Received 17 December 2024, accepted 16 January 2025, date of publication 20 January 2025, date of current version 24 January 2025.

Digital Object Identifier 10.1109/ACCESS.2025.3531559

Fault Estimation Method for Nonlinear Systems

With Perturbations Based on Augmented
Fault Observer
YANXIU SUN 1, YING DU1 , AND YUPING ZHANG2
1 Basic Course Department, Shenyang Institute of Technology, Fushun 113122, China
2 College of Science, Liaoning Petrochemical University, Fushun 113001, China

Corresponding author: Yanxiu Sun ([email protected])

This work was supported by the Basic Scientific Research Project of Liaoning Provincial Department of Education under Grant
LJ212413201004.

ABSTRACT This paper proposes a robust fault estimation method for nonlinear systems with both actuator
and sensor faults. Specifically, a singular system is constructed by combining the state vector of the system
with the fault vector of the sensor, and an augmented fault observer is designed by combining the fault
estimation algorithm of the adaptive actuator. First, an augmented fault observer for linear systems with
perturbations is designed, and the sufficient conditions for the asymptotic stability of the observation
error system are ascertained so that the state of the singular system can be observed. Second, considering
the nonlinear characteristics of the actual control system, sufficient conditions for the existence of the
augmented fault observer are determined. Finally, the design method of augmented fault observer for
nonlinear systems is extended to nonlinear systems with both input interference and sensor disturbance, and
sufficient conditions for the existence of the observer are found. In the design of the observer, the influence of
disturbance on the system fault estimation, convergence speed of fault estimation error, and convenience of
observer design are considered. Moreover, an estimation method is devised for the fault diagnosis of actuators
and sensors in the control system. Finally, a numerical example that tests the design method of augmented
fault observer and effect of system fault estimation is presented. The superiority and effectiveness of the
method are thus verified.

INDEX TERMS Fault observer, fault reconstruction, linear matrix inequalities, singular systems.

I. INTRODUCTION the control system, it is particularly important to detect and

In recent years, with the continuous increasing application deal with faults quickly.
of process control technology to modern industry, urban The purpose of system detection and diagnosis is to deter-
transportation, electric power, aerospace, and other fields, the mine the location and magnitude of the system fault and
system has become increasingly large and more complex. then perform appropriate operations to eliminate or reduce
Because the sensors and actuators in the control system are the impact of the fault on the overall system performance to
in operation for a long time, the possibility of sensor or ensure the smooth and safe operation of the system. In the
actuator failure is greatly increased. If system failure is not past 40 years, dynamic system fault detection and diagno-
be detected and dealt with as soon as possible, there may be sis has attracted significant attention from researchers, and
serious consequences. Based on the urgent need for safety fault detection and diagnosis methods have been successfully
and reliability of the system, to ensure the safe operation of applied to some practical systems [1], [2], [3], [4]. Among
system fault diagnosis methods, adaptive fault estimation
The associate editor coordinating the review of this manuscript and observer has the advantage of estimating both state and fault
approving it for publication was Mohammad AlShabi . information simultaneously and is suitable for most cases
2025 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License.
VOLUME 13, 2025 For more information, see https://creativecommons.org/licenses/by/4.0/ 15301
 Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

where the actual system state variables are not completely perform robust fault estimations based on fault observers.
measurable [5], [6], [7], [8]. At present, in the research on In another study [28], actuator fault estimation for a system
system fault detection and diagnosis, the method of robust with variable parameters was realized by constructing a gen-
fault estimation has achieved notable results. Prior studies eralized learning observer.
have extensively examined the problem of sensor fault esti- The aforedescribed studies have examined the fault esti-
mation in control systems [9], [10], [11], [12], [13], [14], mation and diagnosis of different types of control systems.
[15], [16], [17], [18]. The problem of actuator fault estimation However, most of them have only solved the problem of sen-
in control systems have also been widely studied [19], [20], sor or actuator fault estimation and diagnosis, and relatively
[21], [22], [23], [24], [25], [26], [27], [28], [29], [30]. In the few studies have estimated and diagnosed both types of faults
research of system fault diagnosis, the estimation of system in the system. Problems involving both sensor and actuator
fault based on system state observer plays an important role, faults have been studied [29], [30], [31]. The control methods
and the robust control of system based on observer is often based on state observer are presented respectively in three-
applied to practical problems [31], [32]. level neutral-point-clamped converter and buck converter
A robust estimation method for diagnosing sensor faults system [32], [33]. In [34], the nonlinear part of the system was
was proposed in [9]. A fault observer was designed to achieve not considered. Simultaneously, the degree of self-adaptation
robust estimation of faults. However, the convergence of fault could be further optimized in the actuator fault reconstruc-
estimation errors was not discussed in the process of sensor tion. Kheloufi et al. [35] examined the sensor and actuator
fault estimation, and the design of the fault observer could be fault estimation problem and proposed a fault reconstruction
further optimized. Wang et al. [10] proposed a sensor fault method. Their fault estimation observer is relatively general;
estimation method based on ellipsoid beam for generalized however, several unknown matrices need to be solved, and the
systems with unknown but bounded uncertainties. Yan-xiu problem of the convergence rate of estimation errors warrants
and Hong [11] developed a fault observer-based sensor further investigation.
fault estimation method for nonlinear discrete systems with On the basis of the aforedescribed analyses and previous
unknown disturbances. Further, Hosseini et al. [12] proposed findings, an adaptive augmentation fault observer, which can
a reset unknown input observer to complete the robust esti- achieve simultaneous estimation of sensor and actuator faults
mation of sensor faults. For a class of nonlinear multi-agent for nonlinear control systems with both external and sensor
systems with sensor faults, Chen et al. [13] proposed a new disturbances, is proposed in this paper. Compared with the
dynamic output feedback controller design method to achieve existing methods, the proposed method is more concise in
robust system stabilization. Sun and Li [14] devised a sensor solving the gain matrix, and the designed adaptive augmen-
fault estimation method based on fault observer for a class tation fault observer is easier to implement in engineering.
of interconnected large systems. Further, a design method of Finally, the method is applied to the fault diagnosis of a
unknown input observer has been proposed, and sufficient manipulator system and nonlinear DC motor system; and the
conditions for the existence of unknown input observer have simulation results verify the feasibility and effectiveness of
been deduced, thus providing a basis for fault estimation [15], the proposed method.
[16]. A distributed fault estimation method has been proposed In this paper, the problem of robust fault estima-
for time-varying systems based on digital communication tion of nonlinear systems based on an augmented
channels, successfully realizing fault estimation [17]. fault observer is examined. The main contributions are
Robust estimations of actuator faults contained in the con- three-fold:
trol system have been reported in prior studies [18], [19]. 1) For a class of linear systems with input perturbations,
Reference [18] presents the effective estimation of actuator an augmented fault observer design scheme is proposed, and
faults existing in the system, accounting for the convergence sufficient conditions for the existence of the fault observer
rate of fault estimation errors of linear systems; however, are provided in the form of linear matrix inequalities. On the
only the actuator faults existing in the system are considered basis of the fault observer, the simultaneous estimation of
in this system. In [19], although the nonlinear part of the sensor and actuator faults is realized, and the stability margin
actual system was considered, the convergence rate of the ε is introduced to apply a quadratic constraint on the observer
fault estimation error was not. In further studies [20], [21], gain matrix, which improves the fault estimation performance
[22], a proportional integral unknown input observer based on of the fault observer.
Lyapunov theory was proposed to achieve robust estimation 2) Considering the actual environment of the system, the
of actuator faults. A four-rotor unmanned aircraft system fault estimation problem of the nonlinear system is studied.
with actuator faults has been studied in [23]. Considering By analyzing the nonlinear term, sufficient conditions for the
the initial conditions and noise sensitivity, a fractional order existence of the augmented fault observer for the nonlinear
sliding mode observer has been designed to achieve fault esti- system are ascertained; thus, the designed fault observer is
mation. To address the problem of actuator fault estimation robust and practical. Simultaneously, the method is extended
in a linear parameter change model [24], networked system to double fault estimation with input and sensor disturbance;
model [25], singular system [26], and time-delay system [27], this helps considerably advance the theory of adaptive aug-
a Lyapunov-Krasovskii functional technique was used to mentation fault observer.

