Available online at www.sciencedirect.

com

Journal of the Franklin Institute 351 (2014) 3104–3125

www.elsevier.com/locate/jfranklin

Observer-based fault tolerant control design for a class

of LPV descriptor systems
Mickael Rodriguesa,n, Habib Hamdib, Naceur BenHadj Braiekb,
Didier Theilliolc
a
Laboratoire d'Automatique et Génie des Procédés, Universitéde Lyon, Lyon F-69003,
France Université Lyon1, CNRSUMR 5007, Villeurbanne F-69622, France
b
Laboratory for Advanced Systems at the Tunisia Polytechnic School, University of Carthage, Tunisia
c
Centre de Recherche en Automatique de Nancy, Université de Lorraine, CNRS UMR 7039, France
Received 5 April 2013; received in revised form 9 January 2014; accepted 27 February 2014
Available online 11 March 2014

Abstract

This paper presents a new Fault Tolerant Control (FTC) methodology for a class of LPV descriptor
systems that are represented under a polytopic LPV form. The aim of this FTC strategy is to compensate the
effects of time-varying or constant actuator faults by designing an Adaptive Polytopic Observer (APO)
which is able to estimate both the states of the system and the magnitude of the actuator faults. Based on the
information provided by this APO, a new state feedback control law is derived in order to stabilize the
system. Stability conditions of the designed observer and the state-feedback control are provided and solved
through a set of Linear Matrix Inequalities (LMI) under equality constraints. The performance of the
proposed Fault Tolerant Control scheme is illustrated using a two-phase flash system.
& 2014 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.

1. Introduction

As control of systems become more and more complex, the security remains a key point and
the development of new control theory which integrates the faults that can occur on a system is
of a great interest. Fault Diagnosis [1,2] and Fault Tolerant Control [3] have become challenging
n
Corresponding author.
E-mail addresses: [email protected] (M. Rodrigues), [email protected] (H. Hamdi),
[email protected] (D. Theilliol).

http://dx.doi.org/10.1016/j.jfranklin.2014.02.016
0016-0032/& 2014 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3105

problems in the area of modern control theory. The concept of Fault Tolerant Control is based on
the fact that when a fault (a sensor or an actuator for example) occurs on the system and provides
an undesirable effect, the system can become unstable or be damaged. By the way, the basic idea
of this concept is to be robust against such fault or to take into account the fault occurrence into a
new control which will become tolerant to this fault by canceling its bad effects.
Fault Tolerant Control (FTC) techniques can be classified into two categories [4]: passive and
active approaches. In passive FTC systems, a single controller with fixed structure or parameters
is used to deal with all possible failure scenarios which are assumed to be known a priori.
Consequently, the passive controller is usually conservative. Furthermore, if a failure that would
not be considered in the design occurs, the stability and performance of the closed-loop system
cannot be guaranteed. Such potential limitations of passive approaches provide a strong
motivation for the development of methods and strategies for Active FTC (AFTC) systems [5].
In contrast to passive FTC systems, AFTC techniques rely on a real time fault detection and
isolation (FDI) scheme and a controller reconfiguration mechanism. Such techniques allow a
flexibility to select different controllers according to different component failures, and therefore
better performance of the closed-loop system can be expected. However, this holds true when the
FDI process does not make an incorrect decision [6]. A FTC strategy is designed so as to
reconfigure automatically the control law by ensuring the system stability and to get acceptable
system performances [7,8]. Observer based Fault Tolerant Control methods are also developed in
order to estimate the fault and to reconfigure the control law [9]. The authors in [10,11] have
developed a FDI/FTC strategy for regular Takagi–Sugeno (TS) systems where both the observer
and the control are designed at the same time. However, all these previous above-mentioned FTC
techniques are devoted only for normal (regular) systems whereas here in this paper, the main
goal is to design an Active FTC strategy for descriptor (singular) systems.
Note that Takagi–Sugeno fuzzy systems have always been used in the past membership
functions that were computed by the fuzzy logic theory [12]. But recently, both polytopic LPV
systems and a part of fuzzy systems converge to a same structure. The community of people
working on TS models uses the name “TS FUZZY systems” even if with the recent modeling
approaches (for example sector nonlinearity transformation), the obtained model is no longer
“fuzzy” because the weighting functions are completely deterministic which corresponds to LPV
or quasi-LPV systems [13].
Generally speaking, most of the control research works for physical systems use a normal (or
regular) model, i.e., there is no algebraic relations between the system variables. However,
Differential-Algebraic Equations (DAE) or implicit systems or singular systems or descriptor
systems are of quite importance for the physical representation of some systems [14]. Such
systems appear for example in electrical circuits, mechanical systems with holonomic or non-
holonomic constraints, robotic systems with kinematical constraints and chemical systems [15].
Some practical problems must take into account physical constraints or algebraic relations and
more generally impulsive behaviors caused by an improper transfer matrix: see the following
books on singular systems [16,17].
Concerning FDI for descriptor systems, some authors have considered this problem as in
Darouach and Boutayeb [18], Youssouf and Kinnaert [19] for the general linear case, Astorga-
Zaragoza et al. [20] and Wang et al. [21] for linear descriptor systems in discrete case by
designing an observer through LMI study. New recent works on robust HInfinity control design by
LMI for discrete-time descriptor systems can be found in Chadli and Darouach [22,23].
The concept of Linear Parameter Varying systems (LPV) allows the convenience associated
with LTI models, and yet guarantees performance and stability over a more wide operating
3106 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

envelope. Some results about FDI have been developed for normal LPV systems as in Alwi et al.
[24] by a sliding mode observer, Bokor and Balas [25], Armeni et al. [26], Bokor and Szabo [2]
with a geometrical approach. Nonlinear systems are sometimes represented by a LPV
modelization [27–30] in order to use the technique developed in the linear case like the tools
for stability purposes as LMI Toolboxes.
In Wu and Zheng [31], and Wu et al. [32], the authors have developed a technique for state
estimation and sliding-mode control of Markovian jump singular systems and also by
considering time-delay in Wu et al. [33]. In Li and Zhang [34], the authors have developed a
robust H 1 filtering for singular LPV systems with time varying delays so as to estimate the states
of the system but without any FDI purposes. In Hamdi et al. [35], the authors have developed a
robust FDI method based on a multiple models concept. In Marx et al. [36], the authors have
developed a robust fault tolerant control for descriptor systems but only with constant matrices.
In the paper of Koenig [37], the author has introduced some useful necessary observability
conditions for the design of unknown input observers for descriptor systems. A Fault Tolerant
Control technique is presented for normal LPV systems under sensor faults in Oca et al. [38].
In Hamdi et al. [39], the authors have proposed a polytopic unknown inputs and proportional
integral observers for LPV descriptor systems. However, this technique cannot ensure a correct
fault estimation if the fault is time-varying. By this way, the authors in Rodrigues et al. [40] have
performed their previous works by designing an Adaptive Observer in order to take into account
time-varying faults for descriptor LPV systems. In a similar way, the authors in Wang and Daley
[41] have presented an adaptive fault diagnosis observer approach dedicated to regular LTI
systems which can detect and estimate only constant faults. In Zhang et al. [42], the authors have
performed this previous adaptive observer so as to estimate time-varying faults but only in a LTI
case for regular systems.
In the papers [43,44], the authors have developed an active FTC strategy to avoid actuator
fault/failure effects on polytopic LPV systems; however FDI was not performed and was
supposed to be available and perfect. Moreover, very few contributions are dealing with Fault
Tolerant Control for polytopic LPV descriptor system with a FDI scheme designed at the same
time. The main contributions of this paper are the following:

