 1. Franklin, G.F., Powell, J.D., Workman, M.L., et al.: Digital Control of
the sample interval is an important parameter for stabilizing the Dynamic Systems, Vol. 3. Addison-Wesley, Menlo Park, CA (1998)
control system. In conclusion, although the stability condition 2. Ogata, K., et al.: Discrete-Time Control Systems, Vol. 2. Prentice Hall,
of Theorem 2.1 is conservative, it is confirmed that the stability Englewood Cliffs, NJ (1995)
3. Joo, Y.: Robust stabilisation of sampled-data control systems with non-
condition is valid.
linear perturbations via digital redesign. IET Control Theory Appl. 3(10),
1070–1080 (2009)
4. Yamamoto, Y.: A function space approach to sampled data control sys-
4 CONCLUSION tems and tracking problems. IEEE Trans. Autom. Control 39(4), 703–713
(1994)
This paper has presented a stability analysis based on the local 5. Raff, T., Kogel, M., Allgower, F.: Observer with sample-and-hold updating
for lipschitz nonlinear systems with nonuniformly sampled measurements.
Lipschitz continuity on a compact set of aperiodic sampling
In: 2008 American Control Conference, pp. 5254–5257. IEEE, Piscataway
feedback systems with time- and state-dependent disturbance. (2008)
Asymptotic stability cannot be guaranteed owing to the pres- 6. Pertew, A.M., Marquez, H.J., Zhao, Q.: Sampled-data stabilization of a class
ence of time- and state-dependent disturbance; however, the of nonlinear systems with application in robotics. J. Dyn. Syst. Meas. Contr.
ultimate boundedness of the state is guaranteed. Additionally, 131(2), 021008 02 (2009)
7. Malik, F.M., Malik, M.B., Munawar, K.: Sampled-data state feedback sta-
local Lipschitz continuity have been assumed, which allows the
bilization of a class of nonlinear systems based on euler approximation.
application of the proposed analysis to complex nonlinear sys- Asian J. Control 13(1), 186–197 (2011)
tems. As an example, we applied the analysis to a two-link 8. Etienne, L., Hetel, L., Efimov, D.: Observer analysis and synthesis
manipulator and verified the effectiveness of the analysis by for perturbed lipschitz systems under noisy time-varying measurements.
conducting numerical simulations. Automatica 106, 406–410 (2019)
9. Milosavljevic, C., Perunicic-Drazenovic, B., Veselic, B., Mitic, D.: Sampled
The numerical simulation results have revealed that the esti-
data quasi-sliding mode control strategies. In: 2006 IEEE International
mation of the ultimate boundedness of the state is conservative. Conference on Industrial Technology, pp. 2640–2645. IEEE, Piscataway
Therefore, the conservativeness needs to be relaxed. (2006)
UMEMOTO ET AL. 143

10. Rubagotti, M., Raimondo, D.M., Ferrara, A., Magni, L.: Robust model pre-
dictive control with integral sliding mode in continuous-time sampled-data
nonlinear systems. IEEE Trans. Autom. Control 56(3), 556–570 (2011) How to cite this article: Umemoto, K., Endo, T.,
11. Wang, J., Yang, C., Xia, J., Wu, Z.-G., Shen, H.: Observer-based slid- Matsuno, F.: Local robust stability on compact set for
ing mode control for networked fuzzy singularly perturbed systems nonlinear systems with continuous time controller
under weighted try-once-discard protocol. IEEE Trans. Fuzzy Syst. 30(6), against to aperiodic sampling and disturbance. IET
1889–1899 (2022)
Control Theory Appl. 17, 133–143 (2023).
