Hindawi Publishing Corporation

Journal of Applied Mathematics

Volume 2013, Article ID 829715, 12 pages
http://dx.doi.org/10.1155/2013/829715

Research Article
Robust Fault Detection of Linear Uncertain Time-Delay Systems
Using Unknown Input Observers

Saeed Ahmadizadeh,1 Jafar Zarei,2 and Hamid Reza Karimi3

1
Electrical Engineering Department, Iran University of Science and Technology, Narmak, Tehran 16846, Iran
2
Department of Electrical and Electronic Engineering, Shiraz University of Technology, Shiraz 71557-13876, Iran
3
Department of Engineering, Faculty of Engineering and Science, University of Agder, 4898 Grimstad, Norway

Correspondence should be addressed to Hamid Reza Karimi; hamid.r.karimi@uia.no

Received 18 July 2013; Accepted 13 September 2013

Academic Editor: Baocang Ding

Copyright © 2013 Saeed Ahmadizadeh et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

This paper deals with the problem of fault detection for linear uncertain time-delay systems. The proposed method for Luenberger
observers is developed for unknown input observers (UIOs), and a novel procedure for the design of residual based on UIOs
is presented. The design procedure is carried out based on the 𝐻∞ model matching approach which minimizes the difference
between generated residuals by the optimal observer and those by the designed observer in the presence of uncertainties. The
optimal observer is designed for the ideal system and works so that the fault effect is maximized while the exogenous disturbances
and noise effects are minimized. This observer can give disturbance decoupling in the presence of noise and uncertainties for linear
uncertain time-delay systems. The developed method is applied to a numerical example, and the simulation results show that the
proposed approach is able to detect faults reliably in the presence of modeling errors, disturbances, and noise.

1. Introduction However, model based approaches are based on some

idealized assumptions, one of which is that the mathematical
Fault detection and isolation (FDI) is an essential and model of the plant is a faithful replica of the plant dynamics
challenging problem in many industrial applications. Among [5]. As the mathematical model of a plant hardly represents its
the various reported methods, much attention has been paid complete behavior, due to the existence of model uncertainty,
to model based approaches in the field of control engineering noise, and unknown disturbances, it is essential to design a
in recent years. For example, fault detection problem for fault diagnosis system to take these effects into consideration.
discrete-time Markov jump systems and switched systems Motivated by the abovementioned issues, a robust fault detec-
is investigated in [1, 2], respectively. The problem of fault tion scheme is exploited to design fault detection systems
reconstruction for a class of descriptor linear systems using so that high sensitivity to faults as well as low sensitivity
sliding mode observers is presented in [3]. The sliding mode to uncertainties and perturbation can be obtained. Opti-
observers have been designed such that the actuator fault mization techniques are widely used to solve this problem.
can be reconstructed using output measurements. The data- One of the commonly used approaches to design such FDI
driven scheme for FDI is presented in [4] which exploits an scheme is representing the design procedure by 𝐻∞ and 𝐻−
adaptive residual generator and a bank of isolation observers. performance indexes. The main advantage of this approach
The designed scheme obtains observer parameters without is that it can be solved by linear matrix inequality (LMI) [6].
identification of complete process model. In [7, 8], a two-step FDI design methodology is presented.
2 Journal of Applied Mathematics

In this methodology, the optimal fault detection filter (FDF) The notation 𝑃 > 0 means that 𝑃 is real symmet-
has firstly been designed, neglecting the existence of model ric and positive definite; the symbol ∗ denotes the ele-
uncertainty. Next, the FDF, which is used as residual gener- ments below the main diagonal of a symmetric block
ator, has been obtained via 𝐻∞ model matching technique. matrix.
The same approach is used in [9]; however, different dynamics
is considered for the FDF. 2. Problem Statement
Time delay is an inherent characteristic of many indus-
trial systems; therefore, robust FDI problem for LTI systems Many different industrial systems such as mechanical, elec-
with time delay received great attention over recent decades, trical, meteorological, chemical, economic, and biological
and numerous articles have been presented. One approach systems include time delay. In many studies linearized model
is to solve the formulated design procedure using the eigen- of these systems around point of operation is considered.
structure assignment approach in which the residual signal However, there are always some discrepancies between the
is thoroughly decoupled from delay-free unknown input. real dynamics of the system and linearized model. These
Then the effect of the unknown input is minimized using differences arise from systems uncertainty, as a consequence
𝐻∞ norm [10]. In the presence of uncertainty, the same of neglecting dynamics, and changes in system parameters.
approach as delay-free case can be employed to obtain a Therefore, the following linear uncertain system with additive
robust FDI system. Indeed, solving the 𝐻∞ model matching disturbances and time delay is considered to represent the
problem results in achieving an FDF which acts the same described model:
as the optimal one [11–13]. Although the same approach is
considered in these references, solving procedures are com- 𝑥̇ (𝑡) = (𝐴 + Δ𝐴 (𝑡)) 𝑥 (𝑡) + (𝐴 𝑑 + Δ𝐴 𝑑 (𝑡)) 𝑥 (𝑡 − 𝜏)
pletely different owing to the difference between the dynamic
+ (𝐵 + Δ𝐵 (𝑡)) 𝑢 (𝑡) + 𝐸𝑑 (𝑡) + 𝐹𝑥 𝑓 (𝑡) + 𝑅𝑛 (𝑡) , (1)
of the filter and system. In [14], the problem of robust
FDF design for the class of linear systems has been inves- 𝑦 (𝑡) = 𝐶𝑥 (𝑡) + 𝐹𝑦 𝑓 (𝑡) + 𝐷 𝑛 (𝑡) ,
tigated. The system is subjected to mixed neutral and dis-
crete time-varying delays and some nonlinear perturbations. where 𝑥(𝑡) ∈ R𝑛 is the state vector, 𝑦(𝑡) ∈ R𝑝 is the
The Luenberger type observer has been utilized to design output vector, 𝑢(𝑡) ∈ R𝑞 is the input vector, 𝑑(𝑡) ∈ R𝑚
FDF such that the residual signals effectively show fault is an unknown scalar function representing the disturbance
occurrence. that belongs to 𝐿𝑚2 (0, ∞), 𝑓(𝑡) ∈ R𝑓 denotes the faults, and
Another approach commonly used to robust FDI scheme 𝑛(𝑡) ∈ R𝑟 represents the noise. Note that Δ𝐴(𝑡), Δ𝐵(𝑡), and
is to employ the unknown input observers (UIO), in which Δ𝐴 𝑑 (𝑡) are the norm bounded time-varying uncertainties of
the residual is designed to be sensitive to faults but insensitive the matrices 𝐴, 𝐵, and 𝐴 𝑑 , respectively, and 𝜏 ≥ 0 is a constant
to unknown disturbances. Although the UIO has been widely delay. It is assumed that the characteristics of uncertainty
used in estimation problems in both time delay and delay- matrices belong to
free systems [15–18], there are few references that handle the
problem of designing robust FDI [19, 20]. In [20], a design Ω1 = {Δ𝐴 (𝑡) | Δ𝐴 (𝑡) = 𝑀1 Σ1 (𝑡) 𝑁1 , Σ𝑇1 (𝑡) Σ1 (𝑡) ≤ 𝐼} ,
procedure has been proposed for delay-free systems so that
perturbations and exogenous signals have less effect on the Ω2 = {Δ𝐵 (𝑡) | Δ𝐵 (𝑡) = 𝑀2 Σ2 (𝑡) 𝑁2 , Σ𝑇2 (𝑡) Σ2 (𝑡) ≤ 𝐼} ,
residual signal and the fault has a detectable effect on the
residual; however, the problem has not been presented in 𝐻∞ Ω3 = {Δ𝐴 𝑑 (𝑡) | Δ𝐴 𝑑 (𝑡) = 𝑀3 Σ3 (𝑡) 𝑁3 , Σ𝑇3 (𝑡) Σ3 (𝑡) ≤ 𝐼} ,
model matching technique. Motivated by this consideration, (2)
a robust FDF design using UIO for uncertain systems with
time delay is presented and solved using 𝐻∞ model matching where 𝑀𝑖 and 𝑁𝑖 are predefined matrices. It is supposed that
approach. In contrast to our previous work [18], we are all over the paper the dimensions of matrices are compatible
concerned to design a robust FDF for the case in which if they are not explicitly mentioned.
the dynamic characteristic of fault signal is known. For The dynamic characteristic of fault signal can be
this purpose, a two-step design procedure is developed. In described by [21]
the first step, the optimal fault detection based on UIO
is designed for the system without uncertainty. Next, the 𝜃̇ (𝑡) = 𝐴 𝜃 𝜃 (𝑡) , 𝑡 ≥ 𝑡𝑓 ,
UIO-based fault detection filter is approached to optimal
one in the sense of 𝐻∞ norm. It is demonstrated through 𝜃 (𝑡) = 0, 𝑡 ∈ [0, 𝑡𝑓 ] ,
simulation that the presented fault detection observer is (3)
robust against uncertainty and sensitive enough to the 𝜃 (𝑡𝑓 ) = 𝜃0 ,
faults.
𝑓 (𝑡) = 𝐹𝜃 𝜃 (𝑡) ,
Notation. The notations used throughout the paper are
fairly standard. I and 0 represent identity matrix and zero where 𝑡𝑓 is the time when a fault occurs and 𝐴 𝜃 and 𝐹𝜃
matrix; the superscript “𝑇” stands for matrix transposition. are known matrices with appropriate dimensions. The initial
‖ ⋅ ‖ refers to the Euclidean vector norm or the induced time, 𝑡𝑓 , and initial state, 𝜃0 , are supposed to be unknown.
matrix 2-norm. diag{⋅} represents a block diagonal matrix. The dynamic equation (3) represents any fault with known
Journal of Applied Mathematics 3

dynamic characteristic and unknown amplitude and phase Condition 3:

[21].
Unknown input observer for the class of time-delay 𝐺 = (𝐼 − 𝐿 1 𝐶) 𝐴 𝑑 − 𝐿 3 𝐶. (9)
system (1) is considered as [22]
Condition 4:
𝑧̇ (𝑡) = 𝐹𝑧 + 𝐺𝑧 (𝑡 − 𝜏) + 𝐻𝑢 (𝑡) + 𝐾1𝑧 𝑦 (𝑡) + 𝐾2𝑧 𝑦 (𝑡 − 𝜏) ,
𝐻 = (𝐼 − 𝐿 1 𝐶) 𝐵. (10)
𝑥̂ (𝑡) = 𝑧 (𝑡) + 𝐿 1 𝑦 (𝑡) ,
(4) Condition 5:

̂ is the estimated state vector. The dynamic of 𝑥(𝑡)

where 𝑥(𝑡) ̂ (𝐼 − 𝐿 1 𝐶) 𝐸 = 0, (11)
is governed by
where 0 denotes a null matrix with compatible dimension.
̂̇ (𝑡) = 𝐹𝑥̂ + 𝐺𝑥̂ (𝑡 − 𝜏) + 𝐻𝑢 (𝑡) + 𝐿 1 𝑦̇ (𝑡)
𝑥 Using these relationships, and considering definitions in (12),
(5) the state estimation error dynamic (6) is transformed to (13):
+ 𝐿 2 𝑦 (𝑡) + 𝐿 3 𝑦 (𝑡 − 𝜏) ,
𝑇 = (𝐼 − 𝐿 1 𝐶) ,
where 𝐹, 𝐺, 𝐻, and 𝐿 1 are the observer matrices and 𝐿 2 =
𝐾1𝑧 − 𝐹𝐿 1 , 𝐿 3 = 𝐾2𝑧 − 𝐺𝐿 1 . 𝐹 = [(𝐼 − 𝐿 1 𝐶) 𝐹𝑥 𝐹𝜃 − 𝐿 2 𝐹𝑦 𝐹𝜃 − 𝐿 1 𝐹𝑦 𝐹𝜃 𝐴 𝜃 −𝐿 3 𝐹𝑦 𝐹𝜃 ] ,
The observer matrices will be designed such that the
disturbance and input are decoupled from the estimation 𝑅 = [−𝐿 1 𝐷 (𝐼 − 𝐿 1 𝐶) 𝑅 − 𝐿 2 𝐷 −𝐿 3 𝐷] ,
̂
error defined by 𝑒(𝑡) = 𝑥(𝑡) − 𝑥(𝑡). When UIO-based filter
defined by (4) is applied to the system described in (1), the 𝑇
state estimation error will be 𝜃 = [𝜃𝑇 (𝑡) 𝜃𝑇 (𝑡 − 𝜏)] ,

𝑒 ̇ (𝑡) = 𝐹𝑒 (𝑡) + 𝐺𝑒 (𝑡 − 𝜏) + ((𝐼 − 𝐿 1 𝐶) 𝐴 − 𝐿 2 𝐶 − 𝐹) 𝑥 (𝑡) 𝑇

𝑛 = [𝑛𝑇̇ (𝑡) 𝑛𝑇 (𝑡) 𝑛𝑇 (𝑡 − 𝜏)] ,
+ ((𝐼 − 𝐿 1 𝐶) 𝐴 𝑑 − 𝐿 3 𝐶 − 𝐺) 𝑥 (𝑡 − 𝜏) (12)

𝑒 ̇ (𝑡) = 𝐹𝑒 (𝑡) + 𝐺𝑒 (𝑡 − 𝜏) + 𝐹 𝜃 + 𝑅 𝑛
+ ((𝐼 − 𝐿 1 𝐶) 𝐵 − 𝐻) 𝑢 (𝑡)
+ 𝑇Δ𝐴 (𝑡) 𝑥 (𝑡) + 𝑇Δ𝐴 𝑑 (𝑡) 𝑥 (𝑡 − 𝜏) (13)
+ (𝐼 − 𝐿 1 𝐶) 𝐸𝑑 (𝑡)
+ 𝑇Δ𝐵 (𝑡) 𝑢 (𝑡) .
+ ((𝐼 − 𝐿 1 𝐶) 𝐹𝑥 𝐹𝜃 − 𝐿 2 𝐹𝑦 𝐹𝜃 − 𝐿 1 𝐹𝑦 𝐹𝜃 𝐴 𝜃 ) 𝜃 (𝑡)
In order to use the UIO for fault detection purposes,
− 𝐿 3 𝐹𝑦 𝐹𝜃 𝜃 (𝑡 − 𝜏) + ((𝑅 − 𝐿 1 𝐶𝑅) − 𝐿 2 𝐷) 𝑛 (𝑡) a residual signal should be defined. Difference between
measured output and estimated output is usually considered
as a residual signal. In current work, a more general form for
− 𝐿 3 𝐷𝑛 (𝑡 − 𝜏) − 𝐿 1 𝐷𝑛̇ (𝑡) residual reference signal is considered as follows:

+ (𝐼 − 𝐿 1 𝐶) Δ𝐴 (𝑡) 𝑥 (𝑡) + (𝐼 − 𝐿 1 𝐶) Δ𝐴 𝑑 (𝑡) 𝑥 (𝑡 − 𝜏) 𝑦̂ (𝑡) = 𝐶𝑥̂ (𝑡) ,

+ (𝐼 − 𝐿 1 𝐶) Δ𝐵 (𝑡) 𝑢 (𝑡) . 𝑟 (𝑡) = 𝑉1 (𝑦 (𝑡) − 𝑦̂ (𝑡)) + 𝑉2 (𝑦 (𝑡 − 𝜏) − 𝑦̂ (𝑡 − 𝜏))

(6)
= 𝑉1 𝐶𝑒 (𝑡) + 𝑉2 𝐶𝑒 (𝑡 − 𝜏) + 𝐾1 𝜃 (𝑡) + 𝐾2 𝑛 (𝑡) , (14)
In the absence of uncertainties and faults, it is shown that
the observer, defined by (4), is UIO for the predefined system
by (1) if the following conditions are satisfied [22]. 𝐾1 = [𝑉1 𝐹𝑦 𝐹𝜃 𝑉2 𝐹𝑦 𝐹𝜃 ] ,

Condition 1: 𝐾2 = [0 𝑉1 𝐷 𝑉2 𝐷] .

𝑒 ̇ (𝑡) = 𝐹𝑒 (𝑡) + 𝐺𝑒 (𝑡 − 𝜏) is asymptotically stable. (7) The goal of robust fault detection problem is to mini-
mize the performance index defined in (15) for all classes
of model uncertainties belonging to Ω𝑖 . In general, this
Condition 2: performance index is minimized using 𝐻∞ model match-
ing approach which minimizes the difference between the
𝐹 = (𝐼 − 𝐿 1 𝐶) 𝐴 − 𝐿 2 𝐶. (8) residual signal (𝑟(𝑡)) and reference residual signal (𝑟𝑓 (𝑡))
4 Journal of Applied Mathematics

