International Journal of Control, Automation and Systems 16(X) (2018) 1-11 ISSN:1598-6446 eISSN:2005-4092

http://dx.doi.org/10.1007/s12555-017-0546-8 http://www.springer.com/12555

Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno De-

scriptor Nonlinear Systems with Sensor and Actuator Faults
Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji*, and Mohamed Chaabane

Abstract: This paper concerns the problem of state, fault estimation (FE) and Fault Tolerant Control (FTC) of
Takagi-Sugeno (T-S) descriptor systems affected by sensor, actuator and external disturbances simultaneously. An
Adaptive Fuzzy Observer is firstly proposed to achieve a simultaneous estimation of descriptor system states, actu-
ator and sensor faults by using the H∞ optimization technique. A FTC is secondly proposed to stabilize the faulty
descriptor system. Based on Lyapunov method, stability analysis and design conditions of the resulting closed-loop
system are formulated in a set of Linear Matrices Inequalities (LMIs). The adaptive fuzzy observer and the FTC are
independently designed, in order to avoid the coupling problem. Accordingly, the observer and controller gains are
computed separately by solving a set of LMIs and then used to estimate the unmeasured states, sensor and actuator
faults at the same time. Finally, a truck-trailer system application is given to illustrate the validity of the proposed
approach.

Keywords: Adaptive fuzzy observer, T-S descriptor systems, Actuator/sensor faults estimation, FTC, LMI, Lya-
punov functional.

1. INTRODUCTION obtained fault information, an additional controller can be

designed to compensate the faults. More precisely, FTC
An important number of industrial applications such as guarantees the stability of a closed-loop system even in
electrical power lines and manufacturing processes can the presence of component malfunctions. Besides main-
have an undesirable behavior because of sensor, actuator taining stability properties, it could also keep desirable
and/or process faults or even external disturbances. Gen- performances.
erally speaking, to overcome component malfunctions and FTC approach has been firstly adapted for linear sys-
to maintain a certain level of safety, reliability and per- tems (see [1, 11] and references therein). However, most
formance efficiency of a dynamic system, techniques and realistic engineering systems have nonlinear behaviors. It
tools of fault detection and isolation (FDI), fault estima- is well known that T-S fuzzy representation is a good way
tion (FE) and fault tolerant control (FTC) have been es- to approximate a large class of nonlinear dynamic sys-
tablished (see [1–7]). tems [12]. T-S fuzzy models are nonlinear systems rep-
FDI strategy is used to monitor whether a fault occurs resented by a set of local linear models. By fuzzy blend-
and in which component it is occurred. However, it is ing of the linear system representations the overall fuzzy
generally difficult in practical systems to have the exact model of the system is achieved, which greatly facilitates
information of the size of faults from a FDI strategy only. observer/controller synthesis for complex nonlinear sys-
Accordingly, considerable attention has been devoted to tems. One of the primary advantage is that it offers an
FE; it provides further accurate information of the fault, effective and simple design strategy to represent a nonlin-
such as shape, size and duration (see [8–10] and references ear system. In the literature, many important results have
therein). been reported in [13–15] and [16].
Furthermore, FE plays an important role in FTC, which Consequently, many researchers have interested to FTC
is developed to preserve overall system stability as well as approach for T-S fuzzy systems (see [17–19]). In [20] for
acceptable performance. It possesses the ability to accom- example, authors have extended the adaptive observer pro-
modate component failures automatically. By using the posed in [21] to FTC of T-S fuzzy models.

Manuscript received September 6, 2017; revised November 23, 2017; accepted December 8, 2017. Recommended by Editor Jessie (Ju H.)
Park.

Dhouha Kharrat, Hamdi Gassara, and Ahmed El Hajjaji are with Modeling, Information, and Systems Laboratory, University of Pi-
cardie Jules Verne, UFR of Sciences, 33 Rue St Leu Amiens 80000, France (e-mails: [email protected], [email protected],
[email protected]). Dhouha Kharrat, Hamdi Gassara, and Mohamed Chaabane are with Laboratory of Sciences and Techniques of
Automatic control & computer engineering, University of Sfax, ENIS PB 1173, 3038 Sfax, Tunisia (e-mails: [email protected],
[email protected], [email protected]).
* Corresponding author.

⃝ICROS,
c KIEE and Springer 2018
2 Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji, and Mohamed Chaabane

The problem of state/fault estimation for T-S fuzzy sys- element in the symmetric position of a matrix and sym(A)
tems affected by sensor and actuator faults under external signifies A + AT . λmax (A) and λmin (A) represent the max-
disturbances, has been investigated in [8]. In this latter, imum and minimum eigenvalues of A, respectively. A†
a fuzzy reduced-order observer has been designed which stands for the generalized inverse of A and ∀ denotes "for
can estimate system states, sensor and actuator faults si- all".
multaneously. Then a FTC has been developed for stan-
dard T-S fuzzy systems. 2. PROBLEM FORMULATION
Recently, T-S fuzzy model has been extended to deal
with descriptor nonlinear systems, many scholars have Consider a T-S fuzzy descriptor system composed of a
been interested to T-S descriptor systems as well as FE set of If-Then rules. The ith rule of the system is given as
and FTC (see for instance [22–26]). It is well known follows:
that descriptor systems resulting from natural modeling Plant Rule i(i = 1, 2, .., r): If θ1 is µi1 and, · · · , and θ p
approaches, can represent a wide class of practical sys- is µip , Then
tems including electrical circuits, robotic systems, aircraft 