15302 VOLUME 13, 2025

Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

3) To suppress the interference of external disturbances (1) S < 0;

and improve the robustness of fault estimation, in the design (2) S11 < 0, S22 − S12 T S −1 S < 0;
11 12
process of the fault observer, disturbance terms are classified (3) S22 < 0, S11 − S12 S22 S12 < 0.
−1 T

and integrated according to their properties, and the influence To estimate actuator and sensor faults simultaneously, the
of disturbance on fault estimation is weakened through H∞ following augmentation singular system is constructed:
performance indicators. (
The proposed design method of augmented fault observer ˙ = Āx̄(t) + Bu(t) + Fa fa (t) + Dd(t)
E x̄(t)
(2)
makes it more convenient to solve the gain matrix and is y(t) = C̄ x̄(t).
conducive to realizing the fast and effective fault estimation of  
sensors and actuators. Therefore, the fault estimation method x(t)
where x̄(t) = ∈ Rn+s is the augmented vector, and
has wide application prospects.  fs (t)  
E = In 0 , Ā = A 0 , C̄ = C Fs are matrices of
  
The remainder of this article is organized as follows. In the
Section II, the structure of a class of linear systems with appropriate dimensions.
disturbance is described, and the sufficient conditions for the In the singular system (2), the augmentation vector con-
existence of the system fault observer and design method tains the sensor fault vector. If an efficient observer can be
are presented. In Section III, a design method of augmented designed for system (2), then state x(t) and sensor fault fs (t)
fault observer for nonlinear systems with unknown input can be estimated directly.  
perturbations is presented. Further, the method is extended to E
According to Assumption 1, the matrix is a column
nonlinear systems with both input and sensor perturbations, "   # C̄
T
and sufficient conditions for the existence of an augmented E E
full rank matrix, so matrix , is a non-singular
fault observer are provided, thus affording a theoretical basis C̄ C̄
for simultaneous estimation of sensor and actuator faults. matrix, that is, the matrix is invertible. Thus, let
In Section IV, the proposed method is tested by robot arm "   #−1  
T T
system and nonlinear DC motor system models, which show E E E
,
 
T N = (3)
the feasibility of the robust fault estimation method. Finally, C̄ C̄ C̄
Section V summarizes the advantages of the proposed method
and suggests future research directions. Thus, we get the following equation:
Definitions: I represents the identity matrix of appropriate TE + N C̄ = In+s (4)
dimension, 0 represents the scalar or matrix of appropriate
dimension; ∥ · ∥ is the Euclidean norm; for any matrix A, According to (2) and (4),
A > 0 represents a positive definite matrix; A T and A−1
(
˙ = T Āx̄(t) + TBu(t) + TFa fa (t) + TDd(t) + N ẏ(t)
x̄(t)
represent the transpose and inverse of matrix A, respectively;
and ∗ represents a symmetric term in a symmetric matrix. y(t) = C̄ x̄(t).
(5)
II. DESIGN OF LINEAR SYSTEM FAULT OBSERVER
A. SYSTEM DESCRIPTION
where T and N are the matrices of appropriate dimensions
in (4).
Consider the following linear systems with actuator and sen-
According to (2)–(5), the following subsection designs an
sor failures:
( augmented fault observer to complete the robust estimation
ẋ(t) = Ax(t) + Bu(t) + Fa fa (t) + Dd(t) of system faults.
(1)
y(t) = Cx(t) + Fs fs (t).
B. DESIGN OF AUGMENTED FAULT OBSERVER
where
Considering the simultaneous estimation of sensor and actu-
x(t) ∈ Rn , u(t) ∈ Rm , y(t) ∈ Rp , fa (t) ∈ Rq , fs (t) ∈ Rs ,
ator faults, an observer of the following form is constructed
and d(t) ∈ Rl are the system state vector, control input,
based on (5):
output, actuator fault, sensor fault, and unknown disturbance
term, respectively. A, B, C, D, Fa , Fs are constant matrices of


 ż(t) = T Āz(t)+TBu(t)+TFa fˆa (t)+T ĀNy(t)−L[ŷ(t)−y(t)]
appropriate dimensions.

 ˆ = z(t) + Ny(t)
 x̄(t)
Assumption 1: Matrix Fs in system (1) is column full rank. ˆ
ŷ(t) = C̄ x̄(t)
Assumption 2: The first derivative of the actuator fault in


 f˙ˆ (t) = fˆ (t) − Re (t)


system (1) exists, satisfying f˙a (t) ∈ L2 [0, +∞). a a y
Assumption 3: The external disturbance term d(t) in sys- (6)
tem (1) is bounded and satisfies d(t) ∈ L2 [0, +∞).
Lemma 1: (Schur complement lemma) [35] ˆ
where x̄(t) ∈ Rn+s , ŷ(t) ∈ Rp , and fˆa (t) ∈ Rq are the
The following
 conditions  are equivalent for symmetric observer’s state vector, output vector, and actuator fault esti-
S11 S12 mates, respectively. The matrices L, R are the gain matrices
matrices S = T :
S12 S22 to be designed.