in the presence of the worst case disturbance signals. This 0 0 𝑇Δ𝐴 𝑑 𝑇Δ𝐵 0 0 0
performance index has been minimized by the following ̃𝑑 = [0 0
Δ𝐴 0 ], Δ𝐵̃𝜔1 = [ 0 0 0 0] ,
steps. [0 0 Δ𝐴 𝑑 ] [ Δ𝐵 0 0 0]
Step 1. The ideal residual signal generator system has been ̃1 = [𝑉1 𝐶 −𝑉1∗ 𝐶 0] ,
𝐶 ̃2 = [𝑉2 𝐶 −𝑉2∗ 𝐶 0] ,
𝐶
designed for the system without uncertainty defined in
(16). The residual signal shows the maximum sensitivity to ̃ = [0 𝐾1 − 𝐾1∗ 0 𝐾2 − 𝐾2∗ ] ,
the fault signal while it has the minimum sensitivity to 𝐷
disturbance, noise, and unknown inputs.
𝐾3 = [𝐹𝑥 𝐹𝜃 0] , 𝐾4 = [0 𝑅 0] .
Step 2. The residual signal generator system has been de- (20)
signed such that the performance index (17) is minimized and
the overall system (19) is asymptotically stable: ̃ Δ𝐴
̃𝑑 , Δ𝐵̃𝜔
Furthermore, using (2) it is easy to see that Δ𝐴, 1
can be expressed by
󵄩󵄩 󵄩
󵄩󵄩𝐺𝑟[𝑑 𝑛] 󵄩󵄩󵄩∞
𝐽𝑟 = min 󵄩 󵄩󵄩 , (15) 𝑇𝑀1
(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 ) ∈Ω𝑖 󵄩 󵄩󵄩𝐺 󵄩󵄩 ̃= 𝑀 ̃1 = [ 0 ] Σ1 (𝑡) [0 0 𝑁1 ] ,
̃1 Σ1 (𝑡) 𝑁
󵄩󵄩 𝑟𝑓 󵄩󵄩∞ Δ𝐴
[ 𝑀1 ]
∗ ∗
𝑒𝑓̇ (𝑡) = 𝐹∗ 𝑒𝑓 (𝑡) + 𝐺∗ 𝑒𝑓 (𝑡 − 𝜏) + 𝐹 𝜃 (𝑡) + 𝑅 𝑛 (𝑡) ,
𝑇𝑀2
̃𝑑 = 𝑀
Δ𝐴 ̃2 = [ 0 ] Σ2 (𝑡) [0 0 𝑁2 ] ,
̃2 Σ2 (𝑡) 𝑁 (21)
𝑟𝑓 (𝑡) = 𝑉1∗ 𝐶𝑒𝑓 (𝑡) + 𝑉2∗ 𝐶𝑒𝑓 (𝑡 − 𝜏) + 𝐾1∗ 𝜃 (𝑡) + 𝐾2∗ 𝑛 (𝑡) ,
[ 𝑀2 ]
(16)

󵄩󵄩 󵄩󵄩 𝑇𝑀3
󵄩󵄩𝑟𝑒 󵄩󵄩2 Δ𝐵̃𝜔1 = 𝑀 ̃3 = [ 0 ] Σ3 (𝑡) [𝑁3 0 0 0] .
̃3 Σ3 (𝑡) 𝑁
𝐽𝑟𝑒 = sup < 𝛾, (17)
(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 )∈Ω𝑖 ‖𝜔‖2 [ 𝑀3 ]

Before developing theorems that are utilized in designed

where procedure, the following lemmas, which are useful to prove
the theorems, are introduced.

𝑟𝑒 (𝑡) = 𝑟 (𝑡) − 𝑟𝑓 (𝑡) , Lemma 1 (see [22]). Condition 5 is solvable if and only if the
following relation holds:
(18)
𝑇 𝑇
𝜔 = [𝑢 𝑇 𝑇 𝑇] ,
𝜃 𝑑 𝑛 rank (𝐶𝐸) = 𝑚, 𝑚 ≤ 𝑝. (22)

𝜁 ̇ (𝑡) = (𝐴
̃ + Δ𝐴)
̃ 𝜁 (𝑡) + (𝐴
̃𝑑 + Δ𝐴
̃𝑑 ) 𝜁 (𝑡 − 𝜏)
The general solution of condition 5 can be calculated by

+ (𝐵̃𝜔1 + Δ𝐵̃𝜔1 ) 𝜔 (𝑡) , (19) 𝐿 1 = 𝐸(𝐶𝐸)+ + 𝑌 [𝐼 − 𝐶𝐸(𝐶𝐸)+ ]

(23)
̃1 𝜁 (𝑡) + 𝐶
𝑟𝑒 (𝑡) = 𝐶 ̃2 𝜁 (𝑡 − 𝜏) + 𝐷𝜔
̃ (𝑡) , = Θ1 + 𝑌Θ2 ,

where 𝑌 is an arbitrary matrix with an appropriate dimension.

where
Lemma 2. Suppose that M, N, and Σ(𝑡) are compatible and
Σ𝑇 (𝑡)Σ(𝑡) ≤ 𝐼; then there exists a scalar 𝜀 > 0 such that the
𝑇 following equation holds:
𝜁 (𝑡) = [𝑒𝑇 (𝑡) 𝑒𝑓𝑇 (𝑡) 𝑥𝑇 (𝑡)] ,

𝐹 0 0 𝐺 0 0 𝑀Σ𝑁 + (𝑀Σ𝑁)𝑇 ≤ 𝜀𝑀𝑀𝑇 + 𝜀−1 𝑁𝑇 𝑁. (24)

̃ = [ 0 𝐹∗ 0 ] ,
𝐴 ̃𝑑 = [ 0 𝐺∗ 0 ] ,
𝐴
[ 0 0 𝐴] [ 0 0 𝐴 𝑑]
(a) Reference Model Selection (Step 1). Reference model selec-
tion is an important key to the design of robust fault detection
0 𝐹 0 𝑅 0 0 𝑇Δ𝐴
∗ ∗ filter for linear uncertain time-delay systems. To this end,
𝐵̃𝜔1 = [ 0 𝐹 0 𝑅 ] , ̃ = [0 0 0 ] ,
Δ𝐴 the analogues procedure as that for fault detection in [6]
[𝐵 𝐾3 𝐸 𝐾4 ] [0 0 Δ𝐴 ] is extended for delay systems considering the UIO as the
Journal of Applied Mathematics 5

fault detection filter. According to (16), the reference residual asymptotic stability of (16). They also provide the conditions
signal can be written as sum of two signals, 𝑟𝑓𝑛 (𝑡) and that increase the sensitivity of the residual signal from faults
𝑟𝑓𝜃 (𝑡). The reference model should be chosen such that the and decrease the sensitivity of residual signal from noise.
effect of exogenous signals on the reference residual signal is
minimized while the effect of fault signal is maximized. These Theorem 3. For given 𝛼 > 0, if there exist symmetric positive
two tasks are described mathematically by matrices 𝑃, 𝑄, 𝑉1∗ , 𝑉2∗ , Φ∗1 , Φ∗2 , and Φ∗3 such that the following
LMI holds:
󵄩󵄩 󵄩
󵄩󵄩𝑇 (𝑟𝑓𝑛 , 𝑛)󵄩󵄩󵄩 ≤ 𝛼,
󵄩 󵄩∞ ∗ 𝑇
(25) 𝑃𝐹∗ + 𝐹∗𝑇 𝑃 + 𝑄 𝑃𝐺∗ 𝑃𝑅 𝐶𝑇 𝑉1∗
󵄩󵄩 󵄩 [ 0 𝐶𝑇 𝑉2∗ ]
𝑇
󵄩󵄩𝑇 (𝑟 , 𝑓)󵄩󵄩󵄩 ≥ 𝛽, [ ∗ −𝑄 ] < 0, (26)
󵄩󵄩 𝑓𝑓 󵄩󵄩− [ 2 ∗𝑇 ]
∗ ∗ −𝛼 𝐼 𝐾2
[ ∗ ∗ ∗ −𝐼 ]
where 𝑇(⋅, ⋅) is the transfer function between two signals. The
following two theorems provide conditions which ensure the where

𝑃𝐹∗ = 𝑃 (𝐴 − Θ1 𝐶𝐴) − Φ∗1 (Θ2 𝐶𝐴) − Φ∗2 𝐶,

𝑃𝐺∗ = 𝑃 (𝐴 𝑑 − Θ1 𝐶𝐴 𝑑 ) − Φ∗1 (Θ2 𝐶𝐴 𝑑 ) − Φ∗3 𝐶,

(27)
∗
𝑃𝑅 = [−𝑃 (Θ1 𝐷) − Φ∗1 (Θ2 𝐷) 𝑃 (𝑅 − Θ1 𝐶𝑅) − Φ∗1 (Ψ2 𝐶𝑅) − Φ∗2 𝐷 −Φ∗3 𝐷] ,

𝐾2∗ = [0 𝑉1∗ 𝐷 𝑉2∗ 𝐷] ,

then the system (28) is asymptotically stable and Lyapunov-Krasovskii function which is defined as 𝑉(𝑡) =
𝑡
‖𝑇(𝑟𝑓𝑛 , 𝑛)‖∞ ≤ 𝛼. Furthermore, the UIO matrices are obtained 𝑇
𝑒𝑓𝑛 𝑇
(𝑡)𝑃𝑒𝑓𝑛 (𝑡) + ∫𝑡−𝜏 𝑒𝑓𝑛 (𝑠)𝑄𝑒𝑓𝑛 (𝑠)𝑑𝑠. Then we have
from conditions 2 to 5, and 𝑌∗ = 𝑃−1 Φ1 , 𝐿 2 = 𝑃−1 Φ2 , 𝐿 3 =
∞
𝑃−1 Φ3 , 𝑇
𝐽𝑟𝑓𝑛 = ∫ (𝑟𝑓𝑛 (𝑡) 𝑟𝑓𝑛 (𝑡) − 𝛼2 𝑛𝑇 (𝑡) 𝑛 (𝑡) + 𝑉̇ (𝑡)) 𝑑𝑡
0
(29)
∗
̇ (𝑡) = 𝐹∗ 𝑒𝑓𝑛 (𝑡) + 𝐺∗ 𝑒𝑓𝑛 (𝑡 − 𝜏) + 𝑅 𝑛 (𝑡) ,
𝑒𝑓𝑛 + 𝑉 (0) − 𝑉 (∞) .
(28)
Assume 𝑟𝑓𝑛 (𝑡) = 0 for 𝑡 ∈ [−𝜏, 0]. Since 𝑉(∞) > 0, it can
𝑟𝑓𝑛 (𝑡) = 𝑉1∗ 𝐶𝑒𝑓𝑛 (𝑡) + 𝑉2∗ 𝐶𝑒𝑓𝑛 (𝑡 − 𝜏) + 𝐾2∗ 𝑛 (𝑡) .
be concluded that
∞
𝑇
𝐽𝑟𝑓𝑛 ≤ ∫ (𝑟𝑓𝑛 (𝑡) 𝑟𝑓 𝑛 (𝑡) − 𝛼2 𝑛𝑇 (𝑡) 𝑛 (𝑡) + 𝑉̇ (𝑡)) 𝑑𝑡. (30)
Proof. Condition ‖𝑇(𝑟𝑓 𝑛 , 𝑛)‖∞ ≤ 𝛼 is equivalent to 𝐽𝑟𝑓𝑛 : 0
∞ 𝑇 2 𝑇
∫0 (𝑟𝑓𝑛 (𝑡)𝑟𝑓𝑛 (𝑡) − 𝛼 𝑛 (𝑡)𝑛(𝑡))𝑑𝑡 ≥ 0. Now Consider the Taking derivative from 𝑉(𝑡) and considering (28) yield