 E ẋ(t) = Ai x(t) + Bi (u(t) + fa (t)) + Di d(t),
systems and chemical process (see for example [27–30]
y(t) = Cx(t) + F fs (t), (1)
and references therein). It should be pointed that the sta- 

bility and stabilization problems for descriptor systems are z(t) = CLi x(t).
much more difficult than for regular ones because it affects where θ j (x(t)) are the premise variables which are sup-
the regularity and impulse-free problems [15, 31, 32]. posed to be measurable, µi j (i = 1, · · · , r, j = 1, · · · , p) are
Reference [23] is an existing work for both state and ac- the fuzzy sets which are characterized by the membership
tuator faults reconstruction in which a FTC has been de- functions, r and p are the total number of If-Then rules
signed to stabilize the closed-loop T-S descriptor system and the premise variables, respectively. x(t) ∈ Rn is the
in presence of actuator faults. The work in [3] deals with state vector, u(t) ∈ Rm is the control input, y(t) ∈ R p is
the problem of simultaneous sensor and actuator fault es- the measured output, d(t) ∈ Rν is the external disturbance,
timation for descriptor linear parameter varying systems. fa (t) ∈ Rm and fs (t) ∈ Rw represent actuator and sensor
However the FTC approach and the stabilization analysis fault, respectively. They can be constant or time-varying
have not been discussed. function. z(t) ∈ R p1 is the controlled output. Matrix
There are, so far as the authors know, no works deal- E ∈ Rn×n is assumed to be singular and we suppose that
ing with adaptive observer-based FTC for T-S descriptor rank(E) = q ≤ n. Ai , Bi , Di ,C, F and CLi are known real
systems considering simultaneously actuator, sensor faults constant matrices of appropriate dimensions. We assume
and external disturbances. Motivated by the above obser- that p ≥ m + w, (Ai , Bi ) are controllable, rank(Bi ) = m,
vations, an adaptive observer allowing to estimate both rank(F) = w, rank([C, F]) = p.
states and sensor/actuator faults is developed. Then an By fuzzy blending, the overall fuzzy system is given as
observer-based FTC is introduced to preserve the stability follows:
of the closed-loop faulty system. Using the H∞ optimiza- r
tion technique, we show that the adaptive observer and the E ẋ(t) = ∑ hi (θ (x(t)))
controller could be independently designed and their gains i=1

computed separately by solving a set of LMIs using LMI × [Ai x(t) + Bi (u(t) + fa (t)) + Di d(t)], (2)
Toolbox or Yalmip of MATLAB software [33, 34]. y(t) = Cx(t) + F fs (t), (3)
The reminder of this paper is structured as follows: In r
the second section an overview of the T-S fuzzy descrip- z(t) = ∑ hi (θ (x(t)))[CLi x(t)], (4)
tor systems is given and some preliminaries are provided. i=1

Section 3 includes the main results. An adaptive fuzzy ob- in which

server is proposed to estimate system states, actuator and
θ (x(t)) = [θ1 (x(t)), ..., θ p (x(t))],
sensor faults simultaneously. Sufficient conditions are ex-
pressed in the form of LMI for the observer-based FTC. νi (θ (x(t)))
hi (θ (x(t))) = r , νi (θ (x(t)))
A simulation example is given in Section 4 to illustrate
the validity of the proposed approach. Finally, Section 5
∑ ν i (θ (x(t)))
i=1
concludes this contribution. p
Notations: The following notations are considered. = ∏ µi j (θi (x(t))),
A ∈ Rn indicates the n-dimensional Euclidean space, j=1

whereas A ∈ Rn×p denotes the set of all n × p real ma- where µi j (θi (x(t)) is the membership degree of θi (x(t))
trices. A real symmetric positive definite matrix (respec- in µi j . It is evident that 0 ≤ hi (θ (x(t))) ≤ 1 and
r
tively, negative definite matrix) is represented by A > 0
(respectively, A < 0). Notation (∗) signifies the transposed ∑ hi (θ (x(t))) = 1.
i=1
 Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno Descriptor Nonlinear Systems with Sensor and ... 3

In the following, for briefness we get hi to stand for 3. MAIN RESULTS

hi (θ (x(t))).
Before giving the main results, the following three as- 3.1. Augmented system
sumptions and two lemmas are needed. Using the descriptor approach and by taking sensor
fault fs as an auxiliary state, an augmented system can
Assumption 1 [35]: System (E, Ai ,C) is observable,
be constructed. The faulty system given by (2- 4) can be
[ ] rewritten as follows:
E
rank = n, (5) 
C  r