 To design an Adaptive Polytopic LPV Observer (APO) that can estimate time-varying
actuator faults. Some previous Fault Detection and Isolation (FDI) techniques presented in
Astorga-Zaragoza et al. [20,45] and in Hamdi et al. [39] can only deal with constant faults for
LPV descriptor systems.
 To integrate the information provided by the APO into a new state-feedback design so as to
cancel the actuator fault effects with Fault Tolerant Control (FTC). The FDI and FTC parts are
designed at the same time whereas most of FTC strategies deal with normal LTI or LPV
systems and assume that the FDI part is perfect and not designed. For LPV descriptor systems,
such strategy has never been used.
 To ensure both the stability of the APO and the Fault Tolerant Control by LMI under equality
constraints for LPV descriptor systems.

So, in this paper, an integrated Fault Diagnosis (FD) and FTC design for polytopic LPV
descriptor systems are provided. Polytopic LPV descriptor system is a particular class of LPV
systems which allows us to describe the system as a convex combination of sub-models defined
by the vertices of a convex polytope. These sub-models are then combined by convex weighting
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3107

functions that yield to a global model. Using an Adaptive Polytopic Observer (APO) that is able
to provide both states and actuator faults estimation, it is possible to address the Fault Diagnosis
(FD), and at the same time to build a new control which takes into account the actuator fault
estimation. The use of such Adaptive Polytopic Observer is motivated by the fact that, if a fault
occurs, it is important to quickly detect and estimate it in order to preserve the system
performance in spite of the presence of fault. Moreover, this APO is able to estimate time-
varying fault which was neither possible with our previous paper [39] nor with [45]. Stability
analysis and sufficient conditions are obtained with the use of Linear Matrix Inequality (LMI)
under equality constraint. A lot of works dealing with quadratic stability have been done as in
Cai et al. [46] by the use of LMI or also for fault detection purposes [47].
The structure of this paper is organized as follows: in Section 2, the class of the LPV
descriptor systems is presented. Section 3 describes the problem statement. A method of
designing the Adaptive Polytopic Observer is described in Section 4. Fault tolerant control by
state feedback is tackled in Section 5; finally, and before concluding, a numerical example that
considers a two-phase flash system is used to assess the validity of the proposed approach.
Notations: For symmetric matrices X40 (X Z 0) indicates that X is positive definite (positive
semi-definite). For any square matrix M, λmax ðMÞ represents the maximum singular value of the
matrix M. In a partitioned matrix, the asterisk ‘n’ denotes the terms induced by symmetry.

2. Polytopic LPV descriptor systems modeling

Consider the following continuous-time LPV descriptor representation in the fault-free case:
(
~
E x_ ðtÞ ¼ AðθðtÞÞxðtÞ ~
þ BðθðtÞÞuðtÞ
ð1Þ
yðtÞ ¼ CxðtÞ

where xðtÞA Rn is the state vector, uðtÞA Rp is the input vector, yðtÞA Rm represents the measured
outputs vector and θðtÞ is a varying parameter vector. Matrix E A Rnxn may be singular and
rankðEÞ ¼ ron.
It is assumed that all parameters θi ðtÞ, i ¼ 1; …; l, are bounded, measurable as in Alwi et al.
[24] and Rodrigues et al. [30], and their values remain in the domain of an hypercube such
that [27]

θðtÞ A Γ ¼ fθi ∣θ i r θi ðtÞr θ i g; 8t Z0 ð2Þ

where θ i and θ i represent the minimum and maximum values of θi ðtÞ respectively.
~
AðÞ; ~ are functions which depend affinely on the time-varying parameter vector θðtÞ A Rl .
BðÞ
~
The matrices AðθðtÞÞ; ~
BðθðtÞÞ of the LPV system (1) with the affine parameter dependence (2)
are represented such that
l l
~
AðθðtÞÞ ¼ A~ 0 þ ∑ θi ðtÞA~ i ; ~
BðθðtÞÞ ¼ B~ 0 þ ∑ θi ðtÞB~ i 8 θðtÞA Γ ð3Þ
i¼1 i¼1

The LPV system (1) with bounded parameters can be represented by a polytopic form where
the summits Si of the polytope are defined such that [44] Si ¼ ½Ai Bi C; 8 iA ½1; …; h where
3108 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

h ¼ 2l . The polytopic coordinates are denoted as ρðθðtÞÞ and vary within the convex set Ω:
 h

Ω ¼ ρðθðtÞÞA R ; ρðθðtÞÞ ¼ ½ρ1 ðθðtÞÞ; …; ρh ðθðtÞÞ ; ρi ðθðtÞÞ Z 0; 8 i; ∑ ρi ðθðtÞÞ ¼ 1
h T
i¼1
ð4Þ
~ and BðÞ
Then, to ease the presentation, it is assumed that the matrices AðÞ ~ are given by convex
combinations 8 t Z 0. Consequently, system (1) can be rewritten by a polytopic representation:
8
> h
< Ex_ ðtÞ ¼ ∑ ρ ðθðtÞÞðA xðtÞ þ B uðtÞÞ
i i i
i¼1 ð5Þ
>
: yðtÞ ¼ CxðtÞ

where Ai A Rnn , Bi A Rnp and C A Rmn are time invariant matrices defined for the ith summit
of the polytope.

3. Problem statement

Let us consider an actuator fault on the previous descriptor system:

8
> h
< Ex_ ðtÞ ¼ ∑ ρ ðθðtÞÞ½A xðtÞ þ B ðuðtÞ þ f ðtÞÞ
i i i
i¼1 ð6Þ
>
: yðtÞ ¼ CxðtÞ

where Ai A Rnn , Bi A Rnp and C A Rmn are time invariant matrices defined for the ith model.
f ðtÞ A Rp is the actuator fault vector. Actuator faults can be represented by an additive or a
multiplicative external signal as in Rodrigues et al. [44]. These malfunctions of an actuator can
be represented by a faulty control input uf ðtÞ ¼ ðI p  γÞuðtÞ which can be rewritten as an external
additive signal: uðtÞ þ f ðtÞ where f ðtÞ ¼  γuðtÞ with
γ 9 diag½γ 1 ; γ 2 ; …; γ p ; 0 r γ k r 1 such that
(
γ k ¼ 1-a total failure of the kth actuator k A ½1; …; p
ð7Þ
γ k ¼ 0-the kth healthy actuator

Note: in the following of the paper, γ k A ½0 1½ i.e a total loss of an actuator is not considered
here. The term γk represents the loss of effectiveness of kth actuator, i.e., for example a loss of
effectiveness 60% of 1st actuator will be represented by γ 1 ¼ 0:60. When an actuator fault
appears on the system, such actuator faults can cause system instability. Before starting the FTC
design, we assume that [18,42,39]

Assumption A1. rankðCBi Þ ¼ rankðBi Þ ¼ p; 8 i ¼ 1; …; h,

Assumption A2. The triple matrix ðE; Ai ; CÞ is R-observable, for all i ¼ 1; …; h, i.e.,
 
sE  Ai
rank ¼ n; 8 sA C: ð8Þ
C

where C denotes the complex plane.