12. Ali, S.A., Langlois, N., Guermouche, M.: Sampled-data disturbance
observer for a class of nonlinear systems. IFAC Proc. Vol. 47(3), https://doi.org/10.1049/cth2.12367
3346–3351 (2014)
13. Park, G., Joo, Y., Lee, C., Shim, H.: On robust stability of disturbance
observer for sampled-data systems under fast sampling: An almost neces-
sary and sufficient condition. In: 2015 54th IEEE Conference on Decision APPENDIX A: DETAILED DERIVATION OF
and Control (CDC), pp. 7536–7541. IEEE, Piscataway (2015) (15)
14. Ferdinando, M.D., Pepe, P., Fridman, E.: Exponential input-to-state
In (14), V (x(ti )) is constant denoting a value of V for initial
stability of globally lipschitz time-delay systems under sampled-data
noisy output feedback and actuation disturbances. Int. J. Control 94(6), time ti of the interval. Using constants k1 and k2 , (14) can be
1682–1692 (2021) described in the following simple form:
15. Hetel, L., Fiter, C., Omran, H., Seuret, A., Fridman, E., Richard, J.-P.,
Niculescu, S.I.: Recent developments on the stability of systems with ż = k1 z + k2 (A1)
aperiodic sampling: An overview. Automatica 76, 309–335 (2017)
16. He, Y.: Overview of recent advances in stability of linear systems with time- where
varying delays. IET Control Theory Appl. 13(15), 1–16 (2019)
17. Fujioka, H.: Stability analysis of systems with aperiodic sample-and-hold k1 ∶=c f + cΔ f (A2)
devices. Automatica 45(3), 771–775 (2009)
18. Fujioka, H., Nakai, T.: Stabilising systems with aperiodic sample-and-hold c f + cΔ f + (cg + cΔg )cu
devices: state feedback case. IET Control Theory Appl. 4(7), 265–272 k2 ∶= V (x(ti ))1∕2 + 𝜌d (A3)
1∕2
(2010) 𝛼1
19. Oishi, Y., Fujioka, H.: Stability and stabilization of aperiodic sampled-data
control systems using robust linear matrix inequalities. Automatica 46(8), We can obtain the following using the transformation z1 ∶=
1327–1333 (2010)
z + k2 ∕k1 :
20. Fridman, E., Seuret, A., Richard, J.-P.: Robust sampled-data stabilization
of linear systems: an input delay approach. Automatica 40(8), 1441–1446
(2004) ż 1 = k1 z1 (A4)
21. Sun, J., Chen, G., Chen, J.: Stability analysis of aperiodic sampled-data sys-
tems: A switched polytopic system method. IEEE Trans. Circuits Syst. II The general solution of (A4) is
Exp. Briefs 67(6), 1054–1058 (2019)
22. Nešić, D., Teel, A., Kokotović, P.: Sufficient conditions for stabilization z1 (t ) = k0 ek1 t (A5)
of sampled-data nonlinear systems via discrete-time approximations. Syst.
Control Lett. 38(4-5), 259–270 (1999) where k0 is a constant of integration. With initial condition
23. Ling, Q., Yan, Z., Ji, H., Wang, Y., Xie, Y.: Sufficient conditions to stabi-
z(ti ) = 0,
lize time-varying nonlinear sampled-data systems via approximation. Int. J.
Robust Nonlinear Control 27(1), 108–120 (2017)
24. Bernuau, E., Moulay, E., Coirault, P.: Stability of homogeneous nonlinear z1 (ti ) = k2 ∕k1 = k0 ek1 ti
systems with sampled-data inputs. Automatica 85, 349–355 (2017)
k2 −k1ti
25. Du, H., Qian, C., Li, S., Chu, Z.: Global sampled-data output feedback k0 = e (A6)
stabilization for a class of uncertain nonlinear systems. Automatica 99, k1
403–411 (2019)
26. Mazenc, F., Malisoff, M., Dinh, T.N.: Robustness of nonlinear systems with can be obtained from z1 (ti ) = k2 ∕k1 . Substituting (A6) in
respect to delay and sampling of the controls. Automatica 49(6), 1925– (A5),
1931 (2013)
27. Li, H., Li, C., Zhang, W., Cao, Z.: Exponential stabilization for nonlin- k2 k1 (t −ti )
ear coupled dynamical systems via impulsive and sampled-data control z1 (t ) = e
with input constraints. Int. J. Robust Nonlinear Control 29(17), 6126–6144
k1
(2019) k2 ( k1 (t −ti ) )
28. Omran, H., Hetel, L., Petreczky, M., Richard, J.-P., Lamnabhi-Lagarrigue, z(t ) = e −1
F.: Stability analysis of some classes of input-affine nonlinear systems with k1
aperiodic sampled-data control. Automatica 70, 266–274 (2016)
V (x(ti ))1∕2 c f + cΔ f + (cg + cΔg )cu ( (c f +cΔ f )(t −ti ) )
29. Mao, J., Xiang, Z., Zhai, G., Guo, J.: Adaptive practical stabilization of a z(t ) = e −1
class of uncertain nonlinear systems via sampled-data control. Nonlinear
𝛼
1∕2 c f + c Δf
Dyn. 92(4), 1679–1694 (2018) 1
30. Chu, H., Qian, C., Yang, J., Xu, S., Liu, Y.: Almost disturbance decoupling 𝜌d ( (c f +cΔ f )(t −ti ) )
for a class of nonlinear systems via sampled-data output feedback control. + e −1 (A7)
Int. J. Robust Nonlinear Control 26(10), 2201–2215 (2016)
c f + cΔ f
31. Li, Z., Zhao, J.: Output feedback stabilization for a general class of non-
linear systems via sampled-data control. Int. J. Robust Nonlinear Control is obtained from z = z1 − k2 ∕k1 , (A2), and (A3). Using (A7),
28(7), 2853–2867 (2018) (16), and (17), Equation (15) can be obtained.

You might also like