 Ē ˙
x̄ (t) = ∑ hi [Āi x̄(t) + Bi (u(t) + fa (t)) + Di d(t)],



 i=1
and
y(t) = C̄x̄(t),
[ ] 

sE − Ai 
 r
= n, ∀s ∈ C, Re(s) ≥ 0, ∀i = [1, · · · , r]. 
rank
C  z(t) = ∑ hi [CLi x(t)].

i=1
(6) (12)

Assumption 2 [36, 37] : where

[ ] [ ] [ ]
Ē E 0 x(t) [ ] [ ]
rank = rank = n + w. (7) x̄(t) = , Ē = E 0n×w , Āi = Ai 0n×w ,
C̄ C F fs (t)
[ ]
C̄ = C F .
Assumption 3 [20]: Actuator fault fa (t) and sensor
fault fs (t) satisfy ∥ fa (t)∥ ≤ αa and ∥ fs (t)∥ ≤ αs , respec- 3.2. Design of adaptive fuzzy observer
tively. The derivative of fa (t) and fs (t) with respect to
time are norm bounded i.e. ∥ f˙a (t)∥ ≤ famax and 0 ≤ αa , In this part, an adaptive fuzzy observer is proposed to a
famax < ∞ and ∥ f˙s (t)∥ ≤ fsmax and 0 ≤ αs , fsmax < ∞. simultaneous estimation of descriptor system states, actu-
ator and sensor faults for system (12).
Lemma 1 [38]: For a symmetric positive definite ma-
trix R and a scalar µ ∈ R+ we have the following inequal-  r

ity 



ẇ(t) = ∑ hi [T Āi x̄ˆ(t) + T Bi (u(t) + fˆa (t))

 i=1



 + Li (y(t) − ŷ(t))],
2u⊺ v ≤
1 ⊺
u Ru + µ v⊺ R−1 v , u, v ∈ Rn . 

(8)  x̄ˆ(t) = w(t) + Hy(t),
µ
(13)

 ey (t) = y(t) − ŷ(t),


Lemma 2 [15, 20]: Given a negative definite matrix 
 ŷ(t) = C̄x̄ˆ(t),


ϒ < 0. Consider a matrix Y of appropriate dimension such 


 r
that Y T ϒ Y < 0, then ∃ λ > 0 such that 
 f˙ˆa (t) = Γ ∑ hi Ni (ėy (t) + σ ey (t)),
i=1
−1
Y ϒ Y ≤ −λ (Y +Y ) − λ ϒ
T T 2
. (9)
where w(t) ∈ Rn+w and x̄ˆ(t) ∈ Rn+w are the observer state
Remark 1: Assumption 2 implies that there exists a and the estimation of state vector, respectively. ŷ(t) ∈ R p
full-row rank matrix is the estimation of the output vector. ey (t) ∈ R p is the
output estimation error and fˆa (t) ∈ Rm is the estimated of
[ ]† the actuator fault fa (t), and T , H, Li and Ni are gain ma-
[ ] Ē
T H =
C̄
(10) trices with appropriate dimensions to be solved. σ ∈ R is
a positive scalar. Γ ∈ Rm×m is the learning rate which is
such that chosen to be symmetric, positive definite matrix and set to
balance the convergence speeds of the states and actuator
[ ]
[ ] Ē fault. Under assumption 2, there exist nonsingular matri-
T H = In+w . (11) ces T ∈ Rn+w×n and H ∈ Rn+w×p such that
C̄

It also guarantees the impulse observability of the triple T Ē + HC̄ = In+w . (14)
matrix (Ē, Āi , C̄), ∀i ∈ [1, · · · , r]. It should be noted that the
designed observer (13) require necessary conditions intro- State and fault estimation errors are given as follows:
duced in Assumptions 1 and 2. Similar assumptions can
be also found in [35] and [29] and the references therein. ex (t) = x̄(t) − x̄ˆ(t), e f (t) = fa (t) − fˆa (t).
4 Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji, and Mohamed Chaabane
∫ ∞
≤ γ12 d T (s)M3 d(s) ds, (20)
0

if there exist symmetric positive definite matrix P1 and

positive definite matrices Ni and M such that ∀i ∈ [1, · · · , r]
the following conditions hold:

(T Bi )T P1 = NiC̄, i = 1, 2, · · · , r, (21)
ϕi j + ϕ ji < 0, i, j = 1, 2, · · · , r, i ≤ j, (22)

where
 
φi11 φi12j φi13
Fig. 1. Adaptive fuzzy observer-based fault tolerant con- ϕi j =  ∗ φi22j φi23j , (23)
trol scheme. ∗ ∗ −ρ1 M3

in which
By taking into account (12), (13) and by using relation
(14), estimation error dynamics ex (t) and output estima- φi11 = sym(P1 T Āi −YiC̄) + M1 ,
tion error ey (t) are given by: 1
φi12j = − (ĀTj T T P1 T Bi − C̄T Y jT T Bi ),
σ
r
ėx (t) = ∑ hi [(T Āi − LiC̄)ex (t) + T Bi e f (t) φi13 = P1 T Di ,
i=1 M 1
φi22j = − sym((T Bi )T P1 (T B j )) + M2 ,
+ T Di d(t)], (15) σµ σ
ey (t) = C̄ex (t). (16) 1
φi23j = − (T Bi )T P1 T D j ,
σ
Matrices T and H can be found simultaneously from (10).
In contrast to the constant fault giving in [21] and The observer gains can then be computed from Li = P1−1Yi .
[38]. In this work, time-varying faults are considered (i.e.
̸ 0). Consequently, the dynamic of fault estimation
f˙(t) = Proof: Taking the following Lyapunov function:
error is given by the following expression:
1 T
V1 (t) = eTx (t)P1 ex (t) + e (t)Γ−1 e f (t).
ė f (t) = f˙a (t) − f˙ˆa (t), (17) σ f