VOLUME 13, 2025 15303

 Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

Let satisfying the linear matrix inequalities (12) and (13), then

ˆ − x̄(t) the error dynamic system (11) is asymptotically stable and
 ex̄ (t) = x̄(t)

satisfies the conditions ∥ẽ(t)∥ ≤ α d̃(t) .
efa (t) = fˆa (t) − fa (t) (7)

ey (t) = ŷ(t) − y(t)
 T 
à P − C̃ T X T + Pà − X C̃ PD̃

In+s+q
∗ −αIl+2q 0 <0
According to (5)–(7), the dynamic equation of observation

error can be obtained as follows: ∗ ∗ −αIn+s+q
˙ˆ − x̄(t) (12)
ė (t) = x̄(t)
x̄ ˙
à P − C̃ X + Pà − X C̃ + 2εP < 0
T T T
(13)
= T Āz(t) + TBu(t) + TFa fˆa (t)  
+ T ĀNy(t) − L[ŷ(t) − y(t)] L
where X = PL̃, L̃ = .
R
+ N ẏ(t) − T Āx̄(t) − TBu(t) Proof: Select the Lyapunov function
− TFa fa (t) − TDd(t) − N ẏ(t)
V (t) = ẽT (t)Pẽ(t)
= (T Ā − L C̄)ex̄ (t) + TFa efa (t) − TDd(t) (8)
Taking the derivative with respect to time, we get
Simultaneously, it can be concluded that
V̇ (t) = ẽ˙ T (t)Pẽ(t) + ẽ T (t)Pẽ(t)
˙
ėfa (t) = f˙ˆa (t) − f˙a (t) = [(Ã − L̃ C̃)ẽ(t) + D̃d̃(t)] T Pẽ(t)
= fˆa (t) − Rey (t) − f˙a (t) + ẽT (t)P[(Ã − L̃ C̃)ẽ(t) + D̃d̃(t)]
= −Rey (t) + fˆa (t) − fa (t) + fa (t) − f˙a (t) = ẽT (t)[(Ã− L̃ C̃) T P+ P(Ã− L̃ C̃)]ẽ(t)+2ẽ T (t)PD̃d̃(t)
= −RC̄ex̄ (t) + efa (t) + fa (t) − f˙a (t) (9) (14)

According to (8) and (9), the following fault estimation error Considering the influence of system disturbance on fault
system can be obtained: reconstruction, it is possible to ensure
       ∥ẽ(t)∥ < α d̃(t) (15)
ėx̄ (t) T Ā − L C̄ TFa ex̄ (t) TDd(t)
= − ˙
ėfa (t) −RC̄ Iq efa (t) fa (t) − fa (t) Let
(10) Z ∞ ẽT (t)ẽ(t) − α 2 d̃ T (t)d̃(t)
J= dt
For ease of expression, (10) is rewritten as follows: 0 α
According to the Lyapunov function and (14), the following
(
˙ = (Ã − L̃ C̃)ẽ(t) + D̃d̃(t)
ẽ(t)
(11) inequality is established:
efa (t) = Ĩq ẽ(t)
ẽ (t)ẽ(t) − α 2 d̃ T (t)d̃(t)
Z ∞ T
where J< [ + V̇ ]dt (16)
0 α
   
T Ā TFa L To give sufficient conditions for inequality (16) to hold,
à = , L̃ = ,
0 Iq R let
 
−TD 0 0
C̃ = C̄ 0 , D̃ = ,
 
0 Iq −Iq ẽT (t)ẽ(t) − α 2 d̃ T (t)d̃(t)
J̃ = + V̇ (17)
  α
  d(t)
ėx̄ (t) T 2 T
, d̃(t) = fa (t) , Ĩq = 0 Iq . Then, J̃ = ẽ (t)ẽ(t)−αα d̃ (t)d̃(t) + V̇ < 0 can be deduced by
 
ẽ(t) =
ėfa (t)
f˙a (t) J < 0.
Further, we can guarantee
To suppress the influence of unknown input d̃(t) on fault esti-
mation, The performance indicator α > 0 is now introduced, 1 T
ẽ (t)ẽ(t) < α d̃ T (t)d̃(t) (18)
giving ∥ẽ(t)∥ ≤ α d̃(t) . α
As evident, a smaller value of α can weaken the influence and
of unknown input on fault estimation. ∥ẽ(t)∥ < α d̃(t)
Next, sufficient conditions for asymptotically stable fault
estimation error of system (11) are given in the form of linear According to (14) and (17), we get
matrix inequalities.
Theorem 1: Considers a nonlinear dynamic system (5) and 1 T
J̃ = ẽ (t)ẽ(t) − α d̃ T (t)d̃(t) + ẽT (t)[(Ã − L̃ C̃) T P
observer (6). For a given scalar α, ε > 0, if there is a α
symmetric positive definite matrix P and gain matrices L, R + P(Ã − L̃ C̃)]ẽ(t) + 2ẽT (t)PD̃d̃(t)

15304 VOLUME 13, 2025

Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

Algorithm 1 Design Process of Augmented Fault Observer III. DESIGN OF AUGMENTED FAULT OBSERVER FOR
(1) Construct an augmentation system as shown in (2). NONLINEAR SYSTEMS
(2) Matrices T and N are obtained according to (3). A. CONSIDERING THE CASE OF UNKOWN INPUT
(3) Set H∞ performance index α and stability margin ε, DISTURBANCE
and verify the existence of the augmented fault observer A class of Lipschitz nonlinear dynamic systems with sensor
according to Theorem 1. and actuator faults is considered. The system form is as
(4) If feasible solutions for the matrix inequalities (12) follows:
and (13) exist, the gain matrices L and R are calculated. (
(5) Complete the design of the augmented fault observer, ẋ(t) = Ax(t) + Bu(t) + g[x(t), t] + Fa fa (t) + Dd(t)
complete the effective estimation of the sensor fault by fˆs (t) = y(t) = Cx(t) + Fs fs (t).
ˆ
 
0 Is x̄(t), and complete the effective estimation of the actu- (19)
ator fault according to (6).
where g[x(t), t] are nonlinear functions that satisfy the
Lipschitz condition, i.e.,
1
= ẽT (t)[(Ã − L̃ C̃) T P + P(Ã − L̃ C̃) + I ]ẽ(t) ∥g[x(t), t] − g[x̃(t), t]∥ ≤ Lg ∥x(t) − x̃(t)∥ (20)
α
+ 2ẽ (t)PD̃d̃(t) − α d̃ (t)d̃(t)
T T
where Lg is the Lipschitz constant, and the other symbols are
= ξ (t)6ξ (t)
T defined in accordance with (1).
Based on (2), system (19) can be rewritten as follows:
where (
˙ = Āx̄(t) + Bu(t) + g[E x̄(t), t] + Fa fa (t) + Dd(t)
E x̄(t)
 
ẽ(t)
ξ (t) = , y(t) = C̄ x̄(t).
d̃(t)
(21)
(Ã − L̃ C̃)T P + P(Ã − L̃ C̃) + α1 I PD̃
 