𝑇 ∗
𝑡 𝑒𝑓𝑛 (𝑡) 𝑃𝐹∗ + 𝐹∗𝑇 𝑃 + 𝑄 + 𝐶𝑇 𝑉1∗𝑇 𝑉1∗ 𝐶 𝑃𝐺∗ + 𝐶𝑇 𝑉1∗𝑇 𝑉2∗ 𝐶 𝑃𝑅 + 𝐶𝑇 𝑉1∗𝑇 𝐾2∗ 𝑒𝑓 𝑛 (𝑡)
[ ] [ ]
𝐽𝑟𝑓 𝑛 ≤∫ 𝑒𝑓𝑛 (𝑡 − 𝜏) [ ∗ 𝐶𝑇 𝑉2∗𝑇 𝑉2∗ 𝐶 − 𝑄 𝐶𝑇 𝑉2∗𝑇 𝐾2∗ ] [𝑒𝑓 𝑛 (𝑡 − 𝜏)] . (31)
𝑡−𝜏
[ 𝑛 (𝑡) ] [ ∗ ∗ −𝛼2 𝐼 + 𝐾2∗𝑇 𝐾2∗ ] [ 𝑛 (𝑡) ]
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
Ξ
6 Journal of Applied Mathematics

Hence, Ξ < 0 implies 𝐽𝑟𝑓 𝑛 < 0. Using the Schur Φ∗2 = 𝑃𝐿∗2 , and Φ∗3 = 𝑃𝐿∗3 . Using conditions 2, 3 and (23) it
complement theorem (26) is concluded from (31). Indeed, the can be shown that (27) makes the LMI feasible. It completes
inequality (26), without considering (27), includes nonlinear the proof.
terms of 𝑃𝑌∗ , 𝑃𝐿∗2 , and 𝑃𝐿∗3 which lead the LMI to be
infeasible. To overcome this problem, define Φ∗1 = 𝑃𝑌∗ , Theorem 4. For given 𝛽 > 0, if there exists symmetric positive
matrices 𝑃, 𝑄, 𝑉1∗ , 𝑉2∗ ,Φ∗1 , Φ∗2 , and Φ∗3 such that the following
LMI holds:

∗
𝑃𝐹∗ + 𝐹∗𝑇 𝑃 − 𝑄 + 2𝜑1 (𝑉1∗ , 𝑉1𝑐𝑛 ) −𝑃𝐺∗ −𝑃𝐹 𝐶𝑇 𝑉1∗𝑇
[ ∗ ∗ 𝑛
−𝑄 + 2𝜑2 (𝑉2 , 𝑉2𝑐 ) 0 𝐶𝑇 𝑉2∗𝑇 ]
[ ] < 0, (32)
[ ∗ ∗ 𝛽 𝐹𝜃 𝐹𝜃 + 2𝜑3 (𝐾1 , 𝐾1 ) 𝐾1∗𝑇 ]
2 𝑇 ∗ 𝑛

[ ∗ ∗ ∗ −𝐼 ]

where

𝑃𝐹∗ = 𝑃 (𝐴 − Θ1 𝐶𝐴) − Φ∗1 (Θ2 𝐶𝐴) − Φ∗2 𝐶,

𝑃𝐺∗ = 𝑃 (𝐴 𝑑 − Θ1 𝐶𝐴 𝑑 ) − Φ∗1 (Θ2 𝐶𝐴 𝑑 ) − Φ∗3 𝐶,

∗
(33)
𝑃𝐹 = [𝑃 (𝐹𝑥 𝐹𝜃 − Θ1 𝐶𝐹𝑥 𝐹𝜃 ) − Φ∗1 (Θ2 𝐶𝐹𝑥 𝐹𝜃 ) − Φ∗2 𝐹𝑦 𝐹𝜃 − 𝑃 (Ψ1 𝐹𝑦 𝐹𝜃 𝐴 𝜃 ) − Φ∗1 (Ψ2 𝐹𝑦 𝐹𝜃 𝐴 𝜃 ) −Φ∗3 𝐹𝑦 𝐹𝜃 ] ,
∗ ∗
𝐾1∗ = [𝑉1 𝐹𝑦 𝐹𝜃 𝑉2 𝐹𝑦 𝐹𝜃 ] ,
𝑛−1 𝑛−1 𝑛−1 𝑛−1
𝑉1𝑐𝑛 = 𝑉1∗ 𝐶, 𝑉2𝑐𝑛 = 𝑉2∗ 𝐶, 𝑛
𝐾11 = 𝑉1∗ 𝑛
𝐹𝑦 𝐹𝜃 , 𝐾12 = 𝑉2∗ 𝐹𝑦 𝐹𝜃 , 𝑓𝑜𝑟 𝑛 = 1, 2, . . . ,

𝐾1𝑛 = [𝐾11
𝑛 𝑛
𝐾12 ],
𝑇 𝑇 𝑇
𝜑1 (𝑉1∗ , 𝑉1𝑐𝑛 ) = (𝑉1𝑐𝑛 ) 𝑉1𝑐𝑛 − (𝑉1𝑐𝑛 ) 𝑉1∗ 𝐶 − 𝐶𝑇 𝑉1∗ 𝑉1𝑐𝑛 , (34)
𝑇 𝑇
𝜑2 (𝑉2∗ , 𝑉2𝑐𝑛 ) = (𝑉2𝑐𝑛 ) 𝑉2𝑐𝑛 − (𝑉2𝑐𝑛 ) 𝑉2∗ 𝐶 − 𝐶𝑇 𝑉2∗𝑇 𝑉2𝑐𝑛 ,
𝑇 𝑇
𝜑3 (𝐾1∗ , 𝐾1𝑛 ) = (𝐾1𝑛 ) 𝐾1𝑛 − (𝐾1𝑛 ) 𝑉1∗ 𝐹𝑦 𝐹𝜃 − 𝐹𝜃𝑇 𝐹𝑦𝑇 𝑉1∗𝑇 𝐾1𝑛 ,

𝑇 𝑇 𝑡
then the system (35) is asymptotically stable, and 𝑒𝑓𝜃 (𝑡)𝑃𝑒𝑓𝜃 (𝑡) + ∫𝑡−𝜏 𝑒𝑓𝜃 (𝑠)𝑄𝑒𝑓𝜃 (𝑠)𝑑𝑠. Then we have
‖𝑇(𝑟𝑓𝜃 , 𝑓)‖ ≥ 𝛽. Moreover, the UIO matrices are obtained
−
∞
from conditions 2 to 5, and 𝑌∗ = 𝑃−1 Φ1 , 𝐿 2 = 𝑃−1 Φ2 , 𝐿 3 = 𝑇
𝐽𝑟𝑓𝜃 = ∫ (𝑟𝑓𝜃 (𝑡) 𝑟𝑓𝜃 (𝑡) − 𝛽2 𝑓 (𝑡)
𝑇

𝑃−1 Φ3 , 0 (36)
̇ (𝑡) = 𝐹∗ 𝑒𝑓 𝜃 (𝑡) + 𝐺∗ 𝑒𝑓𝜃 (𝑡 − 𝜏) + 𝐹 𝜃 (𝑡) ,
𝑒𝑓𝜃
∗ ×𝑓 (𝑡) − 𝑉̇ (𝑡)) 𝑑𝑡 − 𝑉 (0) + 𝑉 (∞) .
(35)
Assume 𝑟𝑓𝜃 (𝑡) = 0 for 𝑡 ∈ [−𝜏, 0]. Since 𝑉(∞) > 0, we
𝑟𝑓𝜃 (𝑡) = 𝑉1∗ 𝐶𝑒𝑓𝜃 (𝑡) + 𝑉2∗ 𝐶𝑒𝑓𝜃 (𝑡 − 𝜏) + 𝐾2∗ 𝜃 (𝑡) . have
∞
Proof. Condition ‖𝑇(𝑟𝑓𝑓 , 𝑓)‖ ≥ 𝛽 is equivalent to 𝐽𝑟𝑓𝑓 : 𝑇
−
𝑇
𝐽𝑟𝑓 𝜃 ≥ ∫ (𝑟𝑓𝜃 (𝑡) 𝑟𝑓𝜃 (𝑡) − 𝛽2 𝑓 (𝑡) 𝑓 (𝑡) − 𝑉̇ (𝑡)) 𝑑𝑡. (37)
∞𝑇 𝑇 0
∫0 (𝑟𝑓𝜃 (𝑡)𝑟𝑓𝜃 (𝑡) − 𝛽2 𝑓 (𝑡)𝑓(𝑡))𝑑𝑡 ≥ 0. Now consider the
Lyapunov-Krasovskii function which is defined as 𝑉(𝑡) = Taking derivative from 𝑉(𝑡) and considering (35) yield