Then The time derivative of V1 (t) can be shown to be

r
V̇1 (t) =ėTx (t)P1 ex (t) + eTx (t)P1 ėx (t)
ė f (t) = f˙a (t) − Γ ∑ hi Ni (ėy (t) + σ ey (t)). (18)
2
i=1
+ eTf (t)Γ−1 ė f (t). (24)
σ
By taking into consideration (15) and (16), one can obtain
( Considering Lemma 1 we have
r r [
ė f (t) = fa (t) − Γ ∑ ∑ hi h j NiC̄ (T Ā j − L jC̄)ex (t)
˙ 2 T 1 T
i=1 j=1 e f (t)Γ−1 f˙a (t) ⩽ e (t)Me f (t)
) σ σµ f
] µ
+ T B j e f (t) + T D j d(t) + σ ex (t) . (19) + f˙aT (t)Γ−1 M −1 Γ−1 f˙a (t), (25)
σ
2 T 1 T
3.3. Stability and stabilization analysis e (t)Γ−1 f˙a (t) ⩽ e (t)Me f (t) + δ , (26)
σ f σµ f
3.3.1 Stability analysis of the adaptive fuzzy observer
Theorem 1: Consider system (12) under assumptions where
1−3. Given a real positive scalar γ1 > 0, two scalar tuning µ 2
δ= f λmax (Γ−1 M −1 Γ−1 ). (27)
parameters σ and µ > 0, and definite positive matrices σ amax
M1 , M2 and M3 , the adaptive fuzzy observer can ensure
By substituting (15), (19) into equation (24) and by con-
the asymptotic stability of the error dynamics under H∞
sidering (26), one can obtain
performance level γ1 , that is
∫ ∞
r r

(eTx (s)M1 ex (s) + eTf (s)M2 e f (s)) ds V̇1 (t) ≤ ∑ ∑ hi h j {eTx (t)[P1 (T Āi − LiC̄)
0 i=1 j=1
 Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno Descriptor Nonlinear Systems with Sensor and ... 5

+ (T Āi − LiC̄)T P1 ]ex (t) + 2eTx (t)P1 T Bi e f (t) − γ12 d T (t)M3 d(t) ≤ 0,
+ 2eTx (t)P1 T Di d(t) − 2eTf (t)NiC̄ex (t) for ζ ∥ξ (t)∥2 > δ , (34)
2
− eTf (t)NiC̄(T Ā j − L jC̄)ex (t) when d(t) = 0, (34) means V̇1 (t) ≤ 0 for ζ ∥ξ (t)∥2 > δ
σ
1 and under the Lyapunov stability theory, ξ (t) will con-
− eTf (t)sym(NiC̄T B j )e f (t) verge to a small set Ψ = {ξ (t)/∥ξ (t)∥2 ≤ ζδ }; thus ξ (t) is
σ
1 T uniformly bounded in the case of d(t) = 0.
+ e (t)Me f (t) When d(t) ̸= 0, integrating both sides of (34) with re-
σµ f
2 spect to t over time period [0 ∞] yields
− eTf (t)NiC̄T D j d(t) + δ }. (28) ∫ ∫
σ ∞ ∞
V̇1 (s)ds + eTx (s)M1 ex (s)ds
Taking into account equation (21), one can obtain 0
∫ 0
∫
∞ ∞
r r + eTf (s)M2 e f (s)ds − γ12 d T (s)M3 d(s)ds
V̇1 (t) ≤ ∑ ∑ hi h j {eTx (t)sym(P1 (T Āi − LiC̄))ex (t) 0 0

i=1 j=1 ≤ 0, for ζ ∥ξ (t)∥2 > δ . (35)