6= . According to (4) and (21), the nonlinear augmentation fault
∗ −αIl+2q
observer can be designed as follows:
According to Lemma 1, the inequality 6 < 0 is equiv- 
alent to inequality (12), and the error dynamic system 
 ˆ
ż(t) = T Āz(t) + TBu(t) + Tg[E x̄(t), t] + TFa fˆa (t)
(11) is asymptotically stable and satisfies ∥ẽ(t)∥ <

+ T ĀNy(t) − L[ŷ(t) − y(t)]




α d̃(t) . ˆx̄(t) = z(t) + Ny(t)
Considering the convergence rate of fault estimation error, ˆ



 ŷ(t) = C̄ x̄(t)
the gain matrix of the fault observer is quadratically con-

 ˙ˆ

f (t) = fˆ (t) − Re (t)
a a y
strained here, and the constraint condition is inequality (13).
In this way, the fault estimation error can converge with the (22)
stability margin index of ε, achieving a better fault estimation
According to (4), (21), and (22), the dynamic equation for
effect. Thus, Theorem 1 is proved.
estimating the error can be derived as follows:
Remark 1: Theorem 1 designs an augmented fault (
observer for linear system fault estimation with external per- ėx̄ (t) = (T Ā − L C̄)ex̄ (t) + 1ḡ + TFa efa (t) − TDd(t)
turbations, and gives sufficient conditions for the existence ėfa (t) = −RC̄ex̄ (t) + efa (t) + fa (t) − f˙a (t)
of the observer in the form of linear matrix inequalities.
(23)
In the design of the observer, considering the influence of
disturbance on fault estimation and the convergence rate where 1ḡ(t) = Tg[E x̄(t),ˆ t] − Tg[E x̄(t), t].
of estimation error, the influence of disturbance on fault Equation (23) can be rewritten as follows:
estimation is suppressed by setting H∞ performance indexes. (
˙ = (Ã − L̃ C̃)ẽ(t) + 1g̃(t) + D̃d̃(t)
Simultaneously, the stability margin of ε is introduced to ẽ(t)
(24)
speed up the convergence of estimation errors and improve efa (t) = Ĩq ẽ(t)
the performance of fault estimation. h iT
The following section discusses the design problem of the where 1g̃(t) = 1ḡT (t) 0Tq×1 , and the other matrices and
augmented fault observer for nonlinear systems on the basis vectors are defined in the same way as in (11).
of the aforedescribed design method of the augmented fault The following are the sufficient conditions for the existence
observer. By analyzing the Lipschitz nonlinear term in the of augmented fault observers for nonlinear systems:
system, sufficient conditions for the existence of the aug- Theorem 2: For the system (21), according to Assump-
mented fault observer are ascertained so as to realize the tions 1–3 and the conditions of (20), for a given scalar α, ε >
robust fault estimation of the sensor and actuator in the non- 0, if there is a positive definite matrix P and gain matrices
linear system. L, R, such that the following linear matrix inequalities (25)

VOLUME 13, 2025 15305

 Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

and (26) are both true, then the error dynamic system (24) Considering the influence of system disturbance on fault
is robust and asymptotically stable, and condition ∥ẽ(t)∥ ≤ estimation of nonlinear systems, according to the content of
T 2 T
α d̃(t) is satisfied: the previous section, inequality J̃ = ẽ (t)ẽ(t)−αα d̃ (t)d̃(t) +
V̇ (t) < 0 can be derived from inequality ∥ẽ(t)∥ ≤ α d̃(t) ,
911
 
P PD̃ In+s+q
and the influence of disturbance on fault estimation can be
 ∗ −In+s+q 0 0  < 0 (25) weakened by setting H∞ performance index α.


 ∗ ∗ −αIl+2q 0 
Next, a sufficient condition is given to make the error sys-
∗ ∗ ∗ −αIn+s+q
tem (24) asymptotically stable under the condition ∥ẽ(t)∥ ≤
ÃT P − C̃ T X T + PÃ − X C̃ + 2εP < 0 (26) α d̃(t) .
where According to (27), (29) becomes
911 = ÃT P − C̃ T X T + PÃ − X C̃ + M̃ , 1 T
J̃ = ẽ (t)ẽ(t) − α d̃ T (t)d̃(t) + ẽT (t)[(Ã − L̃ C̃) T P
∥T ∥2 (Lg E)T (Lg E) 0 α
 
M̃ = , + P(Ã − L̃ C̃)]ẽ(t) + 2ẽT (t)P1g̃(t) + 2ẽT (t)PD̃d̃(t)
0 0
1
X = PL̃, = ẽT (t)[(Ã − L̃ C̃) T P + P(Ã − L̃ C̃) + I ]ẽ(t)
  α
L
L̃ = . + 2ẽT (t)P1g̃(t) + 2ẽT (t)PD̃d̃(t) − α d̃ T (t)d̃(t)
R
1
Proof: Select the Lyapunov function ≤ ẽT (t)[(Ã − L̃ C̃) T P + P(Ã − L̃ C̃) + I ]ẽ(t)
α
V (t) = ẽT (t)Pẽ(t) + 2ẽT (t)P1g̃(t) + 2ẽ T (t)PD̃d̃(t) − α d̃ T (t)d̃(t)
The derivative of this function can be written as follows: + ẽT (t)M̃ ẽ(t) − 1g̃T (t)In+s+q 1g̃(t)
1
V̇ (t) = ẽ˙ T (t)Pẽ(t) + ẽ T (t)Pẽ(t)
˙ = ẽT (t)[(Ã − L̃ C̃) T P + P(Ã − L̃ C̃) + I + M̃ ]ẽ(t)
α
= [(Ã − L̃ C̃)ẽ(t) + 1g̃(t) + D̃d̃(t)]T Pẽ(t) + 2ẽT (t)P1g̃(t) + 2ẽ T (t)PD̃d̃(t) − α d̃ T (t)d̃(t)
+ ẽT (t)P[(Ã − L̃ C̃)ẽ(t) + 1g̃(t) + D̃d̃(t)] − 1g̃T (t)In+s+q 1g̃(t)
T T
= ẽ (t)[(Ã − L̃ C̃) P + P(Ã − L̃ C̃)]ẽ(t) = ηT (t)4η(t)
T T
+ 2ẽ (t)P1g̃(t) + 2ẽ (t)PD̃d̃(t) (27)
where η(t) and 4, as shown at the bottom of the next page.
According to (20), the following relation can be obtained: According to Lemma 1, matrix inequality 4 < 0 is
1g̃T (t)1g̃(t) = 1g̃T (t)In+s+q 1g̃(t) equivalent to inequality (25), which makes the error system
 T  of (24) asymptotically stable and satisfies ∥ẽ(t)∥ ≤ α d̃(t) .
ˆ
= g[E x̄(t), t] − g[E x̄(t), t] T 0 T Simultaneously, in order to accelerate the convergence rate
"  # of the fault estimation error, the gain matrix is constrained
ˆ
T g[E x̄(t), t] − g[E x̄(t), t] quadratic by inequality (26) to further improve the estimation
×
0 performance of the fault observer. Thus, theorem 2 is proved.
 T Remark 2: In this section, considering the general phe-
≤ ∥T ∥2 g[E x̄(t),
ˆ t] − g[E x̄(t), t] nomenon of nonlinear terms in the system, by analyzing