 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3109

Assumption A3. The triple matrix ðE; Ai ; CÞ is impulse-observable, for all i ¼ 1; …; h, i.e.,
2 3
E Ai
6 7
rank 4 0 E 5 ¼ n þ rankðEÞ ð9Þ
0 C
Assumption A4. The fault f(t) satisfies Jf ðtÞJ r α1 and the derivative of f(t) with respect to time
is norm bounded, i.e., J f_ ðtÞJ r α2 and 0 r α1 ; α2 o1.

Assumption A5. Only partial actuator faults are considered, i.e., γ k A ½0 1½.

Noting that, the R-observability characterizes the capacity to reconstruct only the state of the
dynamic part and the Impulse-observability guarantees the capacity to estimate the state of static
part of the descriptor system (6).
The main objective of an Active FTC is to find a control law such that the system remains
stable despite the presence of actuator faults [44,4]. For this purpose, a FDI procedure is
necessary for estimating both the states and faults. In the following, an AFTC with a state
feedback will be used such that
h
uðtÞ ¼  ∑ ρi ðθðtÞÞK i x^ ðtÞ f^ ðtÞ ð10Þ
i¼1

The following section is dedicated to synthesize an adaptive observer for Polytopic LPV
descriptor systems.

4. Adaptive polytopic observer design

Consider the following Adaptive Polytopic Observer (APO) as

8
>
>
h
>
> z_ ðtÞ ¼ ∑ ρi ðθðtÞÞðN i zðtÞ þ Gi uðtÞ þ Ri yðtÞ þ Bi f^ ðtÞÞ
>
>
>
>
i¼1
>
> ^ ðtÞ ¼ zðtÞ þ T 2 yðtÞ
>
< x
y^ ðtÞ ¼ C x^ ðtÞ ð11Þ
>
>
>
> _ h
>
>
> f^ ðtÞ ¼ Γ ∑ ρi ðθðtÞÞU i ð_e y ðtÞ þ sey ðtÞÞ
>
> i¼1
>
>
: ey ðtÞ ¼ yðtÞ y^ ðtÞ ¼ Cex ðtÞ

where z(t) is the state vector, f^ ðtÞ is an estimate of the fault f(t) and y^ ðtÞ ¼ C x^ ðtÞ is the estimated
output vector. The matrix Γ A Rf is a symmetric positive definite learning rate matrix. N i , Gi , Ri
and T2 are unknown matrices of appropriate dimensions to be determined. In the case of actuator
faults, the matrix Fi of the Observer developed in Rodrigues et al. [40] is equal to Bi.
Let us define the following state estimation error ex(t) from Eqs. (6) and (11) such that
ex ðtÞ ¼ xðtÞ x^ ðtÞ ¼ ðI n  T 2 CÞxðtÞ zðtÞ ð12Þ
E 
Since for rank C ¼ n, there exist nonsingular matrices T 1 A Rnn and T 2 A Rnm such that
T 1E þ T 2C ¼ In ð13Þ
3110 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

Then, the state estimation error (12) is described by

ex ðtÞ ¼ T 1 ExðtÞ zðtÞ ð14Þ
and the actuator fault estimation error ef(t) is defined by
ef ðtÞ ¼ f ðtÞ f^ ðtÞ ð15Þ
First, the state estimation error dynamic is given by
e_ x ðtÞ ¼ T 1 E x_ ðtÞ  z_ ðtÞ ð16Þ
By using Eq. (6) with the state feedback control law defined in Eq. (10), after some
calculations Eq. (16) becomes
h
e_ x ðtÞ ¼ ∑ ρi ðθðtÞÞ½N i ex ðtÞ þ ðT 1 Ai  N i T 1 E  Ri CÞxðtÞ þ ðT 1 Bi  Gi ÞuðtÞ
i¼1
þ ðBi þ T 1 Bi  Gi Þef ðtÞ þ ðGi  Bi Þf ðtÞ ð17Þ
Then, if the following conditions hold true 8 i ¼ 1; …; h,
T 1 Ai  Ri C  N i T 1 E ¼ 0 ð18Þ

T 1 Bi  Gi ¼ 0 ð19Þ
and by taking into account Eqs. (6), (11) and (16), the state estimation error dynamic can be
written as
h
e_ x ðtÞ ¼ ∑ ρi ðθðtÞÞ½N i ex ðtÞ þ Bi ef ðtÞ þ M i f ðtÞ ð20Þ
i¼1

with M i ¼ ðT 1  I n ÞBi .
The substitution of Eq. (13) into Eq. (18) yields
N i ¼ T 1 Ai þ ðN i T 2  Ri ÞC ¼ T 1 Ai þ Li C ð21Þ
where Li ¼ N i T 2  Ri .
Without loss of generality, considering Eq. (13), one can write the following relationship:
 
E
½T 1 T 2  ¼ ½I n  ð22Þ
C
A solution ½T 1 T 2  exists if [18,39]
 
E
rank ¼n ð23Þ
C
Then, a particular solution of Eq. (22) using the generalized inverse matrix denoted by ðÞþ is
given by
 þ
E
½T 1 T 2  ¼ ð24Þ
C
Based on a fault estimation given by the APO, the objective of AFTC scheme is to design a
feedback control law such that the system remains stable even if a fault occurs. The following
section is dedicated to the stability conditions of this AFTC based on an actuator fault estimation
from the Adaptive Polytopic Observer.
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3111

5. Fault tolerant control by state-feedback and fault estimation

The dynamic of the state estimation error (20) and the closed-loop system with the control law
(10) are defined as follows:
h
e_ x ðtÞ ¼ ∑ ρi ðθðtÞÞ½N i ex ðtÞ þ Bi ef ðtÞ þ M i f ðtÞ ð25Þ
i¼1

h h
Ex_ ðtÞ ¼ ∑ ∑ ρi ðθðtÞÞρj ðθðtÞÞ½Φij xðtÞ þ Bi K j ex ðtÞ þ Bi ef ðtÞ ð26Þ
i¼1i¼1

with N i ¼ T 1 Ai þ Li C and Φij ¼ ðAi  Bi K j Þ. The goal is to ensure the stability of these two
differential equations.