1 T
+ 2eTx (t)P1 T Di d(t) + e (t)Me f (t) As V1 (∞) ≥ 0, and with zero initial condition V1 (0) = 0,
σµ f
one obtains
2 T
− e (t)(T Bi )T P1 (T Ā j − L jC̄)ex (t) ∫ ∞ ∫ ∞
σ f eTx (s)M1 ex (s)ds + eTf (s)M2 e f (s)ds
1 0 0
− eTf (t)sym((T Bi )T P1 T B j )e f (t) ∫ ∞
σ
≤ γ12 d T (s)M3 d(s)ds, for ζ ∥ξ (t)∥2 > δ . (36)
2
− eTf (t)(T Bi )T P1 T D j d(t) + δ }. (29) 0
σ
Therefore, J1 < 0 for ζ ∥ξ (t)∥2 > δ .
Let By posing Yi = P1 Li and ρ1 = γ12 we obtain (23). □
J1 (t) = V̇1 (t) + eTx (t)M1 ex (t) + eTf (t)M2 e f (t) Remark 2: By making a transformation into the fol-
− γ12 d T (t)M3 d(t), (30) lowing optimization problem, inequality (21) in Theorem 22
r r 1 can be easily solved thanks to LMI Toolbox.
J1 (t) ≤ ∑ ∑ hi h j {eTx (t)(sym(P1 (T Āi − LiC̄)) Minimize η > 0 subject to
i=1 j=1
[ ]
+ M1 )ex (t) + 2eTx (t)P1 T Di d(t) η Iq (T Bi )T P1 − NiC̄
> 0, i = 1, 2, · · · , r.
1 T ∗ η In+w
+ e (t)Me f (t)
σµ f (37)
2
− eTf (t)(T Bi )T P1 (T Ā j − L jC̄)ex (t) 3.3.2 Design of observer-based FTC
σ
1 In this part, an observer-based FTC is proposed to com-
− eTf (t)sym((T Bi )T P1 T B j )e f (t)
σ pensate the effects of actuator faults and to stabilize the
2 closed-loop faulty descriptor system.
− eTf (t)(T Bi )T P1 T D j d(t) + eTf (t)M2 e f (t)
σ The observer-based FTC can be given as follows:
− γ12 d T (t)M3 d(t) + δ }. (31) r
[ ]T u(t) = − ∑ hi Ki x̂(t) − fˆa (t). (38)
Let ξ (t) = eTx eTf (t) d T (t) , then we obtain i=1

r r The closed-loop of the T-S Descriptor System becomes

J1 (t) ≤ ∑ ∑ hi h j {ξ T (t)ϕi j ξ (t) + δ }. (32)
i=1 j=1 r r
E ẋ(t) = ∑ ∑ hi h j [(Ai − Bi K j )x(t) + Bi K j ex (t)
? If condition (22) holds we can obtain i=1 j=1 Kxj

+ Bi e f (t) + Di d(t)], (39)

J1 (t) ≤ −ζ ∥ξ (t)∥2 + δ , (33) r
z(t) = ∑ hi (θ (x(t)))[CLi x(t)]. (40)
where ζ = λmin (−ϕi j ). i=1
It follows that
Theorem 2: Given the H∞ performance level γ2 , and
V̇1 (t) + eTx (t)M1 ex (t) + eTf (t)M2 e f (t) positive definite matrices M44 , M5 = diag(M11 , M22 , M33 ),
6 Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji, and Mohamed Chaabane

systems (39)-(40) are robustly asymptotically stable under When η (t) ̸= 0, integrating both sides of (49) with re-
H∞ performance level γ2 , that is spect to t over time period [0 ∞] yields
∫ ∞ ∫ ∞ ∫ ∞ ∫ ∞
zT (s)M4 z(s)ds ≤ γ22 η T (s)M5 η (s)ds (41) V̇2 (s)ds + zT (s)M4 z(s)ds
0 0 0 0
∫ ∞
[ ]T − γ22 η T (s)M5 η (s)ds ≤ 0. (50)
with η (t) = eTx (t)
d (t) eTf (t) T
if there exist posi- 0
tive definite matrices X and Wi (i = 1, · · · , r) such that the
following conditions hold: As V2 (∞) ≥ 0, and with zero initial condition V2 (0) = 0,
one obtains
EX = X T E T ≥ 0, (42) ∫ ∞ ∫ ∞
zT (s)M4 z(s) ds ≤ γ22 η T (s)M5 η (s) ds. (51)
Ξi j + Ξ ji < 0, i, j = 1, 2, · · · , r, i ≤ j, (43) 0 0

where Therefore, J2 < 0.

  By replacing (47) and (40) into (48), one can obtain
ψi11j BiW j Bi Di X T CLTi
 ∗ ψ 22
0 0 0  r r
  J2 (t) = ∑ ∑ hi h j {xT (t)sym(P2T (Ai − Bi K j ))x(t)
Ξi j = 
 ∗ ∗ −ρ M
2 22 0 0 ,
 i=1 j=1
 ∗ ∗ ∗ −ρ2 M33 0 
+ 2xT (t)P2T Bi K j ex (t) + 2xT (t)P2T Bi e f (t)
∗ ∗ ∗ ∗ −M44
(44) + 2xT (t)P2T Di d(t) + xT (t)CLTi M4CLi x(t)
− γ22 η T (t)M5 η (t)}, (52)
in which
then,
ψi11j = sym(Ai X − BiW j ),
ψ 22 = −ρ2 λ (XM11 + M11 X) + ρ2 λ M11 , J2 (t)
{[ ]T[ ][ ]}
r r
x(t) Θ11 Θ12 x(t)
with ρ2 = γ22 . = ∑ ∑ hi h j ij ij
,
In this case, the gains of the controller are given by Ki = i=1 j=1
η (t) ∗ −γ22 M5 η (t)
Wi X −1 .
where
Proof: Considering the Lyapunov function as follows: 

ex (t)
V2 (t) = xT (t)E T P2 x(t). η (t) =  e f (t)  ,
The time derivative of V2 (t) can then be written as d(t)
i j = θi j +CLi M4CLi , θi j = sym(P2 (Ai − Bi K j )),
Θ11 T T
V̇2 (t) = ẋT (t)E T P2 x(t) + xT (t)E T P2 ẋ(t). (45)
[ ]
Define Θ12
ij = P2T Bi K j P2T Bi P2T Di ,
 