ˆ
 the Lipschitz condition satisfied by the nonlinear terms, suf-
× g[E x̄(t), t] − g[E x̄(t), t] ficient conditions for the existence of an augmented fault
≤ ∥T ∥2 [x̄(t)
ˆ − x̄(t)]T E T LgT Lg E[x̄(t)
ˆ − x̄(t)] observer are presented; the proposed fault estimation method
is relatively general and expands its application range.
= ∥T ∥2 [Lg E(In+s 0(n+s)×q )ẽ(t)]T [Lg E(In+s 0(n+s)×q )ẽ(t)]
= ∥T ∥2 ẽT (t)[Lg E 0n×q ]T [Lg E 0n×q ]ẽ(t) B. CONSIDERING THE CASE WITH BOTH UNKNOWN
∥T ∥2 (Lg E)T (Lg E) 0 INPUT AND SENSOR DISTURBANCES
 
= ẽT (t) ẽ(t) Consider a class of nonlinear dynamic systems with unknown
0 0
input and sensor disturbances. The system form is as follows:
= ẽT (t)M̃ ẽ(t). (28)
(
where ẋ(t) = Ax(t) + Bu(t) + g[x(t), t] + Fa fa (t) + Dd(t)
∥T ∥2 (Lg E)T (Lg E) y(t) = Cx(t) + Fs fs (t) + Ds ds (t).
 
0
M̃ = .
0 0 (30)
Thus,
where ds (t) ∈ Rs represents a bounded sensor distur-
ẽ (t)M̃ ẽ(t) − 1g̃ (t)In+s+q 1g̃(t) ≥ 0
T T
(29) bance, the first derivative of the disturbance satisfies ḋs (t) ∈

15306 VOLUME 13, 2025

Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

Algorithm 2 The Design Process of Augmented Fault can be obtained from Theorem 3 and replacing the matrix D̃
Observer for Nonlinear Systems Is as Follows: in inequality (25) with the augmentation vector D̂ in (34).
(1) Construct the augmentation system as shown in (21). Theorem 3: For the system (31), according to Assump-
(2) Matrices T and N are obtained according to (3). tions 1–3 and conditions (20), for a given scalar α, ε > 0,
(3) Set H∞ performance index α and stability margin ε, if there is a positive definite matrix P and again matrices L, R
and verify the existence of the augmented fault observer such that the following linear matrix inequalities (35) and (36)
according to Theorem 2. hold simultaneously, then the error dynamic system (34) is
(4) If there are feasible solutions to the matrix inequali- robust and asymptotically stable, and condition ∥ẽ(t)∥ ≤
ties (25) and (26), the gain matrices L and R are computed. α d̂(t) is satisfied.
(5) Complete the design of the augmented fault observer,
ψ11
 
obtain the sensor fault estimate, and complete the actuator P PD̃ XDs PÑ Ds In+s+q
fault estimate according to (22).  ∗ −In+s+q
 0 0 0 0 

 ∗
 ∗ −αIl+2q 0 0 0 <0

 ∗
 ∗ ∗ −αIs 0 0 

L2 [0, +∞), Ds ∈ Rp×s is a known constant matrix, and other  ∗ ∗ ∗ ∗ −αIs 0 
symbols are the same as those in (19). ∗ ∗ ∗ ∗ ∗ −αIn+s+q
Similar to (2), (30) can be rewritten as follows: (35)
(
˙ = Āx̄(t) + Bu(t) + g[E x̄(t), t] + Fa fa (t) + Dd(t)
E x̄(t) Ã P − C̃ X + PÃ − X C̃ + 2εP < 0
T T T
(36)
y(t) = C̄ x̄(t) + Ds ds (t) where
(31)
911 = ÃT P − C̃ T X T + PÃ − X C̃ + M̃ ,
Every vector and matrix definition is the same as before.
   
−TD 0 0 N
The system (31) fault observer can be designed in the form D̃ , Ñ = ,
0 Iq −Iq 0
of (22), and according to (22) and (31), it can be obtained as
∥T ∥2 (Lg E)T (Lg E) 0(n+s)×q
 
follows: M̃ = ,
 0q×(n+s) 0q×q
 ėx̄ (t) = (T Ā − L C̄)ex̄ (t) + 1ḡ + TFa efa (t) − TDd(t)
 X = PL̃,
+ NDs ḋs (t) + LDs ds (t)  
L


ėfa (t) = −RC̄ex̄ (t) + efa (t) + RDs ds (t) + fa (t) − f˙a (t) L̃ = .
R
(32)  
Proof: Matrix D̂ = D̃ D̄ is replaced by D̃ in (25) to
The error system (32) can be written as follows: complete the proof; the full proof process is omitted herein.
( Remark 3: In addition to the dual faults of the sensor and
˙ = (Ã − L̃ C̃)ẽ(t) + 1g̃(t) + D̃d̃(t) + D̄d̄(t)
ẽ(t)
(33) actuator, the nonlinear system introduced in this section also
efa (t) = Ĩq ẽ(t) adds sensor disturbance, thus strengthening the influence of
the disturbance part on the fault estimation. In the design
In the formulas,

LDs NDs

T process of the augmented fault observer, this study com-
and d̄(t) = dsT (t) ḋsT (t) , the matrices

D̄ = bines the external and sensor disturbances to design the H∞
RDs 0q×s
and vectors are expressed in the same way as (11) and (24). performance indicators of the fault estimation as a whole,
To give sufficient conditions for the asymptotic stability gives the conditions for the existence of the observer more
of (33), it is rewritten as follows: conveniently, which is more conducive to the design of the
( augmented fault observer, and realizes the robust estima-
˙ = (Ã − L̃ C̃)ẽ(t) + 1g̃(t) + D̂d̂(t)
ẽ(t) tion of the system fault. The specific design process of the
(34)
efa (t) = Ĩq ẽ(t) observer is the same as that of Algorithm 2.
T
where D̂ = D̃ D̄ , d̂(t) = d̃ T (t) d̄ T (t) .
  