𝑇 𝑇 ∗
𝑡
𝑒𝑓𝜃 (𝑡) −𝑃𝐹∗ − 𝐹∗ 𝑃 − 𝑄 + 𝐶𝑇 𝑉1∗𝑇 𝑉1∗ 𝐶 −𝑃𝐺∗ + 𝐶𝑇 𝑉1∗𝑇 𝑉2∗ 𝐶 −𝑃𝐹 + 𝐶𝑇 𝑉1∗𝑇 𝐾1∗ 𝑒𝑓 𝜃 (𝑡)
[𝑒 (𝑡 − 𝜏)] [ 𝑇 ∗𝑇 ∗ 𝑇 ∗𝑇 ∗ ] [𝑒 (𝑡 − 𝜏)]
𝐽𝑟𝑓 𝜃 > ∫ [ 𝑓𝜃 ] [ ∗ 𝐶 𝑉2 𝑉2 𝐶 + 𝑄 𝐶 𝑉2 𝐾1 ] [ 𝑓𝜃 ] . (38)
𝑡−𝜏 2 𝑇 ∗𝑇 ∗
[ 𝜃 (𝑡) ] [ ∗ ∗ −𝛽 𝐹 𝐹
𝜃 𝜃 + 𝐾 1 𝐾
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
1][ 𝜃 (𝑡) ]
Ξ
Journal of Applied Mathematics 7

Hence, Ξ > 0 implies 𝐽𝑟𝑓𝜃 > 0. Ξ > 0 is equivalent to

𝑇 ∗
𝑃𝐹∗ + 𝐹∗ 𝑃 + 𝑄 − 𝐶𝑇 𝑉1∗𝑇 𝑉1∗ 𝐶 −𝑃𝐺∗ + 𝐶𝑇 𝑉1∗𝑇 𝑉2∗ 𝐶 −𝑃𝐹 + 𝐶𝑇 𝑉1∗𝑇 𝐾1∗
[ ]
[ ∗ −𝑄 − 𝐶𝑇 𝑉2∗𝑇 𝑉2∗ 𝐶 𝐶𝑇 𝑉2∗𝑇 𝐾1∗ ] < 0. (39)
2 𝑇 ∗𝑇 ∗
[ ∗ ∗ 𝛽 𝐹𝜃 𝐹𝜃 − 𝐾1 𝐾1 ]

Then, (39) can be written as

𝑇 ∗
𝑃𝐹∗ + 𝐹∗ 𝑃 + 𝑄 − 2𝐶𝑇 𝑉1∗𝑇 𝑉1∗ 𝐶 −𝑃𝐺∗ + 𝐶𝑇 𝑉1∗𝑇 𝑉2∗ 𝐶 −𝑃𝐹 + 𝐶𝑇 𝑉1∗𝑇 𝐾1∗ 𝑉1∗ 𝐶
[ ]
] + [𝐶 𝑉1 𝐶 𝑉2 𝐾1 ] 𝐼 [𝑉2 𝐶] < 0. (40)
𝑇 ∗𝑇 𝑇 ∗𝑇 ∗𝑇 ∗
[ ∗ −𝑄 − 2𝐶𝑇 𝑉2∗𝑇 𝑉2∗ 𝐶 𝐶𝑇 𝑉2∗𝑇 𝐾1∗
∗
[ ∗ ∗ 𝛽2 𝐹𝜃𝑇 𝐹𝜃 − 2𝐾1∗𝑇 𝐾1∗ ] [ 𝐾1 ]

Making use of Lemma 2 with 𝜀 = 1, we have developed for the proposed structure. The procedures of this
method are as follows.
(1) Choose appropriate values of 𝛼 and 𝛽.
𝑇 𝑇
−𝐶𝑇 𝑉1∗𝑇 𝑉1∗ 𝐶 ≤ (𝑉1𝑐𝑛 ) 𝑉1𝑐𝑛 − (𝑉1𝑐𝑛 ) 𝑉1∗ 𝐶 − 𝐶𝑇 𝑉1∗𝑇 𝑉1𝑐𝑛 , (2) Solve the LMI (26), and find a feasible solution for 𝑃,
𝑇 𝑇 𝑇 𝑄, 𝑉1∗ , 𝑉2∗ , Φ∗1 , Φ∗2 , and Φ∗3 matrices.
−𝐶 𝑉2∗𝑇 𝑉2∗ 𝐶 ≤ (𝑉2𝑐𝑛 ) 𝑉2𝑐𝑛 − (𝑉2𝑐𝑛 ) 𝑉2∗ 𝐶 −𝐶 𝑇
𝑉2∗𝑇 𝑉2𝑐𝑛 , (41) 𝑛−1
(3) Set 𝑉1𝑐𝑛 = 𝑉1∗𝑛−1 𝐶, 𝑉2𝑐𝑛 = 𝑉2∗𝑛−1 𝐶, 𝐾11
𝑛
= 𝑉1∗ 𝐹𝑦 𝐹𝜃 ,
𝑇 𝑇
−𝐾1∗𝑇 𝐾1∗ ≤ (𝐾1𝑛 ) 𝐾1𝑛 − (𝐾1𝑛 ) 𝑉1∗ 𝐹𝑦 𝐹𝜃 − 𝐹𝜃𝑇 𝐹𝑦𝑇 𝑉1∗𝑇 𝐾1𝑛 . 𝑛 ∗𝑛−1
and 𝐾12 = 𝑉2 𝐹𝑦 𝐹𝜃 . Then, solve (26) and (32) by
increasing 𝑛 to find a feasible solution for 𝑃, Q, 𝑉1∗ ,
𝑉2∗ , Φ∗1 ,Φ∗2 , and Φ∗3 .
Applying Schur complement to (40) and changing variables (4) Increase 𝛽 and decrease 𝛼 and go to step 2. Continue
𝜑1 (𝑉1∗ , 𝑉1𝑐𝑛 ), 𝜑2 (𝑉2∗ , 𝑉2𝑐𝑛 ), and 𝜑3 (𝐾1∗ , 𝐾1𝑛 ), the LMI (32) is this procedure until the feasible solution cannot be
obtained. To overcome the infeasibility of (33), the same found for LMIs (26) and (32).
variables as those selected in Theorem 3 are used. It completes
the proof. (b) Robust UIO Design (Step 2). As mentioned before, the
residual signal generator system is obtained by minimizing
Corollary 5. The system is asymptotically stable and satisfies (17). To this end, Theorem 8 is presented which guarantees
(16) if there exists symmetric positive matrices 𝑃, 𝑄, 𝑉1∗ , 𝑉2∗ , that the overall system (19) is asymptotically stable and
Φ∗1 , Φ∗2 , and Φ∗3 such that the LMIs (26) and (32) hold. performance index (17) is minimized. Before presenting
Theorem 8, the following theorem is presented which helps
Remark 6. It is desired to obtain a reference residual system prove Theorem 8.
which has maximum sensitivity to the fault as well as the Theorem 7. For a given 𝛾 > 0 and the system (37), if
minimum sensitivity to the exogenous signal. This aim can there exist symmetric positive matrices 𝑃, 𝑄 and constants
be formulated by performance index defined by inf 𝛼/𝛽. To 𝜀1 , 𝜀2 , 𝑎𝑛𝑑 𝜀3 such that the LMI (44) holds, then the system (37)
this end, an iterative optimization method presented in [6] is is asymptotically stable and ‖𝜐(𝑡)‖2 ≤ 𝛾‖𝑢(𝑡)‖2 :

̃ + Δ𝐴)
𝜒̇ (𝑡) = (𝐴 ̃ 𝜒 (𝑡) + (𝐴 ̃𝑑 ) 𝜒 (𝑡 − 𝜏) + (𝐵̃𝑢 + Δ𝐵̃𝑢 ) 𝑢 (𝑡) ,
̃𝑑 + Δ𝐴 (42)