M11 0 0
E T P2 = P2T E ≥ 0. (46) M5 =  ∗ M22 0 .
∗ ∗ M33
By considering (39) and (46) we obtain
r r By applying the Schur complement,
V̇2 (t) = ∑ ∑ hi h j {xT (t)sym(P2T (Ai − Bi K j ))x(t) [ 11 ]
r r
Θi j Θ12
i=1 j=1
∑ ∑ hi h j ∗ −γ22 M5 < 0
ij
(53)
+ 2xT (t)P2T Bi K j ex (t) + 2xT (t)P2T Bi e f (t) i=1 j=1

+ 2xT (t)P2T Di d(t)}. (47) can be written as

 
Let θi j P2T Bi K j P2T Bi P2T Di CLTi
 ∗ −γ22 M11 0 0 0 
r r  
J2 (t) = V̇2 (t) + zT (t)M4 z(t) − γ22 η T (t)M5 η (t). (48) 
∑ ∑ hi h j  ∗ ∗ −γ2 M22 20 2
0  
i=1 j=1  ∗ ∗ ∗ −γ2 M33 0 
It proceeds that
∗ ∗ ∗ ∗ −M4−1
V̇2 (t) + zT (t)M4 z(t) − γ22 η T (t)M5 η (t) < 0, (49) < 0. (54)

when η (t) = 0, (49) means V̇2 (t) ≤ 0, then closed-loop Consider the following symmetric matrix:
systems (39)-(40) are robustly asymptotically stable in the
case of η (t) = 0. X = diag(P2−T , P2−T , I, I, I).
 Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno Descriptor Nonlinear Systems with Sensor and ... 7

We can transform inequality (54) by pre and post multi- The model parameters are l = 2.8, L = 5.5, v = −1, t¯ = 2
plying it by X, we obtain: and t0 = 0.5.
  To have the T-S descriptor representation, the following
ϑi j Bi K j P2−1 Bi Di P2−T CLiT
state variable is introduced:
 ∗ ϑ 22 0 0 0 
r r   vt¯
∑ ∑ i jh h  ∗ ∗ −γ 2
M 0 0  x4 (t) = x2 (t) − x1 (t). (58)
2 22  Lt0
i=1 j=1  ∗ ∗ ∗ −γ22 M33 0 
∗ ∗ ∗ ∗ −M4−1 The following fuzzy rules can be employed:
Rule 1: If θ (t) = x2 (t) + Ltvt¯0 x1 (t) is about 0, Then
< 0, (55)
{
Kxj E ẋ(t) = A1 x(t) + B1 u(t) + Dd(t),
where ϑi11j = sym((Ai − Bi K j )P2−1 ) and ϑi22 j =
2 −T −1
y(t) = Cx(t).
−γ2 P2 M11 P2 .
Considering Lemma 2, the following inequality can be Rule 2: If θ (t) = x2 (t)+ Ltvt¯0 x1 (t) is about π or −π , Then
obtained {
E ẋ(t) = A2 x(t) + B2 u(t) + Dd(t),
− γ22 P2−T M11 P2−1 y(t) = Cx(t),
≤ −λ γ22 (P2−T M11 + M11 P2−1 ) + λ 2 γ22 M11 . (56)
where
   
By posing X = P2−1 , Wi = Ki P2−1 , M44 = M4−1 and ρ2 = γ22 1 0 0 0 − Ltvt¯0 0 0 0
0 1 0  vt¯ 0 0 0 
we obtain inequality (44). □ 0 , A1 =  
E =
0 0 1   Ltv20t¯2 vt¯ ,
0  − 2Lt0 t0 0 0 
Remark 3: From augmented estimation vector x̄ˆ(t), 0 0 0 0 − Ltvt¯0 1 0 −1
we can deduce easily the estimation
[ of] states and sen-
 vt¯   
sor faults as follows: x̂(t) = I 0 x̄ˆ(t) and fˆs (t) = − Ltvt¯0 0 0 0
[ ] n w
 vt¯ 0 0 0 
lt0
0n Iw x̄ˆ(t). The actuator faults can be regulated by  0 
B1 =   , A2 = 
Lt0 
,
the adaptive observer proposed in (13).  0   − φ2Lt
2 ¯2
v t φ vt¯
0 0 
0 t0
Algorithm 1: The design of the proposed adaptive ob- 0 − Ltvt¯0 1 0 −1
server and the fault tolerant controller can be summarized  vt¯     
lt0
0 2
as follows.  0   0  1
B2 = 
 0
 , D =  vt¯  , C = I4 , F =   ,
   1
• Compute the matrices T and H by solving equation t0
(10). 0 0 1
[ ] [ ]
• Compute the gains of the adaptive observer by solv- CL1 = 2 1 0 1 , CL2 = 0 −1 −1 1 .
ing the optimization problem given by (37) and the
LMI constraints given by (22). We set φ = 10tπ and d(t) = 0.2sin(θ (t)) − θ (t).
0