IV. SIMULATION EXAMPLES AND ANALYSIS
Here, the sufficient conditions for the existence of a non- To verify the effectiveness of the proposed fault estimation
linear augmentation fault observer with double perturbation method based on the augmented fault observer, a control

T
η(t) = ẽT (t) 1g̃T (t) d̃ T (t) ,


(Ã − L̃ C̃) T P + P(Ã − L̃ C̃) + α1 I + M̃

 
P PD̃
4= ∗ −In+s+q 0 .
∗ ∗ −αIl+2q)

VOLUME 13, 2025 15307

 Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

And
d(t) = 0.333 sin x3 (t),
fs (t) = 0.8(1 − e5−t ) + sin π t, 0 ≤ t ≤ 20.


fa (t) = 0.4 sin t cos(3t) + 3, 0 ≤ t ≤ 6.

Given H∞ performance index α = 0.2 and stability margin
ε = 0.3, solving the matrix inequality in Theorem 1, the
gain matrix of the undetermined matrix and augmented fault
diagnosis observer can be obtained as follows P, P and X , as
shown at the bottom of the next page.
And
 
−6.0417 −39.2917 −16.0509
   6.9721 6.9515 10.5599 


L 2.7400 40.5481 20.4687 
L̃ = P−1 X =

FIGURE 1. Sensor fault and sensor fault estimation curve of linear system. =−22.8161 −140.2178 −56.2634

R  
 8.7755 72.6284 28.0026 
176.4971 307.2981 182.3755
In a linear system, the initial condition of the augmen-
tation state estimation error with double faults may be set
as ẽ(0) = (−0.3, −0.1, 0.1, 0.2, −0.2, 0.3)T . The effect of
sensor fault estimation was tested against the real fault and
fault estimation curve.
As shown in Figs. 1 and 2, by solving the gain matrix
of the augmented fault observer, the designed observer can
effectively estimate the sensor and actuator faults contained in
the system. Simultaneously, owing to the quadratic constraint
on the gain matrix, the proposed fault estimation method
proposed is evidently better than existing methods.

B. SENSOR AND ACTUATOR FAULT ESTIMATION IN

NONLINEAR SYSTEMS
FIGURE 2. Actuator fault and actuator fault estimation curve of linear
system. For the nonlinear DC motor system in [35], the corresponding
matrix is as follows:
 
−0.4751 −0.0396
system with both sensor and actuator faults was simulated and A= ,
113.9706 −0.00952
analyzed. 
0.1277

B= ,
0
A. SENSOR AND ACTUATOR FAULT ESTIMATION FOR  
1 0
LINEAR SYSTEMS WITH DISTURBANCE C= ,
0 1
For the robotic arm system in [34], the state–space equation  
is as follows: 1
Fs = ,
     0
 ẋ1 (t) 0 1 0 0 x1 (t)  
 1
,


 ẋ2 (t) −48.6 −1.25
 48.6 0 x2 (t)
  Fa =


  =   0

 ẋ3 (t)  0 0 0 10 x3 (t)  
 0
,


 ẋ4 (t)
 1.95 0 −1.95 0 x4 (t) D=


       1

 0 0 0 
0.0075 sin x2 (t)

,

21.6 12.5 0 g[x(t), t] =
0
+  u(t) +  fa (t) +   d(t)
      
 
0 0 0





    d(t) = 0.003 sin 5t.

 0 0 −1
And

       

 y1 (t)
 1 0 0 0 x1 (t) 0 (
fs (t) = 5(1 − e4−t ) + sin π t, 0 ≤ t ≤ 30.


=   2  1 fs (t)
+

y (t) 0 1 0 0 x (t)
       

  2  


y3 (t) 0 0 1 1 x3 (t) 0 fa (t) = 2 sin(2t) cos t + 3, 0 ≤ t ≤ 50.

15308 VOLUME 13, 2025

Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

FIGURE 4. Actuator fault and actuator fault estimation curve of nonlinear

system.
FIGURE 3. Sensor fault and sensor fault estimation curve of nonlinear
system.
 
−1 L
L̃ = P X=
R
Given H∞ performance index α = 0.2 and stability margin  
34.3099 88.2970
ε = 0.2, using MATLAB to solve the matrix inequality  15.4061 115.9381 
.
in Theorem 2, the gain matrix of the undetermined matrix =
 48.5190 −67.5565
and augmented fault diagnosis observer can be obtained as
657.3800 162.5867
follows:
  In a nonlinear system, the initial condition of the augmen-
34.6159 −9.9086 15.7327 −0.9651 tation state estimation error with double faults is ẽ(0) =
−9.9086 38.3060 9.9086 0.0000 
, (−0.3, −0.1, 0.1, 0.2)T . The true fault values and fault esti-
P=  15.7327 9.9086 34.6156 −9.9651 mation curves of sensors and actuators are depicted in Figs. 3
−0.9651 0.0000 −9.9651 0.1231 and 4.
 
1.1639 0.6879 Figs. 3 and 4 show the estimation curves of sensor faults
0.7309 2.8968
, and actuator faults for nonlinear systems. Evidently, the pro-
X = 103 

1.7375 0.0425 posed method can be used to effectively estimate the faults of
0.0010 0.0000 nonlinear systems with disturbances, and the effectiveness of

 
208.4742 16.6826 13.7482 2.6627 100.6747 −0.1245
 16.6826 41.7914 32.0571 22.5003 40.8050 −2.3470
 
 13.7482 32.0571 82.8029 34.5125 35.5525 −0.1103
P= 2.6627

 22.5003 34.5125 25.9586 28.9563 −0.3942

100.6747 40.8050 35.5525 28.9563 93.5952 −1.4725
−0.1245 −2.3470 −0.1103 −0.3942 −1.4725 1.0255
 
208.4742 16.6826 13.7482 2.6627 100.6747 −0.1245
 16.6826 41.7914 32.0571 22.5003 40.8050 −2.3470
 
 13.7482 32.0571 82.8029 34.5125 35.5525 −0.1103
P= 2.6627

 22.5003 34.5125 25.9586 28.9563 −0.3942

100.6747 40.8050 35.5525 28.9563 93.5952 −1.4725
−0.1245 −2.3470 −0.1103 −0.3942 −1.4725 1.0255
 
−304.7990 −617.6506 −241.9741
−291.0975
 22.3209 278.3871 
−127.5844 749.1253 846.3836 
X = 
−172.3904
 −206.7442 179.7287 
−325.5519 54.5321 265.8755 
161.1556 247.5644 142.9266