̃1 𝜒 (𝑡) + 𝐶
𝜐 (𝑡) = 𝐶 ̃2 𝜒 (𝑡 − 𝜏) + 𝐷𝑢
̃ (𝑡) , (43)
̃+ 𝐴
𝑃𝐴 ̃ 𝑇𝑁
̃𝑇 𝑃 + 𝑄 + 𝜀−1 𝑁 ̃ 𝑃𝐴̃𝑑 𝑃𝐵̃𝑢 ̃𝑇 𝑃𝑀
𝐶 ̃1 𝑃𝑀̃2 𝑃𝑀 ̃3
1 1 1 1
[ ∗ −1 ̃ 𝑇 ̃
−𝑄 + 𝜀2 𝑁2 𝑁2 0 ̃𝑇
𝐶2 0 0 0 ]
[ ]
[ ̃ 𝑇𝑁
̃ ̃𝑇 ]
[ ∗ ∗ −𝛾2 𝐼 + 𝜀3−1 𝑁3 3 𝐷 0 0 0 ]
[ ]
[ ∗ ∗ ∗ −𝐼 0 0 0 ] < 0. (44)
[ −1 ]
[ ∗ ∗ ∗ ∗ −𝜀1 𝐼 0 0 ]
[ ]
[ ∗ ∗ ∗ ∗ ∗ −𝜀2−1 𝐼 0 ]
[ ∗ ∗ ∗ ∗ ∗ ∗ −𝜀3−1 𝐼]
8 Journal of Applied Mathematics

Proof. Define the following Lyapunov-Krasovskii function: Assume 𝜒(𝑡) = 0 for 𝑡 ∈ [−𝜏, 0]. Since 𝑉(∞) > 0, we have
𝑡
𝑉 (𝑡) = 𝜒𝑇 (𝑡) 𝑃𝜒 (𝑡) + ∫ 𝜒𝑇 (𝑠) 𝑄𝜒 (𝑠) 𝑑𝑠. (45)
𝑡−𝜏
∞
The performance index ‖𝜐(𝑡)‖2 ≤ 𝛾‖𝑢(𝑡)‖2 can be written 𝐽𝜐 ≤ ∫ (𝜐𝑇 (𝑡) 𝜐 (𝑡) − 𝛾2 𝑢𝑇 (𝑡) 𝑢 (𝑡) + 𝑉̇ (𝑡)) 𝑑𝑡. (47)
as 0
∞
𝐽𝜐 = ∫ (𝜐𝑇 (𝑡) 𝜐 (𝑡) − 𝛾2 𝑢𝑇 (𝑡) 𝑢 (𝑡) + 𝑉̇ (𝑡)) 𝑑𝑡
0 (46)
+ 𝑉 (0) − 𝑉 (∞) . Taking derivative from (45) and considering (37) yield

𝑇 𝑇
𝜒 (𝑡) 𝑃 (𝐴 ̃ + Δ𝐴) ̃ + (𝐴 ̃ + Δ𝐴) ̃ 𝑃+𝑄+𝐶 ̃𝑇 𝐶 ̃ ̃ ̃ ̃𝑇 ̃ ̃ ̃ ̃𝑇 ̃
∞
[ 1 1 𝑃 (𝐴 𝑑 + Δ𝐴 𝑑 ) + 𝐶1 𝐶2 𝑃 (𝐵𝑢 + Δ𝐵𝑢 ) + 𝐶1 𝐷]
𝐽𝑟𝑒 ≤ ∫ [ ]
𝜒 (𝑡 − 𝜏) [ ∗ −𝑄 + 𝐶 ̃ 𝐶𝑇 ̃ 0 ]
0 2 2
[ 𝑢 (𝑡) ] [ ∗ ∗ 2
−𝛾 𝐼 + 𝐷 𝐷 ̃ 𝑇̃
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
]
Ξ (48)
𝜒 (𝑡)
× [𝜒 (𝑡 − 𝜏)] 𝑑𝑡.
[ 𝑢 (𝑡) ]

Ξ < 0 implies 𝐽𝑟𝑒 < 0. The Ξ < 0 can be written as Considering (49), (50) and using Schur complement (48)
lead to (44). It completes the proof.

Theorem 8. For a given 𝛾 > 0, if there exist symmetric positive

̃+ 𝐴
𝑃𝐴 ̃𝑇 𝑃 + 𝑄 𝑃𝐴
̃𝑑 𝑃𝐵̃𝑢
matrices 𝑃1 , 𝑃2 , 𝑃3 , 𝑄1 , 𝑄2 , and 𝑄3 , matrices Φ1 , Φ2 , and Φ3 ,
[ ∗ −𝑄 0 ]
2
and constants 𝜀1 , 𝜀2 , and 𝜀3 such that LMI [𝑠𝑖𝑗 ]14×14 < 0 holds,
[ ∗ ∗ −𝛾 𝐼] then the overall system (19) is asymptotically stable, and 𝐽𝑟𝑒 < 𝛾.
̃𝑇 The observer matrices are calculated by considering (4), (5),
𝐶
[
1
̃𝑇 ] 𝐼 [𝐶 ̃2 𝐷]
̃1 𝐶 ̃ and conditions 2 to 5, and 𝑌 = 𝑃1−1 Φ1 , 𝐿 2 = 𝑃1−1 Φ2 , and 𝐿 3 =
+ 𝐶 (49)
̃
2
𝑇 𝑃1−1 Φ3 . The LMI coefficients are defined as
[𝐷 ]
̃ + Δ𝐴
𝑃Δ𝐴 ̃𝑇 𝑃 𝑃Δ𝐴̃𝑑 𝑃Δ𝐵̃𝑢 𝑇
𝑠1,1 = 𝑃1 𝐹 + (𝑃1 𝐹) + 𝑄1 , 𝑠1,4 = 𝑃1 𝐺, 𝑠1,8 = 𝑃1 𝐹,
+[ ∗ 0 0 ] < 0.
[ ∗ ∗ 0 ] 𝑠1,10 = 𝑃1 𝑅, 𝑠1,11 = 𝐶𝑇 𝑉1𝑇 , 𝑠1,12 = 𝑃1 (𝑇𝑀1 ) ,

𝑠1,13 = 𝑃1 (𝑇𝑀2 ) , 𝑠1,14 = 𝑃1 (𝑇𝑀3 ) ,

Using Lemma 2, one can write the following inequality: 𝑇 ∗
𝑠2,2 = 𝑃2 𝐹∗ + (𝑃2 𝐹∗ ) + 𝑄2 , 𝑠2,5 = 𝑃2 𝐺∗ , 𝑠2,8 = 𝑃2 𝐹 ,
∗ 𝑇
𝑠2,10 = 𝑃2 𝑅 , 𝑠2,11 = −𝐶𝑇 𝑉1∗ ,
̃ + Δ𝐴
𝑃Δ𝐴 ̃𝑇 𝑃 𝑃Δ𝐴̃𝑑 𝑃Δ𝐵̃𝑢
[ ∗ 0 0 ] 𝑇
𝑠3,3 = 𝑃3 𝐴 + (𝑃3 𝐴) + 𝑄3 + 𝜀1−1 𝑁1𝑇 𝑁1 , 𝑠3,6 = 𝑃3 𝐴 𝑑 ,
[ ∗ ∗ 0 ]
𝑠3,7 = 𝑃3 𝐵, 𝑠3,8 = 𝑃3 𝐾3 , 𝑠3,9 = 𝑃3 𝐸, 𝑠3,10 = 𝑃3 𝐾4 ,
̃1 𝑃𝑀
𝑃𝑀 ̃2 𝑃𝑀
̃3 𝜀1 𝐼 0 0
≤[ 0 0 0 ] [ 0 𝜀2 𝐼 0 ] 𝑠3,12 = 𝑃3 (𝑀1 ) , 𝑠3,13 = 𝑃3 (𝑀2 ) , 𝑠3,14 = 𝑃3 (𝑀3 ) ,
[ 0 0 0 ] [ 0 0 𝜀3 𝐼]
𝑠4,4 = −𝑄1 , 𝑠4,11 = 𝐶𝑇 𝑉2𝑇 ,
𝑇
(50)
̃1 𝑃𝑀
𝑃𝑀 ̃2 𝑃𝑀
̃3 ̃𝑇 0 0
𝑁 𝑇
[ ] [
1
̃𝑇 0 ] 𝑠5,5 = −𝑄2 , 𝑠5,11 = 𝐶𝑇 𝑉2∗ , 𝑠6,6 = −𝑄3 + 𝜀2−1 𝑁2𝑇 𝑁2 ,
× 0 0 0 + 0 𝑁 2
[ 0 0 0 ] [ 0 0 𝑁 ̃ 𝑇] 𝑇
3 𝑠7,7 = −𝛾𝐼 + 𝜀3−1 𝑁3𝑇 𝑁3 , 𝑠8,8 = −𝛾𝐼, 𝑠8,11 = 𝐾1𝑇 − 𝐾1∗ ,
−1 ̃
𝜀1 𝐼 0 0 𝑁1 0 0 𝑠9,9 = −𝛾𝐼, 𝑠10,10, = −𝛾𝐼, 𝑠10,11 = 𝐾2𝑇 − 𝐾2∗ ,
𝑇
× [ 0 𝜀2 𝐼 0 ] [ 0 ̃2
𝑁 0 ].
[ 0 0 𝜀3 𝐼] [ 0 0 ̃
𝑁3 ] 𝑠11,11 = −𝐼, 𝑠12,12 = −𝜀1−1 𝐼,
Journal of Applied Mathematics 9

𝑠13,13 = −𝜀2−1 𝐼, 𝑠14,14 = −𝜀3−1 𝐼, where

otherwise 𝑠𝑖,𝑗 = 0,
(51)