The membership functions for Rules 1 and 2 are as fol-

• Compute the gains of the FTC by solving the LMI lows:
constraints given by (42) and (43). ( )
1
• Implement the observer (13) to estimate both states h1 (θ (t)) =
1 + exp(−3(θ (t) + 0.5π ))
and sensor/actuator faults, then the FTC (38) to ( )
achieve the objectives of the FTC. 1
× 1− ,
1 + exp(−3(θ (t) − 0.5π ))
4. NUMERICAL EXAMPLE h2 (θ (t)) =1 − h1 (θ (t)).
By solving (10), T and H can be given as follows
In this part we consider a truck-trailer system to show  
the effectiveness of our results. 0.8333 0.1667 0.1667 0
Considering the following dynamic model of the truck-  0.1667 0.5833 0.0833 0 
 
trailer system [23, 38] T =  0.1667 0.0833 0.5833 0  ,
 0.3333 0.1667 0.1667 0 
vt¯ vt¯ −0.3333 −0.1667 −0.1667 0
ẋ1 (t) = − x1 (t) + u(t),
Lt0 lt0  
0.1667 −0.1667 −0.1667 0
vt¯  −0.1667 0.4167 −0.0833 0 
ẋ2 (t) = x1 (t), (57)  
Lt0 H =  −0.1667 −0.0833 0.4167 0  .

vt¯ vt¯  −0.3333 −0.1667 −0.1667 1 
ẋ3 (t) = sin[x2 (t) + x1 (t)],
t0 2L 0.3333 0.1667 0.1667 0
8 Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji, and Mohamed Chaabane

By choosing the tuning parameter values as follows: σ = 0.2 0.2

2, µ = 0.5, η = 10−4 and ρ1 = 1 to satisfy Theorem 1 and
0
Remark 1 then λ = 2, ρ2 = 1 and M11 = eye(4) to satisfy 0.1

Theorem 2. -0.2
x1 (t)
Using MATLAB LMI Toolbox, the optimization prob- -0.4
0
estimation of x 1 (t)
lem of theorem 1 and 2 can be easily solved.
-0.6 -0.1
The gains of the Adaptive Fuzzy Observer: 0 5 10 15 20 25 30 0

  Time(s)
0.2
5.4503 −5.3322 −4.4967 −0.3343 0.2
0.1 0.2
0.2
 1.5950 −2.3439 −0.6450 −0.0142 
 
L01 =   0 x2 (t)
 4.2856 −16.3160 6.5687 0.4703  ,
0
0.1
0 0.1

-0.2  2.3915 −5.0311 −0.7733 0.9286  -0.2

estimation of x 2 (t)
-0.2
x (t) xx1 (t)
−2.5356 4.8958
(t)
0.7645 10.2876 -0.10 3 0
-0.4
 
estimation of x 1 (t) -0.4 estimation of x 1(t)
3
-0.4

5.9981 −6.0296 −5.5375 −0.9673 -0.6

-0.2 -0.1
-0.6
-0.6
 1.7971 −2.5434 15 −1.1143
20 −0.2667  30 -0.1
00 55 00
 5 
0 10 25 10
10 15
15 20
20 25
25 30
30

L2 =  3.9047 −15.5800Time(s)
6.7274 1.3254 ,

Time(s)
Time(s)
0.2
0.1  2.6698 −4.9917 −1.2290 0.8808  0.2
0.1
0.2 0.2

0 −2.5629 5.3867 0.9408 −0.0410 0

x2 (t)
0
0 0.10
estimation of x 2 (t)
and-0.2
controller gains: x1x(t)(t)
-0.2 -0.2
xx3 (t)
(t)
-0.1
-0.4 [ 3
] (t) 0
-0.1
-0.4
4
-0.4
K1 = −2.7152 5.2551 −0.5846 −0.1428
estimation
estimationofofx 1x(t) estimation
estimation of
of xx3 (t)
(t)
, 3 4

-0.6
-0.2 [ ] -0.1
-0.2
-0.6 -0.6
K020 = −2.6972
55 5.7485
10
10 −0.5806
15
15 −0.100325
20
20 25 . 30
30 00 55 10
10 15
15 20
20 25
25 30
30 0
Time(s)
Time(s) Time(s)
Time(s)
us consider in the first time the sensor fault fs1 (t) as a
Let 0.1 0.2
square-wave signal between 10s and 25s, and the actuator
0
fault 0as
{ -0.2
-0.1 0, t<
x3 (t)5, x4 (t)
fa1 (t) = estimation of x (t) -0.4
5(1 − exp(−0.5(t − 5))), 5 ≤ t ≤ 30.3
estimation of x 4 (t)

-0.2 -0.6
0 5 10 15 20 25 (59) 30 0 5 10 15 20 25 30
Time(s) Time(s)
Consider now the sensor fault fs2 (t) as a square-wave
signal between 10 s and 25 s, and the actuator fault as Fig. 2. System states and their estimations under FTC law.