5.1. Stability analysis

In order to investigate the stability criteria, let us consider the following Lemma:
Lemma 1 (Zhang et al. [42]). Given a scalar μ40 and a symmetric positive definite matrix P1,
the following inequality holds
1
2xT yr xT P1 x þ μyT P1 1 y x; yA Rn □ ð27Þ
μ

In contrast to Hamdi et al. [39], here time-varying faults are considered. Then, it follows that
f_ ðtÞa 0 and consequently
_
e_ f ðtÞ ¼ f_ ðtÞ f^ ðtÞ ð28Þ

Theorem 1. Under Assumptions A1–A5, given scalars s, μ, β40, if there exist symmetric
positive definite matrices X, Q, P1, P2 and matrices Wi and Si such that 8 iA ½1; …; h,
8j A ½1; …; h,
0 1
Θij Bi W j Bi 0 0
B n  2δX δI 0 C
B 0 C
B C
B n n  2δI 0 δI Co0 ð29Þ
B C
B n n n Ω Σ C
@ i ij A
n n n n Υ ij

s.t.
ET P1 ¼ PT1 E Z 0 ð30Þ

BTi Q  U i C ¼ 0 ð31Þ
where
Θij ¼ ðAi X  Bi W j Þ þ ðAi X  Bi W j ÞT ð32Þ

1
Ωi ¼ ðQT 1 Ai  Si C Þ þ ðQT 1 Ai  Si CÞT þ P1 ð33Þ
μ
3112 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

1 T T
Σ ij ¼  A T Q  C T STj Bi ð34Þ
s j 1
1 T 2
Υ ij ¼  Bi QBj þ BTj QBi þ P2 ð35Þ
s sμ
then, the state x(t) of the system, the state estimation error ex(t) and the fault estimation error
ef(t) are bounded. The gains of the observer and the state feedback control law are given by
Li ¼ Q  1 Si and K i ¼ W i X  1 respectively. □

Proof. In order to prove the stability of the closed-loop system and the convergence of the state
and fault estimation errors, let us consider the Lyapunov function depending on xðtÞ; ex ðtÞ and
ef(t):
1
V ðt Þ ¼ xT E T P1 x þ eTx ðt ÞQex ðt Þ þ eTf ðt ÞΓ  1 ef ðt Þ ð36Þ
s
where P1 , Q and Γ are symmetric positive definite matrices with appropriate dimensions.
Stability condition for the estimation error yields that the time derivative of the Lyapunov
function (36) should be negative definite. By taking into account Eqs. (25) and (26) the
derivative of V(t) with respect to time is
h h
V_ ðt Þ ¼ ∑ ∑ ρi ðθðt ÞÞρj ðθðt ÞÞ ðE x_ ÞT P1 x þ xT PT1 E x_ þ eTx ðtÞ½N Ti Q þ QN i ex ðtÞ
i¼1i¼1

1 T 1 1 T 1
þ2eTx ðt ÞQM i
f ðt Þ þ þ e_ f ðt ÞΓ ef ðt Þ þ ef ðt ÞΓ e_ f ðt Þ
2eTx ðt ÞQBi ef ðt Þ ð37Þ
s s
By considering Eqs. (30) and (28) and the expression of f^ ðtÞ in Eq. (11), we can obtain
h h

_
V ðt Þ ¼ ∑ ∑ ρi ðθðt ÞÞρj ðθðt ÞÞ xT ðtÞΠ ij xðtÞ þ eTx ðtÞΩi ex ðtÞ þ 2eTx ðtÞQM i f ðtÞ
i¼1i¼1
2
þ 2eTx ðt ÞQBi ef ðt Þ þ 2xT P1 Bi K j ex ðt Þ þ 2xT P1 Bi ef ðt Þ eTf ðt ÞU i e_ y ðt Þ þ sey ðt Þ
 s
2
þ eTf ðt ÞΓ  1 f_ ðt Þ ð38Þ
s
with ey ðtÞ ¼ Cex ðtÞ and the following notations
Π ij ¼ ΦTij P1 þ P1 Φij ð39Þ

Ωi ¼ N Ti Q þ QN i ð40Þ
By using Eq. (25), it follows that
h h

V_ ðt Þ ¼ ∑ ∑ ρi ðθðt ÞÞρj ðθðt ÞÞ xT ðtÞΠ ij xðtÞ þ eTx ðtÞΩi ex ðtÞ þ 2eTx ðtÞQM i f ðtÞ
i¼1i¼1
2
þ 2eTx ðt ÞQBi ef ðt Þ þ 2xT P1 Bi K j ex ðt Þ þ 2xT P1 Bi ef ðt Þ eTf ðt ÞU i CN j ex ðt Þ
s 
2 T 2 T 2 T 1_
 ef ðt ÞU i CBj ef ðt Þ 2ef ðtÞ U i Cex ðt Þ ef ðt ÞU i CM j f ðt Þ þ ef ðt ÞΓ f ðt Þ
T
s s s
ð41Þ
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3113

By using Assumption A1 and the equality (31), it follows that

h h

V_ ðt Þ ¼ ∑ ∑ ρi ðθðt ÞÞρj ðθðt ÞÞ xT ðtÞΠ ij xðtÞ þ eTx ðtÞΩi ex ðtÞ þ 2eTx ðtÞQM i f ðtÞ
i¼1i¼1
2 2
þ 2xT P1 Bi K j ex ðt Þ þ 2xT P1 Bi ef ðt Þ eTf ðt ÞBTi QN j ex ðt Þ eTf ðt ÞBTi QBj ef ðt Þ
s s
2 2
 eTf ðt ÞBTi QM j f ðt Þ þ eTf ðt ÞΓ  1 f_ ðt Þ ð42Þ
s s
Now, using the Assumption A4 and applying Lemma 1 for three terms of the above inequality, it
is shown that
1
2eTx ðt ÞQM i f ðt Þ r eTx ðt ÞP1 ex ðt Þ þ μf ðtÞT M Ti QP1 1 QM i f ðt Þ
μ
1 T
r ex ðt ÞP1 ex ðt Þ þ μα21 λmax M Ti QP1 1 QM i ð43Þ
μ
2 T 1 T μ T
e ðt ÞΓ  1 f_ ðt Þr e ðt ÞP2 ef ðt Þ þ f_ ðt Þ Γ  1 P2 1 Γ  1 f_ ðt Þ
s f sμ f s
1 T μ
r ef ðt ÞP2 ef ðt Þ þ α22 λmax Γ  1 P2 1 Γ  1 ð44Þ
sμ s
2 1 T μ 
 eTf ðt ÞBTi QM j f ðt Þr ef ðt ÞP2 ef ðt Þ þ f ðtÞT M Tj QBi P2 1 BTi QM j f ðt Þ
s sμ s
1 T μ 2 
r ef ðt ÞP2 ef ðt Þ þ α1 λmax M Tj QBi P2 1 BTi QM j ð45Þ
sμ s
h h

V_ ðt Þr ∑ ∑ ρi ðθðt ÞÞρj ðθðt ÞÞ xT ðtÞΠ ij xðtÞ þ eTx ðtÞΩi ex ðtÞ þ 2xT P1 Bi ef ðtÞ
i¼1i¼1
2 2 1
þ 2xT P1 Bi K j ex ðt Þ eTf ðt ÞBTi QN j ex ðt Þ  eTf ðt ÞBTi QBj ef ðt Þ þ eTx ðt ÞP1 ex ðt Þ
 s s μ
2 T
þ ef ðt ÞP2 ef ðt Þ þ δ ð46Þ
sμ
where
h μ
δ ¼ max μα21 λmax M Ti QP1 1 QM i þ α22 λmax Γ  1 P2 1 Γ  1
i;j s
μ 2  i
1 T
þ α1 λmax M j QBi P2 Bi QM j
T
ð47Þ
s
The inequality (46) can be reformulated as follows:
h h
V_ ðtÞr x~ T ðtÞ ∑ ∑ ρi ðθðtÞÞρj ðθðtÞÞΞ ij x~ ðtÞ þ δ ð48Þ
i¼1i¼1

where
0 1
xðtÞ
B C
x~ ðtÞ ¼ @ ex ðtÞ A
ef ðtÞ
3114 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

and
0 1
Π ij P1 Bi K j P1 Bi
B Ωi þ μ1 P1  s1 BTi QN j C
Ξ ij ¼ @ n A ð49Þ
n n Υ ij