VOLUME 13, 2025 15309

 Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

the fault estimation method for nonlinear systems with dis- [3] J. Marzat, H. Piet-Lahanier, F. Damongeot, and E. Walter, ‘‘Model-based
turbances based on the augmented fault observer is verified. fault diagnosis for aerospace systems: A survey,’’ Proc. Inst. Mech. Eng.,
G, J. Aerosp. Eng., vol. 226, no. 10, pp. 1329–1360, Oct. 2012, doi:
10.1177/0954410011421717.
V. CONCLUSION [4] S. M. M. Alavi and M. Saif, ‘‘Fault detection in nonlinear stable systems
A robust estimation method based on augmented fault over lossy networks,’’ IEEE Trans. Control Syst. Technol., vol. 21, no. 6,
pp. 2129–2142, Nov. 2013, doi: 10.1109/TCST.2012.2236839.
observer was developed for a class of nonlinear systems [5] A. Martínez-Sibaja, A. Alvarado-Lassman, R. Posada-Gómez,
with sensor and actuator faults. Under the condition of exter- J. M. Méndez-Contreras, B. E. G. Sánchez, and O. Sandoval-González,
nal disturbance, an adaptive augmentation fault observer ‘‘Dedicated observer scheme for fault diagnosis and isolation in
instruments of an anaerobic reactor,’’ Proc. Technol., vol. 7, pp. 173–180,
was designed, and sufficient conditions for the existence Jan. 2013, doi: 10.1016/J.PROTCY.2013.04.022.
of the observer were given in the form of linear matrix [6] T. A. Najafabadi, F. R. Salmasi, and P. Jabehdar-Maralani, ‘‘Detection and
inequalities, which provides a basis for realizing robust fault isolation of Speed-, DC-link Voltage-, and current-sensor faults based on an
adaptive observer in induction-motor drives,’’ IEEE Trans. Ind. Electron.,
estimation of nonlinear systems. By applying quadratic con- vol. 58, no. 5, pp. 1662–1672, May 2011, doi: 10.1109/TIE.2010.2055775.
straints on the observer gain matrix, the convergence rate of [7] X. Yao, G. Tao, R. Qi, and B. Jiang, ‘‘An adaptive actuator fail-
observation errors of the augmentation fault observer was ure compensation scheme for a class of nonlinear MIMO systems,’’
J. Franklin Inst., vol. 350, no. 9, pp. 2423–2441, Nov. 2013, doi:
further improved. Thus, the performance of fault estimation 10.1016/J.JFRANKLIN.2013.01.016.
was improved. Simultaneously, considering the disturbances [8] K. Zhang, M. Staroswiecki, and B. Jiang, ‘‘Static output feedback
existing in the system, the influence of disturbance on fault based fault accommodation design for continuous-time dynamic sys-
estimation was weakened by setting the H∞ performance tems,’’ Int. J. Control, vol. 84, no. 2, pp. 412–423, Feb. 2011, doi:
10.1080/00207179.2011.557396.
index. [9] S. Yan-Xiu, ‘‘Observer-based sensor fault estimation method for nonlinear
Specifically, the fault estimation methods of linear sys- systems,’’ Firepower Command Control., vol. 45, no. 12, pp. 43–47, 2020.
tems, nonlinear systems, nonlinear systems with input and [10] W. Wang, Y. Geng, S. Huang, Y. Yu, and Y. Yang, ‘‘Sensor fault
estimation based on ellipsoid bundles for dynamic point-the-bit rotary
sensor disturbances were studied. The proposed methods are steerable systems,’’ Measurement, vol. 235, Aug. 2024, Art. no. 114973,
progressive and enrich the theory of simultaneous estimation doi: 10.1016/J.MEASUREMENT.2024.114973.
of sensor and actuator faults. Finally, the proposed fault esti- [11] S. Yan-Xiu and L. Hong, ‘‘A Fault observer design method for nonlinear
discrete systems with perturbations,’’ Electr. Light Control., vol. 31, no. 10,
mation method was tested, and the simulation results show pp. 91–95, 2024.
that the proposed method has a good estimation effect, which [12] I. Hosseini, A. Karami, and M. Fiacchini, ‘‘Reset unknown input observer
verifies its effectiveness and feasibility. In future work, based for robust actuator and sensor fault estimation based on a descriptor system
approach,’’ Eur. J. Control, vol. 77, May 2024, Art. no. 100995.
on the results of this study, the fault estimation methods of
[13] J. Chen, J. Li, Y. Zhang, and Y. Cao, ‘‘Observer-based fault tolerant
nonlinear interconnected large-scale systems with perturba- control for a class of nonlinear multi-agent systems with sensor faults,’’
tions can be studied to further enrich the theory of system Int. J. Robust Nonlinear Control, vol. 34, no. 6, pp. 4132–4156, Jan. 2024.
fault estimation and diagnosis. [14] Y. Sun and H. Li, ‘‘An estimation method for sensor faults based on
observer in interconnected systems,’’ PLoS ONE, vol. 19, no. 3, Mar. 2024,
Art. no. e0296848, doi: 10.1371/JOURNAL.PONE.0296848.
AUTHOR CONTRIBUTIONS [15] S. Venkateswaran and C. Kravaris, ‘‘Linear unknown input observers
Conceptualization: Yanxiu Sun; methodology: Yanxiu Sun; for sensor fault estimation in nonlinear systems,’’ IFAC-PapersOnLine,
vol. 56, no. 1, pp. 61–66, 2023.
formal analysis: Yanxiu Sun, Ying Du, and Yuping Zhang; [16] S. Venkateswaran and C. Kravaris, ‘‘Design of linear unknown input
writing: Yanxiu Sun. All authors have read and agreed to the observers for sensor fault estimation in nonlinear systems,’’ Automatica,
published version of the manuscript. vol. 155, Sep. 2023, Art. no. 111152.
[17] Y. Ju, G. Wei, and W. Li, ‘‘Distributed fault estimation over sensor net-
works: A bit rate allocation scheme,’’ Inf. Sci., vol. 633, pp. 597–612,
INSTITUTIONAL REVIEW BOARD STATEMENT Jul. 2023, doi: 10.1016/J.INS.2023.03.045.
Not applicable. [18] Y. Peng, ‘‘Observer-based robust fault diagnosis and fault-tolerant control
of aircraft,’’ Missile Space Launch Technol., vol. 5, no. 2, pp. 74–78, 2022.
[19] S. Yanxiu, ‘‘Actuator fault estimation for nonlinear generalized systems
INFORMED CONSENT STATEMENT with interference terms,’’ Firepower Command Control., vol. 47, no. 6,
Not applicable. pp. 93–96, 2022.
[20] P. Gasga, M. Bernal, S. Gómez-Peñate, F. R. López-Estrada,
G. Valencia-Palomo, and I. Santos-Ruiz, ‘‘Actuator multiplicative
DATA AVAILABILITY STATEMENT and additive simultaneous faults estimation using a qLPV proportional
Not applicable. integral unknown input observer,’’ Int. J. Adapt. Control Signal Process.,
vol. 38, no. 8, pp. 2930–2950, Aug. 2024, doi: 10.1002/ACS.3859.
[21] R. El Abbadi and H. Jamouli, ‘‘Actuator fault estimation of a networked
CONFLICTS OF INTEREST control system with packet loss,’’ J. Control, Autom. Electr. Syst., vol. 32,
The authors declare no conflicts of interest. no. 6, pp. 1553–1562, Dec. 2021.
[22] J. Guzman, F.-R. López-Estrada, V. Estrada-Manzo, and
G. Valencia-Palomo, ‘‘Actuator fault estimation based on a
REFERENCES proportional-integral observer with nonquadratic Lyapunov functions,’’
[1] M. Witczak, Fault Diagnosis and Fault-Tolerant Control Strategies for Int. J. Syst. Sci., vol. 52, no. 9, pp. 1938–1951, Jul. 2021.
Non-Linear Systems. Cham, Switzerland: Springer, 2014. [23] V. Borja-Jaimes, A. Coronel-Escamilla, R. F. Escobar-Jiménez,
[2] S. Huang, K. K. Tan, and T. H. Lee, ‘‘Fault diagnosis and fault-tolerant M. Adam-Medina, G. V. Guerrero-Ramírez, E. M. Sánchez-Coronado,
control in linear drives using the Kalman filter,’’ IEEE Trans. and J. García-Morales, ‘‘Fractional-order sliding mode observer for
Ind. Electron., vol. 59, no. 11, pp. 4285–4292, Nov. 2012, doi: actuator fault estimation in a quadrotor UAV,’’ Mathematics, vol. 12, no. 8,
10.1109/TIE.2012.2185011. p. 1247, Apr. 2024, doi: 10.3390/MATH12081247.