𝑃1 𝑇 = 𝑃1 (𝐼 − Θ1 𝐶) − Φ1 (Θ2 𝐶) ,

𝑃1 𝐹 = 𝑃1 (𝐴 − Θ1 𝐶𝐴) − Φ1 (Θ2 𝐶𝐴) − Φ2 𝐶,

𝑃1 𝐺 = 𝑃1 (𝐴 𝑑 − Θ1 𝐶𝐴 𝑑 ) − Φ1 (Θ2 𝐶𝐴 𝑑 ) − Φ3 𝐶, (52)

𝑃1 𝐹 = [𝑃1 (𝐹𝑥 𝐹𝜃 − Θ1 𝐶𝐹𝑥 𝐹𝜃 ) − Φ1 (Θ2 𝐶𝐹𝑥 𝐹𝜃 ) − Φ2 𝐹𝑦 𝐹𝜃 − 𝑃1 (Θ1 𝐹𝑦 𝐹𝜃 𝐴 𝜃 ) − Φ1 (Θ2 𝐹𝑦 𝐹𝜃 𝐴 𝜃 ) −Φ3 𝐹𝑦 𝐹𝜃 ] ,

𝑃1 𝑅 = [𝑃1 (𝑅 − Θ1 𝐶𝑅) − Φ1 (Θ2 𝐶𝑅) − Φ2 𝐷 −Φ3 𝐷 −𝑃1 (Θ1 𝐷) − Φ1 (Θ2 𝐷)] .

Proof. In Theorem 7 assume that 𝑃 = diag(𝑃1 𝑃2 𝑃3 ) and 𝐽th,𝑛 can be computed offline, and under the assumption that
𝑄 = diag(𝑄1 𝑄2 𝑄3 ). Then, using system dynamic (19) it is 𝑛 ∈ 𝐿 2 we have sup(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 )∈Ω𝑖 ‖𝑟𝑛 (𝑡)‖2 = 𝑀𝑛 . Since the
straight forward to see that 𝑠𝑖𝑗 are the same as (37). Without signal 𝑢 is supposed to be known online, the value of 𝐽th,𝑢
considering (52), the inequality (51) includes nonlinear terms can be determined online by 𝐽th,𝑢 = 𝛾𝑢 ‖𝑢(𝑡)‖2 , where 𝛾𝑢 =
of 𝑃𝑌, 𝑃𝐿 2 , and 𝑃𝐿 3 which lead the LMI to be infeasible. To sup(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 )∈Ω𝑖 (‖𝑟𝑢 (𝑡)‖2 )/(‖𝑢(𝑡)‖2 ). 𝛾𝑢 can be computed by
overcome this problem, define Φ1 = 𝑃1 𝑌, Φ2 = 𝑃1 𝐿 2 , and Theorem 7. Therefore, the threshold value can be evaluated
Φ3 = 𝑃1 𝐿 3 . Using conditions 2, 3 and (23) it can be seen that by
(52) makes the obtained LMI feasible. It completes the proof.
𝐽th = 𝑀𝑛 + 𝛾𝑢 ‖𝑢(𝑡)‖2 . (55)
Since values of (53) and (54) increase by passing the time and,
Remark 9. It should be noted that the present work differs
consequently, need more memory in real application, one can
from [18] in the following perspectives.
use root-mean-square (RMS) norm of 𝑟(𝑡), defined in (56), to
(a) The results in [18] are obtained without considering detect the fault signals:
dynamic characteristic for fault; however, the current 𝑡2
results are achieved by considering dynamic char- ‖𝑟 (𝑡)‖𝑇2 = ∫ 𝑟𝑇 (𝑡) 𝑟 (𝑡) 𝑑𝑡, 𝑇 = 𝑡2 − 𝑡1 , (56)
𝑡1
acteristic that is modeled by (3). Hence, the design
procedure in [18] is not applicable for the current case. where T is designed parameter.
(b) The residual signal 𝑟(𝑡) is constructed based on
(14) which uses both estimation error and delay 3. Simulation Results
in estimation error; however, the residual signal in
The main objective of this section is to investigate the
[18] is constructed using estimation error. Since two
effectiveness of the designed UIO. To this end, a numerical
design parameters 𝑉1 and 𝑉2 appear in the LMIs, the
example is used and simulation results are presented. Con-
obtained LMIs are more flexible.
sider a system which is defined by (1) with the following
Remark 10. After designing the FDI system, residual eval- matrices:
uation methods and appropriate level of threshold should −3.8 1.5 −0.5 0.4 0.1 −0.2
be selected to take a decision about the occurrence of fault. 𝐴 = [ 0.5 −3 1 ] , 𝐴 𝑑 = [0.1 −0.8 0.2 ] ,
According to (13) and (14), the residual signal for fault-free [−0.3 0.7 −2.4] [0.7 −0.1 0.5 ]
system 𝑟0 (𝑡) satisfies the following equation:
0.1 0.6 −0.4
󵄩󵄩 0 󵄩󵄩 𝐵 = [ 0.2 ] , 𝐹𝑥 = [−0.5] , 𝐸 = [ 0.1 ] ,
󵄩󵄩𝑟 (𝑡)󵄩󵄩 = 󵄩󵄩󵄩󵄩𝑟𝑛 (𝑡) + 𝑟𝑢 (𝑡)󵄩󵄩󵄩󵄩2
󵄩 󵄩2 [−0.4] [ 0.4 ] [−0.3]
(53)
󵄩 󵄩 󵄩 󵄩
≤ 󵄩󵄩󵄩𝑟𝑛 (𝑡)󵄩󵄩󵄩2 + 󵄩󵄩󵄩𝑟𝑢 (𝑡)󵄩󵄩󵄩2 ≤ 𝐽th,𝑛 + 𝐽th,𝑢 , 1 0 0 0.2 0.1
𝐶 = [0 1 0] , 𝐹𝑦 = [ 0.8 ] , 𝑅 = [ 0.2 ] ,
where
[0 0 1] [−1.2] [−0.4]
󵄩󵄩 󵄩
𝐽th,𝑛 = sup 󵄩󵄩𝑟𝑛 (𝑡)󵄩󵄩󵄩2 , 0.9
(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 ) ∈Ω𝑖
𝐷 = [0.2] .
󵄩󵄩 󵄩 (54)
𝐽th,𝑢 = sup 󵄩󵄩𝑟𝑢 (𝑡)󵄩󵄩󵄩2 . [0.7]
(Δ𝐴,Δ𝐵,Δ𝐴 𝑑 ) ∈Ω𝑖 (57)
10 Journal of Applied Mathematics

Uncertainties are also defined by the following matrices Abrupt fault signal
in (2):

0.1 0.1 −0.1 1

𝑀1 = [0.2] , 𝑀2 = [ 0 ] , 𝑀3 = [ 0.2 ] ,
[0.1] [−0.1] [ 0.1 ] 0.8

𝑁1 = [0 0.1 0.3] ,

Amplitude
0.6
𝑁2 = [0.1 0 0] ,
𝑁3 = 0.1. 0.4
(58)
0.2
The dynamic characteristic of fault is considered as
0
𝐴 𝜃 = 0, 𝐹𝜃 = 1. (59) 0 1 2 3 4 5 6 7 8 9 10
Time (s)
The first step to design the fault detection system is to
Figure 1: An abrupt fault occurs at 𝑡 = 5.
solve the LMIs (26) and (32) in Theorems 3 and 4. The
Yalmip LMI toolbox is used to solve the LMIs. To start the
iterative optimization method presented in Remark 6, the
initial values 𝛼int = 3 and 𝛽int = 0 are selected. Using this 0.0372 −1.1695 0.8939
procedure, the following results are obtained: 𝐿 1 = [−0.2145 0.7355 0.1978] ,
[−0.7462 −0.9036 1.6937]
−0.0994 −0.0532 0.2714 −0.7701 −0.1957 0.9507
𝑉1∗ = [−0.0532 −5.4751 1.3859 ] , 𝐾1𝑧 = [−0.1730 −0.0567 0.2159] ,
[ 0.2714 1.3859 −0.7651] [−0.5857 −0.1278 0.7254]
−0.6315 −0.8304 0.1871 0.2122 −0.3395 −0.7167
𝑉2∗ = [−0.8304 0.6711 0.5971 ] , 𝐾2𝑧 = [0.0496 −0.0777 −0.1591] .
[ 0.1871 0.5971 −0.5845] [0.1603 −0.2648 −0.5534]
662.5 −250.8 −2628.5 (61)
Φ∗1 = [ −250.8 3855.7 −281.2 ] ,
To verify the sensitivity of designed UIO, an abrupt fault,
[−2628.5 −281.2 2055.8 ] (60)
shown in Figure 1, occurs in the 4 seconds elapsed from
984.8 −1453.4 −842.7 running of the system. The step disturbance signal exerted
Φ∗2 = [−1453.4 2292 1253.8 ] , to the system between 3 to 7 seconds. The noise signal is
[ −842.7 1253.8 733.6 ] assumed to be white Gaussian noise with power 0.0005, and
the uncertainty Σ𝑖 (𝑡) is considered sinusoidal signal. The
−4.7395 13.0831 0.7977 residual signals are shown in Figure 2. It can be seen that the
Φ∗3 = [ 13.0831 −8.5691 −14.9116] , residual signals change when the fault occurs; however, the
[ 0.7977 −14.9116 −3.0645 ] residual signals do not show any sensitivity to the exerted dis-
turbance. The value of threshold 𝐽th is presented in Figure 3.
𝛼 = 2.4, 𝛽 = 1. This figure indicates that the fault is detected rapidly and
the difference between threshold and norm of faulty residual
Using these values, the LMI (51) is solved and the observer signal is high enough to detect the occurrence of fault in the
dynamic matrices are obtained as follows: system.
The RMS of residual signals (56) has been depicted
−3.6557 −3.3806 2.9608 in Figure 4. It can be seen that the RMS of faulty signals
𝐹 = [−0.3878 −1.4059 0.3008] , suddenly changes in contrast to RMS of fault-free signals.
[−2.4404 −3.2245 1.9455] Therefore, the occurrence of fault can be effectively realized.
−0.0559 −0.7284 −0.1686
𝐺 = [−0.0153 −0.1629 −0.0350] , 4. Conclusions
[−0.0445 −0.557 −0.1273]
In this paper, a novel UIO-based residual generator is
0.6878 developed for robust fault detection purposes. The developed
𝐻 = [0.1535] , method is applicable to a variety of linear uncertain time-
[0.5328] delay systems. The proposed approach is able to decouple
Journal of Applied Mathematics 11