1 T 2
Υ ij ¼  B QBj þ BTj QBi þ P2 ð50Þ
s i sμ
Then, by taking into account Assumption A1 and if the following inequality holds
h h
∑ ∑ ρi ðθðtÞÞρj ðθðtÞÞΞ ij o0 ð51Þ
i¼1i¼1

We can obtain that

V_ ðtÞr  ε J x~ J 2 þ δ ð52Þ
where ε40 is given by
h h

ε ¼ min λmin  ∑ ∑ ρi ðθðtÞÞρj ðθðtÞÞΞ ij o0 ð53Þ
i¼1i¼1

which can also be bounded as follows:

ε r min λmin ð Ξ ij Þo0 ð54Þ
i;j

Then, V_ ðtÞo0 if εJ x~ J 2 4δ, 8 t Z 0 which means that the state x(t), the state estimation error ex(t)
and the fault estimation error ef(t) converge to a small set according to the Lyapunov stability
theory and lie in it.
To complete the proof by considering Eq. (51), let us introduce the following notations:
h
Z ξ ¼ ∑ ρi ðθðtÞÞZ i ð55Þ
i¼1

h h
Z ξξ ¼ ∑ ∑ ρi ðθðtÞÞρj ðθðtÞÞZ ij ð56Þ
i¼1i¼1

where Z ξ and Z ξξ are given matrices. By using these notations, the inequality (51) becomes
!
Π ξξ Dξξ
Δξξ ¼ ð57Þ
DTξξ Λξξ

with
Dij ¼ ðP1 Bi K j P1 Bi Þ ð58Þ
!
Ωi þ μ1 P1  s1 BTi QN j
Λij ¼ ð59Þ
n Υ ij
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3115

Consider a symmetric matrix X which is defined as

! !
P1 1 0 P1 1 0
X¼ ; X1 ¼ ð60Þ
0 X1 0 I

By considering that for any positive definite matrix P and for any full column rank matrix Q,
then QPQT is a positive definite matrix. Then, by post- and pre-multiplying the inequality (57) by
X , we can obtain that
!
P1 1 Π ξξ P1 1 P1 1 Dξξ X 1
ð61Þ
n X 1 Λξξ X 1

The term X 1 Λξξ X 1 can be replaced by considering the following inequality which holds for any
scalar β such that
1 T 1
ðX 1 þ βΛξξ Þ Λξξ ðX 1 þ βΛξξ Þr0
1
3X 1 Λξξ X 1 r  2βX 1  β2 Λξξ ð62Þ
Considering Eq. (62) and with the Schur Complement, the inequality (61) becomes
0 1 1
P1 Π ξξ P1 1 P1 1 Dξξ X 1 0
B n  2βX 1 βI C
@ Ao0 ð63Þ
n n Λξξ

Using the notations (55) and (56) and the definitions of the matrices Π ξξ , Dξξ and Λξξ given by
Eqs. (40), (58) and (59), by making the change of variables X ¼ P1 1 , Wi ¼ KiX, Si ¼ QLi, we can
obtain the inequalities given in Theorem 1 under equality constraint (31) which ends the
proof. □

It can be noticed that the conservatism introduced by the use of a common Lyapunov function
could be reduced by the use of parameter-dependent Lyapunov function so as to get others
solutions [30]. Another strategy based on Polya's Theorem [48] could also be used even if the
number of LMI to be solved will increase to reduce the conservatism.

6. Illustrative example

The proposed example considers a descriptor model of a two-phase flash system [49]
represented in Fig. 1 in which a volatile component flashes out of a dilute binary mixture. It is
assumed that the level control is nearly instantaneous and that the liquid and vapor phases are at
the same temperature. Since only one component is volatile, the gas phase contains the pure
volatile component. Accumulation of energy and matter in the gas phase is neglected because the
mass of liquid in the flash vessel is considerably larger than the mass of gas.
The continuous isothermal reactor can be modeled by using a LPV descriptor representation as
follows:
(
~
E x_ ðtÞ ¼ AðθðtÞÞxðtÞ ~
þ BðθðtÞÞuðtÞ
ð64Þ
yðtÞ ¼ CxðtÞ
3116 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

P Pref

P, T
LT

Q, TQ , x Q LC
x

QL

Fig. 1. Flash tank system.

where
2 3
 M w ðQL0  ðkm Aþθ1 ðtÞÞÞ M w ðk m Aþθ1 ðtÞÞ M w x0
0 0
6 ρV ρV ρV 7
6 7
6  ðk m Aþθ1 ðtÞÞðhv þθ2 ðtÞÞ M w Q L0 ðhv þθ2 ðtÞÞðk m Aþθ1 ðtÞÞ
0 MwT 0 7
6 ρVC p ρV ρVC p ρV 7
6 7
A~ ðθðt ÞÞ ¼ 6
6 0 0 1 1
H 0 7
7
6 b 7
6 0 0 1
P0 ln ð10Þ 0 7
6 ðT 0 þcÞ2 7
4 ðk m Aþθ1 ðtÞÞ2 ðhv þθ2 ðtÞÞ  ðkm Aþθ1 ðtÞÞM w QL0  ðk m Aþθ1 ðtÞÞ2 ðhv þθ2 ðtÞÞ  ðk m Aþθ1 ðtÞÞM w T 0
5
ρVC p ρV ρVC p 0 ρV

2 3 2  M w xQ 3
1 0 0 0 0 ρv
60 6 7 2 3
6 1 0 0 077 6
6
 MwT Q
ρv
7
7 0 0 1 0 0
6 7 6 7 6 7
E¼6
60 0 0 0 077; B~ ðθðt ÞÞ ¼ 6 0 7; C ¼40 0 0 1 05
6 7 6 7
40 0 0 0 05 6 0 7 0 0 0 0 1
4 5
ðk m Aþθ1 ðtÞÞM w T Q
0 0 0 0 0 ρv

where x1(t), x2 ðtÞ, x3 ðtÞ, x4 ðtÞ and x5 ðtÞ are the liquid mole fraction of volatile component, flash
tank temperature, equilibrium mole fraction, pressure in flash tank and liquid out-flow rate
respectively. So, we get ½x1 x2 x3 x4 x5 T ¼ ½xL T xn P QL T . The system parameters are listed in
Table 1.
Parameters kmA and hv are considered as varying variables denoted respectively as θ1 ðtÞ and
θ2 ðtÞ which vary such that θ1 A ½  0:018; 0:018 and θ2 A ½ 2:32; 2:32, l ¼ 2. As usually
done in LPV framework and as in system (1), these parameters are assumed to be available as in
Alwi et al. [24] or in Rodrigues et al. [30]. As there are two parameters which vary in this LPV
descriptor system (64), then 22 ¼ 4 models are considered as explained in Section 2.
The parameters evolution functions ρðθðtÞÞ vary within a convex set like in Eq. (4) and are depicted
in Fig. 2. In this case, the descriptor polytopic LPV representation (64) can be rewritten as follows:
8
> 4
< Ex_ ðtÞ ¼ ∑ ρ ðθðtÞÞðA xðtÞ þ B ðuðtÞ þ f ðtÞÞ
i i i
i¼1 ð65Þ
>
: yðtÞ ¼ CxðtÞ
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3117

Table 1
Parameters and values.