15310 VOLUME 13, 2025

Y. Sun et al.: Fault Estimation Method for Nonlinear Systems With Perturbations

[24] A. Sanjuan, F. Nejjari, and R. Sarrate, ‘‘Actuator fault estimation YANXIU SUN received the M.S. degree in
using optimization-based learning techniques for linear parameter applied mathematics from Liaoning University,
varying systems with unreliable scheduling parameters,’’ Eng. Shenyang, China, in 2010. He is currently a
Appl. Artif. Intell., vol. 127, Jan. 2024, Art. no. 107247, doi: Teacher with Shenyang Institute of Technology,
10.1016/J.ENGAPPAI.2023.107247. Fushun, China. His main research interests include
[25] J. Guo, H. Liu, J. Hu, and B. Song, ‘‘Joint state and actuator fault system observer design, robust control of non-
estimation for networked systems under improved accumulation-based linear systems, and fault estimation of nonlinear
event-triggered mechanism,’’ ISA Trans., vol. 127, pp. 60–67, Aug. 2022,
systems. He presided more than the scientific
doi: 10.1016/J.ISATRA.2022.04.011.
research funding project of Liaoning Provincial
[26] J. Martinez, B. Aguiar, V. Estrada-Manzo, and M. Bernal, ‘‘Actuator
fault detection for discrete-time descriptor systems via a convex unknown Department of Education, in 2020, presided more
input observer with unknown scheduling variables,’’ Math. Problems Eng., than the teaching reform project of Shenyang Institute of Technology,
vol. 2021, pp. 1–16, Mar. 2021. in 2021, participated in the scientific research project of Liaoning Provincial
[27] Y. Wu, D. Du, B. Liu, and Z. Mao, ‘‘Actuator fault estimation Department of Education, in 2022, and presided more than the basic appli-
for two-stage chemical reactor system based on delta operator cation research project of Liaoning Science and Technology Department,
approach,’’ J. Process Control, vol. 107, pp. 37–46, Nov. 2021, doi: in 2023. He will be hosting the Youth Backbone Fund project of Shenyang
10.1016/J.JPROCONT.2021.09.013. Institute of Technology.
[28] I. I. Zetina-Rios, G. Osorio-Gordillo, R. A. Vargas-Méndez,
G. Madrigal-Espinosa, and C. Astorga-Zaragoza, ‘‘Actuator fault
estimation based on generalized learning observer for quasi-linear
parameter varying systems,’’ Int. J. Adapt. Control Signal Process.,
vol. 35, no. 5, pp. 828–845, May 2021, doi: 10.1002/ACS.3229.
[29] S. Gao, G. Ma, Y. Guo, and W. Zhang, ‘‘Fast actuator and sensor fault esti- YING DU received the M.S. degree in applied
mation based on adaptive unknown input observer,’’ ISA Trans., vol. 129, mathematics from Northeastern University,
pp. 305–323, Oct. 2022, doi: 10.1016/J.ISATRA.2022.01.019. Shenyang, China, in 2010. She is currently a
[30] H. Zhenggao, ‘‘Fault diagnosis of continuous system based on adaptive Teacher with Shenyang Institute of Technology,
singular observer,’’ Comput. Meas. Control, vol. 23, no. 7, pp. 2298–2301, Fushun, China. Her research interests include sys-
2015, doi: 10.16526/J.CNKI.11-4762/TP.2015.07.017. tem observer design, robust control theory, and
[31] Y. Zhu, N. Xu, F. Wu, X. Chen, and D. Zhou, ‘‘Fault estimation for a applications.
class of Markov jump piecewise-affine systems: Current feedback based
iterative learning approach,’’ IEEE/CAA J. Autom. Sinica, vol. 11, no. 2,
pp. 418–429, Feb. 2024, doi: 10.1109/JAS.2023.123990.
[32] X. Shen, J. Liu, Z. Liu, Y. Gao, J. I. Leon, S. Vazquez, L. Wu, and
L. G. Franquelo, ‘‘Sliding-mode control of neutral-point-clamped power
converters with gain adaptation,’’ IEEE Trans. Power Electron., vol. 39,
no. 8, pp. 9189–9201, Aug. 2024, doi: 10.1109/TPEL.2024.3386800.
[33] Z. Liu, X. Lin, Y. Gao, R. Xu, J. Wang, Y. Wang, and J. Liu, ‘‘Fixed-
time sliding mode control for DC/DC buck converters with mismatched YUPING ZHANG received the M.S. degree in
uncertainties,’’ IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 70, no. 1, applied mathematics from Harbin Institute of
pp. 472–480, Jan. 2023, doi: 10.1109/TCSI.2022.3217612. Technology, Harbin, China. She is currently a
[34] H. Zhenggao, ‘‘Fault diagnosis of nonlinear dynamic systems based Teacher with Liaoning Petrochemical University,
on adaptive unknown input observer,’’ Control Decis., vol. 31, no. 5, Fushun, China. Her research interests include deep
pp. 901–906, 2016, doi: 10.13195/J.KZYJC.2015.0224. learning and image processing.
[35] H. Kheloufi, A. Zemouche, F. Bedouhene, and H. Souley-Ali, ‘‘A robust
H∞ observer-based stabilization method for systems with uncertain param-
eters and Lipschitz nonlinearities,’’ Int. J. Robust Nonlinear Control,
vol. 26, no. 9, pp. 1962–1979, Jun. 2016, doi: 10.1002/RNC.3391.

VOLUME 13, 2025 15311