Residual signals RMS norm of residual signal (50)

3 2.5

1 2

−1 1.5

Amplitude
Amplitude

−2

−3 1

−4

−5 0.5

−6

−7 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time (s) Time (s)

Figure 2: Residual signals of UIO fault detection filter. Without fault

With fault

Figure 4: Detecting occurrence of fault using RMS norm of ‖𝑟(𝑡)‖𝑇2 .

Euclidean norm of residual signals and value of Jth
9

8
the data framework for designing unknown input observer
7 for time-delay system is an interesting area.

6
Conflict of Interests
Amplitude

5
The authors declare that there is no conflict of interests
4
regarding the publication of this paper.
3

2 References
1 [1] L. Zhang, E.-K. Boukas, L. Baron, and H. R. Karimi, “Fault
detection for discrete-time Markov jump linear systems with
0 partially known transition probabilities,” International Journal
0 1 2 3 4 5 6 7 8 9 10
of Control, vol. 83, no. 8, pp. 1564–1572, 2010.
Time (s)
[2] D. Dongsheng, B. Jiang, P. Shi, and H. R. Karimi, “Fault detec-
Without fault tion for continues-time switched systems under asynchronous
With fault switching,” International Journal of Robust and Nonlinear Con-
Jth trol, 2013.
Figure 3: Euclidean norm of residual fault-free and faulty signals [3] J. Yu, G. Sun, and H. R. Karimi, “Fault reconstruction based
and 𝐽th . cascaded sliding mode observers for descriptor linear systems,”
Mathematical Problems in Engineering, vol. 2012, Article ID
623426, 20 pages, 2012.
[4] S. Yin, X. Yang, and H. R. Karimi, “Data-driven adaptive
thoroughly exogenous disturbances while minimizing uncer- observer for fault diagnosis,” Mathematical Problems in Engi-
tainties and noise effects. The fault effect is also maximized at neering, vol. 2012, Article ID 832836, 21 pages, 2012.
the same time. To this end, first, the optimal fault detection [5] J. Chen and R. J. Patton, Robust Model-Based Fault Diagnosis for
filter is designed for system without considering uncertain- Dynamic Systems, Kluwer Academic Publishers, 1999.
ties. Then, the fault detection filter is designed so that the 𝐻∞ [6] J. Guo, X. Huang, and Y. Cui, “Design and analysis of robust
norm between the fault detection filter and the optimal one is fault detection filter using LMI tools,” Computers and Mathe-
minimized. Superiority of the proposed approach has been matics with Applications, vol. 57, no. 11-12, pp. 1743–1747, 2009.
verified through a numerical example. Simulation results [7] M. Zhong, S. X. Ding, T. Bingyong, T. Jeinsch, and M. Sader,
show that the proposed approach is able to detect dynamic “An LMI approach to design robust fault detection observers,”
faults. As a future work, one can extend this approach to non- in Proceedings of the 4th World Congress on Intelligent Control
linear systems and descriptor system. Moreover, developing and Automation, pp. 2705–2709, June 2002.
12 Journal of Applied Mathematics

[8] M. Zhong, S. X. Ding, J. Lam, and H. Wang, “An LMI approach

to design robust fault detection filter for uncertain LTI systems,”
Automatica, vol. 39, no. 3, pp. 543–550, 2003.
[9] T. Li, L. Wu, and X. Wei, “Robust fault detection filter design
for uncertain LTI systems based on new bounded real lemma,”
International Journal of Control, Automation and Systems, vol. 7,
no. 4, pp. 644–650, 2009.
[10] M. Zhong, S. X. Ding, J. Lam, and C. Zhang, “Fault detection
filter design for LTI system with time delays,” in Proceedings of
the 42nd IEEE Conference on Decision and Control, pp. 1467–
1472, December 2003.
[11] L. Bai, Z. Tian, and S. Shi, “Design of H∞ robust fault detection
filter for linear uncertain time-delay systems,” ISA Transactions,
vol. 45, no. 4, pp. 491–502, 2006.
[12] M. Zhong, H. Ye, T. Sun, and G. Wang, “An iterative LMI
approach to RFDF for linear system with time-varying delays,”
Asian Journal of Control, vol. 8, no. 1, pp. 86–90, 2006.
[13] H. Wang, C. Wang, and H. Gao, “An LMI approach to robust
fault detection filter design for uncertain state-delayed systems,”
in Advances in Machine Learning and Cybernetics, vol. 3930 of
Lecture Notes in Computer Science, pp. 112–121, 2006.
[14] H. R. Karimi, M. Zapateiro, and N. Luo, “A linear matrix
inequality approach to robust fault detection filter design of
linear systems with mixed time-varying delays and nonlinear
perturbations,” Journal of the Franklin Institute, vol. 347, no. 6,
pp. 957–973, 2010.
[15] M. Witczak, A. Obuchowicz, and J. Korbicz, “Genetic program-
ming based approaches to identification and fault diagnosis of
non-linear dynamic systems,” International Journal of Control,
vol. 75, no. 13, pp. 1012–1031, 2002.
[16] G.-R. Duan and R. J. Patton, “Robust fault detection using
Luenberger-type unknown input observers—a parametric
approach,” International Journal of Systems Science, vol. 32, no.
4, pp. 533–540, 2001.
[17] P. M. Frank and X. Ding, “Survey of robust residual generation
and evaluation methods in observer-based fault detection
systems,” Journal of Process Control, vol. 7, no. 6, pp. 403–424,
1997.
[18] S. Ahmadizadeh, J. Zarei, and H. R. Karimi, “Robust unknown
input observer design for linear uncertain time delay systems
with application to fault detection,” Asian Journal of Control, vol.
16, pp. 1–14, 2014.
[19] W. Chen and M. Saif, “Fault detection and isolation based on
novel unknown input observer design,” in Proceedings of the
American Control Conference, pp. 5129–5134, June 2006.
[20] S. Mondal, G. Chakraborty, and K. Bhattacharyya, “LMI
approach to robust unknown input observer design for continu-
ous systems with noise and uncertainties,” International Journal
of Control, Automation and Systems, vol. 8, no. 2, pp. 210–219,
2010.
[21] G.-Y. Tang and J. Li, “Optimal fault diagnosis for systems with
delayed measurements,” IET Control Theory and Applications,
vol. 2, no. 11, pp. 990–998, 2008.
[22] Y.-M. Fu, G.-R. Duan, and S.-M. Song, “Design of unknown
input observer for linear time-delay systems,” International
Journal of Control, Automation and Systems, vol. 2, no. 4, pp.
530–535, 2004.
Advances in Advances in Journal of Journal of
Operations Research
Hindawi Publishing Corporation
Decision Sciences
Hindawi Publishing Corporation
Applied Mathematics
Hindawi Publishing Corporation
Algebra
Hindawi Publishing Corporation
Probability and Statistics
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

The Scientific International Journal of

World Journal
Hindawi Publishing Corporation
Differential Equations
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

International Journal of Advances in

Combinatorics
Hindawi Publishing Corporation
Mathematical Physics
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of Journal of Mathematical Problems Abstract and Discrete Dynamics in

Complex Analysis
Hindawi Publishing Corporation
Mathematics
Hindawi Publishing Corporation
in Engineering
Hindawi Publishing Corporation
Applied Analysis
Hindawi Publishing Corporation
Nature and Society
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

International
Journal of Journal of
Mathematics and
Mathematical
Discrete Mathematics
Sciences

Journal of International Journal of Journal of

Hindawi Publishing Corporation Hindawi Publishing Corporation Volume 2014

Function Spaces
Hindawi Publishing Corporation
Stochastic Analysis
Hindawi Publishing Corporation
Optimization
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014