Variable Definition Values/units

State x(t)
xL(t) Liquid mole fraction of volatile component –
T(t) Flash tank temperature K
x⋆ðtÞ Equilibrium mole fraction –
P(t) Pressure in fash tank kPa
QL(t) Liquid out-flow rate mol/s

Input u(t)
Q(t) Feed flow rate 4377 mol/s
Parameters
ρV Mass of liquid 23.7 kg
kmA Mass-transfer coefficient 0.12 815% mol/s
hv Heat of vaporization 23.24 8 10% kJ/mol

Known constants
Cp Average heat capacity 4.2 kJ/K g
Mw Average molecular weight of liquid 50 g/mol
Cv Valve flow coefficient 16:8 ðkPaÞ  1=2 mol=s
V Liquid volume 19 m3
xQ Volatile component mole-fraction in feed 1
TQ Temperature of feed 500 K
H Henry's law constant 313 K
Pref Downstream pressure 10 KPa
b Antoine equation constant 130.63 1C  1
c Antoine equation constant 23.426 1C

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 2. Parameters evolutions ρi.

3118 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

The matrices of the system can be determined at the vertices of the polytope for extrema values of
parameters ρi. So, matrices Ai are defined as follows:
2 3
 0:8223 0 0:5279 0 1:0127
6  0:3011 0:6751 201:1814 7
6 0:3011 0 7
6 7
A1 ¼ 66 0 0 1  0:0484 0 7
7
6 7
4 0  0:0013 0 0:3967 0 5
0:2355  0:5279  0:2355 0  157:3239
2 3
 0:8223 0 0:5279 0 1:0127
6  0:3376 201:1814 7
6 0:6751 0:3376 0 7
6 7
A2 ¼ 6
6 0 0 1  0:0484 0 7
7
6 7
4 0  0:0013 0 0:3967 0 5
0:2640  0:5279  0:2640 0  157:3239
2 3
 0:7980 0 0:5522 0 1:0127
6  0:3150 201:1814 7
6 0:6751 0:3150 0 7
6 7
A3 ¼ 6
6 0 0 1  0:0484 0 7
7
6 7
4 0  0:0013 0 0:3967 0 5
0:2577  0:5522  0:2577 0  164:5664
2 3
 0:7980 0 0:5522 0 1:0127
6  0:3531 201:1814 7
6 0:6751 0:3531 0 7
6 7
A4 ¼ 6
6 0 0 1  0:0484 0 7 and
7
6 7
4 0  0:0013 0 0:3967 0 5
0:2889  0:5522  0:2889 0  157:3239
2 3
 1:2152
6  251:8143 7
6 7
6 7
6
Bi ¼ B ¼ 6 0 7
7
6 7
4 0 5
201:4515

The weighting functions ρi ðθðtÞÞ are defined as combinations of θj [39] and are given as follows:
θ1 ðtÞ θ 1 θ2 ðtÞ  θ 2 ðθ1 ðtÞ þ 0:018Þðθ2 ðtÞ þ 2:32Þ
ρ1 ð θ ð t Þ Þ ¼ ¼
θ1  θ1 θ2  θ2 0:167
θ1 ðtÞ θ 1 θ 2  θ2 ðtÞ ðθ1 ðtÞ þ 0:018Þð2:32 θ2 ðtÞÞ
ρ2 ð θ ð t Þ Þ ¼ ¼
θ1  θ1 θ2  θ2 0:167
θ 1  θ1 ðtÞ θ2 ðtÞ  θ 2 ð0:018  θ1 ðtÞÞðθ2 ðtÞ þ 2:32Þ
ρ3 ð θ ð t Þ Þ ¼ ¼
θ1  θ1 θ2  θ2 0:167
θ 1  θ1 ðtÞ θ 2  θ2 ðtÞ ð0:018  θ1 ðtÞÞð2:32  θ2 ðtÞÞ
ρ4 ð θ ð t Þ Þ ¼ ¼
θ1  θ1 θ2  θ2 0:167
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3119

Remark. Note that parameters kmA and hv can be measured by the distinct ways. For the
measurement of hv (enthalpy of vaporization), let us consider that the change of state of a pure
substance is made with constant pressure P and constant temperature T. The heat of the reaction
corresponds to a change of enthalpy hv since the pressure is constant. With constant pressure P, it
is called latent heat of state change. The molar enthalpy change of state or mass enthalpy change
of state corresponds to the amount of heat required per unit of amount of substance (mol) or mass
(kg) of body so it changes state. For example, for the passage from the liquid state to the vapor
state, one speaks about vaporization enthalpy (or latent heat of vaporization). The latent heat or
enthalpy (in Joules) can be expressed as follows:
Δhv ¼ nC p ΔT
Δhv : variation of the heat of vaporization
ΔT : variation of temperature
n : number of moles
C p : average heat capacity

The molar latent heat or molar enthalpy (in Joules/mol) is given by

Δhv
Δhmolar
v ¼
n
By the way, we measure hv by measuring the temperature variations ΔT (with constant pressure).
The physical measurement method of the coefficient of mass transfer kmA consists of
measuring the oxygen concentration Ce before the entry of the reactor, and the oxygen
concentration inside the reactor CL. The liquid phase in the reactor is assumed to be perfectly
mixed. The coefficient kmA can be deduced from the following expression:
km A:VðC n  CL Þ ¼ QL ðCL  Ce Þ
C n : saturation of oxygen concentration in the liquid
C L : oxygen concentration inside the reactor
C e : oxygen concentration before entering in the reactor
QL : liquid outflow rate
V : liquid volume

6.1. Fault tolerant control design for a two-phase flash system

Let us consider an additive actuator fault signal f(t) affecting the polytopic LPV descriptor
system (6) which is defined as follows:
f ðtÞ ¼ 0; to15 s
f ðtÞ ¼ 25 sin ð2:5tÞ; 15 sr to25 s
f ðtÞ ¼ 15; 25 sr to35 s
f ðtÞ ¼ 0; t435 s

The observer based control law given by Eqs. (11) is designed by solving the LMI problem
defined in Theorem 1. One can check that the necessary Assumptions (A2) and (A3) are verified.
3120 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

Afterwards and according to the proposed methodology defined in Section 4, matrices T1 and T2
can be computed from Eq. (24). The gains matrices of the APO and the controllers are obtained
by solving the LMIs (29) with parameter value δ ¼ 1442:
2 3
 0:7905  0:7147  2:1275
6  174:7141  231:1559  537:6705 7
6 7
6 7
6
L1 ¼ 6  12:8643  0:8337  0:7042 77;
6 7
4  0:3478  12:4030  0:2258 5
 233:1793  222:3492  118:8419
2 3
 1:5590  1:5266  2:7960
6  437:1092  507:7838  786:0505 7
6 7
6 7
L2 ¼ 6
6  13:0101  0:9684  0:8103 7 7
6 7
4  0:3173  12:3811  0:1930 5
 213:1735  202:8134  115:9478
2 3
 5:5749 0:3908  4:6427
6  307:8523  332:0796  549:2024 7
6 7
6 7
L3 ¼ 6
6  14:1713  2:8035  2:1476 7 7;
6 7
4  3:0905  13:3019  2:4077 5
 122:3051  68:0098  94:6731
2 3
 0:0586  1:0962  1:4895
6  163:1733  149:9693  511:7788 7
6 7
6 7
L4 ¼ 6
6  12:6399  0:6354  0:3416 7 7
6 7
4  1:0167  12:8870  2:1877 5
 201:5105  114:0819  93:7544
K 1 ¼ ½ 0:0382  0:0429 0:0250 0 0:4862
K 2 ¼ ½ 0:0424  0:0433 0:0281 0 0:4856
K 3 ¼ ½ 0:0395  0:0430 0:0263 0 0:4860
K 4 ¼ ½ 0:0439  0:0435 0:0295 0 0:4853
U ¼ ½ 0:8457 2:9547 0:5406; for i ¼ 1; …; 4

Simulations have been realized by applying a random noise with maximal magnitude 0.01 in
the output measurements. The actuator fault and its estimate are depicted in Fig. 3. One can see
the good estimation of this time-varying actuator fault by the use of the APO despite an additive
noise. This actuator fault estimation is also used in the control law given in Eq. (10) so as to
make the system robust against this actuator fault. Moreover, it should be noticed that the fault
estimation error ef(t) depicted in Fig. 4 is zero-mean that underlines an accurate fault magnitude
estimation. Note that abrupt changes of the fault can generate some peaks in Fig. 4 at time 15 s,
25 s and 35 s. Similar peaks have been noticed in the example part of Rodrigues et al. [30] for the
same reasons.
Figs. 5–7 illustrate a comparison between the outputs of the nominal model (i.e., without any
fault), the outputs of the faulty system without FTC (with a classical control law by a state
feedback) and finally the outputs with our proposed FTC. It can be noticed that the outputs
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3121

30
original actuator fault estimated actuator fault

20

Q(t) (mol/s) 10

−10

−20

−30
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 3. Actuator fault and its estimated.

8
f(t) − f(t) estimated

2
ef(t)

−2

−4

−6
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 4. Fault estimation error ef(t).

without FTC do not converge to the nominal dynamic: it underlines that the system is perturbed
by the actuator fault and it is not robust against such faults. The proposed APO under the Fault
Tolerant control law (10) makes the system robust against actuator fault since the outputs's
trajectories of the system with FTC reach the outputs of nominal model.
For comparison, the nominal state feedback controller (without taking into account faults
occurrences) and the proposed FTC control are plotted simultaneously in Figs. 5–7. The FTC
scheme can well accommodate the actuator fault. Here, the Adaptive Polytopic Observer shows
good results for the estimation of both time-varying or abrupt actuator fault in spite of the
3122 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

0.01

0.008

0.006

0.004

0.002

−0.002

−0.004

−0.006

−0.008

−0.01
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 5. y1 ðtÞ of the system: nominal output, output without FTC and output with FTC.

0.2

0.15

0.1

0.05
P(t) (KPa)

−0.05

−0.1

−0.15

−0.2
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 6. y2 ðtÞ of the system: nominal output, output without FTC and output with FTC.

presence of an additive noise. An extension of this paper should consider Fault-Tolerant Control
for Markovian systems as in Liu et al. [50] with our FTC strategy.

7. Conclusion

In this paper, an actuator Fault Tolerant Control methodology to address polytopic LPV
descriptor system has been studied. The FTC scheme is based on an Adaptive Polytopic
Observer that is able to simultaneously estimate time varying faults and state variables with a
good accuracy. The stability analysis has been formulated and solved within a set of linear matrix
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3123

60

50

40

30
QL(t) (mol/s)
20

10

−10

−20

−30

−40
0 5 10 15 20 25 30 35 40 45 50
t(s)

Fig. 7. y3 ðtÞ of the system: nominal output, output without FTC and output with FTC.

inequalities under equalities constraints. The developed scheme has been applied to a two-phase
flash system with an additive actuator fault so as to illustrate the effectiveness of this method.

References

[1] J. Chen, R. Patton, Robust Model-based Fault Diagnosis for Dynamic Systems, Kluwer Academic Publishers, 1999.
[2] J. Bokor, Z. Szabo, Fault detection and isolation in nonlinear systems, Annu. Rev. Control 33 (2009) 113–123.
[3] M. Blanke, M. Kinnaert, J. Lunze, M. Staroswiecki, Diagnosis and Fault Tolerant Control, 2nd ed., Springer,
Berlin, Germany, 2006.
[4] Y. Zhang, J. Jiang, Bibliographical review on reconfigurable fault-tolerant control systems, Annu. Rev. Control 32
(2008) 229–252.
[5] Z. Gao, S.X. Ding, Actuator fault robust estimation and fault-tolerant control for a class of nonlinear descriptor
systems, Automatica 43 (2007) 912–920.
[6] Shanbin Li, Dominique Sauter, Christophe Aubrun Joseph Yamé, Stability guaranteed active fault-
tolerant control of networked control systems, J. Control Sci. Eng. vol. 2008, Article ID 189064, 9 p. 2008.
http://dx.doi.org/10.1155/2008/189064.
[7] D. Theilliol, H. Noura, J. Ponsart, Fault diagnosis and accommodation of three-tank system based on analytical
redundancy, ISA Trans. 41 (2002) 365–382.
[8] Q. Li, Z. Ren, S. Dai, W. Wang, A new robust fault-tolerant controller for self-repairing flight control system, J.
Frankl. Inst. 350 (2013) 2509–2518.
[9] Z. Mao, B. Jiang, P. Shi, Observer based fault-tolerant control for a class of nonlinear networked control systems,
J. Frankl. Inst. 347 (2010) 940–956.
[10] M. Witczak, L. Dziekan, V. Puig, J. Korbicz, Design of a fault-tolerant control scheme for takagi-sugeno fuzzy
systems, in: The 16th Mediterranean Conference on Control and Automation, MED'08, Ajaccio, 2008.
[11] D. Ichalal, B. Marx, J. Ragot, D. Maquin, New fault tolerant control strategies for nonlinear Takagi–Sugeno
systems, Int. J. Appl. Math. Comput. Sci. 22 (2012) 197–210.
[12] H. Wang, K. Tanaka, M. Griffin, An approach to fuzzy control of nonlinear systems: stability and design issues,
IEEE Trans. Fuzzy Syst. 4 (1996) 14–23.
[13] A. Nagy Kiss, B. Marx, G. Mourot, G. Schutz, J. Ragot, State estimation of two-time scale multiple models.
Application to wastewater treatment plant, Control Eng. Pract. 19 (2011) 1354–1362.
3124 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125

[14] F.L. Lewis, A survey of linear singular systems, Circuits Syst. Signal Process. 5 (1986) 3–36.
[15] S. Mattson, H. Elmqvist, M. Otter, Physical system modeling with modelica, Control Eng. Pract. 6 (1998) 501–510.
[16] L. Dai, Singular Control Systems, Springer, Germany, 1989.
[17] G.-R. Duan, Analysis and Design of Descriptor Linear Systems Advances in Mechanics and Mathematics, Springer, New
York, 2010.
[18] M. Darouach, M. Boutayeb, Design of observers for descriptor systems, IEEE Trans. Autom. Control 75 (1995)
490–501.
[19] A. Youssouf, M. Kinnaert, Observer based residual generator for singular systems, in: Proceedings of the 35th
Conference on Decision and Control, Kobe, Japan, 1996, pp. 1175–1180.
[20] C.-M. Astorga-Zaragoza, D. Theilliol, J.-C. Ponsart, M. Rodrigues, Observer synthesis for a class of a descriptor
LPV systems, in: American Control Conference (ACC'2011), Etats-Unis, USA, 2011.
[21] Z. Wang, Y. Shen, X. Zhang, Q. Wang, Observer design for discrete-time descriptor systems: an LMI approach,
Syst. Control Lett. 61 (2012) 683–687.
[22] M. Chadli, M. Darouach, Novel bounded real lemma for discrete-time descriptor systems: application to h-infinity
control design, Automatica 48 (2012) 449–453.
[23] M. Chadli, M. Darouach, Further enhancement on robust hinfinity control design for discrete-time singular systems,
IEEE Trans. Autom. Control 59 (2) (2014) http://dx.doi.org/10.1109/TAC.2013.2273266. Date of Publication:
February 2014.
[24] H. Alwi, C. Edwards, A. Marcos, Fault reconstruction using a LPV sliding mode observer for a class of LPV
systems, J. Frankl. Inst. 349 (2012) 510–530.
[25] J. Bokor, G. Balas, Detection filter design for LPV systems: a geometric approach, Automatica 40 (2004) 511–518.
[26] S. Armeni, A. Casavola, E. Mosca, Robust fault detection and isolation for LPV systems under a sensitivity
constraint, Int. J. Adapt. Control Signal Process. 23 (2009) 55–72.
[27] F. Wu, Control of Parameter Varying Systems, University of California at Berkeley, USA, 1995.
[28] L. Wu, Z. Wang, H. Gao, h1 and l2  l1 filtering for two-dimensional linear parameter-varying systems, Int. J.
Robust Nonlinear Control 17 (2007) 1129–1154.
[29] F. Bruzelius, Linear Parameter-Varying Systems an Approach to Gain Scheduling, Department of Signals and
Systems, Chalmers University of Technology, Goteborg, Sweden, 2004.
[30] M. Rodrigues, M. Sahnoun, D. Theilliol, J.-C. Ponsart, Sensor fault detection and isolation filter for polytopic LPV
systems: a winding machine application, J. Process Control 23 (2013) 805–816.
[31] L. Wu, W. Zheng, Passivity-based sliding mode control of uncertain singular time-delay systems, Automatica 45
(2009) 2120–2127.
[32] L. Wu, P. Shi, H. Gao, State estimation and sliding-mode control of Markovian jump singular systems, IEEE Trans.
Autom. Control 55 (2010) 1213–1219.
[33] L. Wu, X. Su, P. Shi, Sliding mode control with bounded l2 gain performance of Markovian jump singular time-
delay systems, Automatica 48 (2012) 1929–1933.
[34] F. Li, X. Zhang, Delay-range-dependent robust h1 filtering for singular LPV systems with time variant delay, Int. J.
Innov. Comput. Inf. Control 9 (2013) 339–353.
[35] H. Hamdi, M. Rodrigues, C. Mechmeche, N.B. Braiek, Robust fault detection and estimation for descriptor systems
based on multi-models concept, Int. J. Control Autom. Syst. 10 (2012) 1260–1266.
[36] B. Marx, D. Koenig, D. Georges, Robust fault tolerant control for descriptor systems, IEEE Trans. Autom. Control
49 (2004) 1869–1876.
[37] D. Koenig, Observers design for unknown input nonlinear descriptor systems via convex optimization, IEEE Trans.
Autom. Control 51 (2006) 1047–1052.
[38] S.M.D. Oca, D. Rotondo, F. Nejjari, V. Puig, Fault estimation and virtual sensor FTC approach for LPV systems, in: 50th
IEEE Conference on Decision and Control and European Control Conference (CDC-ECC), CD-Rom, Orlando, USA,
2011.
[39] H. Hamdi, M. Rodrigues, C. Mechmeche, D. Theilliol, N.B. Braiek, Fault detection and isolation in linear
parameter-varying descriptor systems via proportional integral observer, Int. J. Adapt. Control Signal Process. 26
(2012) 224–240.
[40] M. Rodrigues, H. Hamdi, D. Theilliol, C. Mechmeche, N.B. Braiek, Fault diagnosis based on adaptive polytopic
observer for LPV descriptor systems, in: Proceedings of 8th IFAC Symposium SAFEPROCESS: Fault Detection,
Supervision and Safety for Technical Processes, CD-Rom, Mexico City, Mexico, 2012.
[41] H. Wang, S. Daley, Actuator fault diagnosis: an adaptive observer-based technique, IEEE Trans. Autom. Control 41
(1996) 1073–1078.
 M. Rodrigues et al. / Journal of the Franklin Institute 351 (2014) 3104–3125 3125

[42] K. Zhang, B. Jiang, V. Cocquempot, Adaptive observer-based fast fault estimation, Int. J. Control Autom. Syst. 6
(2008) 320–326.
[43] M. Rodrigues, D. Theilliol, D. Sauter, Design of an active fault tolerant control for nonlinear systems described by a
multi-model representation, in: Proceedings of the 20th IEEE International Symposium on Intelligent Control
(ISIC'05) and the 13th Mediterranean Conference on Control and Automation (MED'05), Limassol, Cyprus.
[44] M. Rodrigues, D. Theilliol, S. Aberkane, D. Sauter, Fault tolerant control design for polytopic LPV system, Int. J.
Appl. Math. Comput. Sci. 17 1 (2007) 27–37.
[45] C.-M. Astorga-Zaragoza, D. Theilliol, J.C. Ponsart, M. Rodrigues, Fault diagnosis for a class of descriptor linear
parameter-varying systems, Int. J. Adapt. Control Signal Process. 26 (2012) 208–223.
[46] G. Cai, C. Hu, G. Duan, Efficient LMI-based quadratic stability and stabilization of parameter-dependent interval
systems with applications, Int. J. Innov. Comput. Inf. Control 8 (2012) 1943–1954.
[47] D. Zhang, H. Wang, B. Lu, Z. Wang, LMI-based fault detection fuzzy observer design with multiple performance
constraints for a class of non-linear systems: comparative study, Int. J. Innov. Comput. Inf. Control 8 (2012)
633–645.
[48] A. Sala, C. Arino, Asymptotically necessary and sufficient conditions for stability and performance in fuzzy control:
applications of Polya's theorem, Fuzzy Sets Syst. 158 (2007) 2671–2686.
[49] A. Ben-Zvi, P.J. McLellan, K.B. McAuley, Identifiability of non-linear differential algebraic systems via a
linearization approach, Can. J. Chem. Eng. 84 (2006) 590–596.
[50] M. Liu, P. Shi, L. Zhang, X. Zhao, Fault-tolerant control for nonlinear Markovian jump systems via proportional and
derivative sliding mode observer technique, IEEE Trans. Circuits Syst. Part I 58 (2011) 2755–2764.