 0, t ≤ 5, 7

fa2 (t) = 5 + 0.5sin(0.2π (t − 5)), 5 < t ≤ 17, 5


 4

5.5, 17 < t ≤ 30. 3

(60) 1

0 actuator fault f a1 (t)

-1
estimation of fa1 (t)

The simulation [initial states in ]the first case -2

T
are set as x0 = −0.1 −0.1 −0.2
-3

and w0 = -4

[ ]T 0 5 10 15
Time(s)
20 25 30

0.1 0.1 0.1 0.2 0 . And in the second case,

[ ]T
they are set as x0 = −0.1 −0.1 −0.2 and w0 = Fig. 3. Actuator fault fa1 (t) and its estimated fˆa1 (t) under
[ ]T FTC law.
0.2 0.1 0.2 0.1 0 . By choosing Γ = 2.5 in
the simulation example, for the two types of actuator
faults, the derivative of fa1 (t) and fa2 (t) over time are sensor/actuator faults and external disturbances simulta-
norm bounded by fa1max = 2.5 and fa2max = 0.31, respec- neously. The learning rate should be minutely selected,
tively. δ1 = σµ fa21max λmax (Γ−1 M −1 Γ−1 ) = 8.6.10−3 and otherwise, increasing or decreasing this value will lead to
δ2 = σµ fa22max λmax (Γ−1 M −1 Γ−1 ) = 1.35.10−4 reduce the an unsatisfactory accuracy and rapidity of estimation. The
radius of the ball in which the estimation errors converge. evolutions of system states, sensor and actuator faults as
It should be noted that when selecting the learning rate well as their estimated are depicted in Figs. 2-7 in the two
Γ = 2.5, based on the proposed adaptive fault estima- cases. It is quite clear to remark that the adaptive observer
tion algorithm, the fault-tolerant controller can rapidly re- designed in this work leads to a good estimation of system
cover the performance of the system in the presence of states, actuator and sensor faults. Based on the simulation
 Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno Descriptor Nonlinear Systems with Sensor and ... 9

0.2 8

0.15
6

0.1
4
0.05

2
0

-0.05 0

-0.1
actuator fault f (t)
-2 a2
sensor fault fs1 (t)
-0.15 estimation of f (t)
a2
estimation of fs1 (t) -4
-0.2

-0.25 -6
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time(s) Time(s)

Fig. 4. Sensor fault fs1 (t) and its estimated fˆs1 (t) under Fig. 6. Actuator fault fa2 (t) and its estimated fˆa2 (t) under
FTC law. FTC law.

0.2

0.2 0.15
0.1
0.1
0
0.05

-0.2 00
x1 (t) -0.05
x (t)
2
-0.4 estimation of x (t)
-0.1

1 -0.1
-0.15
estimation offaultxf 2(t)(t)
sensor s2

estimation of fs2 (t)

-0.6 -0.2

0 5 10 15 20 25 30 -0.25 0 5 10 15 20 25 30
0 5 10 15 20 25 30

Time(s) Time(s)
Time(s)

0.1
0.2
0.1 Fig. 7. Sensor fault fs2 (t) and its estimated fˆs2 (t) under
0.1
0
0 0 FTC law.
0
-0.2 0
x (t) -0.2 xx2 (t)(t)
-0.1 xx12(t)
(t)
-0.4 3
estimation of
4
ion of x 1 (t) -0.1 estimation of xxx1 (t)
estimation of (t)
(t) -0.1 estimation
estimationofofx x2 (t)(t)
tems. This kind of systems may be affected by actuator,
2
3 4
-0.6 -0.4
5 30 -0.20 5 10 15
15 20
20 25
25 30
30 sensor faults
0 5 and external
10 disturbances.
15 20 By 25considering 30
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time(s)
Time(s)
Time(s)
the sensor faults as an auxiliary
Time(s) state vector, the original
Time(s)
0.1 descriptor system has been transformed into another one.
0.1 Using the H∞ optimization technique, an Adaptive Fuzzy
0 0
Observer has been firstly proposed to achieve a simulta-
0
neous
-0.2
estimation of descriptor system states, actuator and
-0.2 xx3x(t)
(t) x4 (t)
-0.1 2(t)
ion of x (t) -0.1
4
estimationofof
estimation ofxxx(t)(t)
sensor faults. Secondly, we have introduced an observer-
estimation of x 4 (t)
on of x 31(t) estimation 3 2(t) 4
-0.4 based FTC to stabilize the closed-loop faulty descriptor
-0.4
-0.2
5 30 00 5 10 15 20 25 30
5 30 0 5
5 10
10 15
15 20
20 25
25 30
30 system.
0 Considering
5 10 the H∞15performance
20 index,
25 it is 30im-
Time(s)
Time(s) Time(s)
Time(s) portant to mention that the coupling issue resulting from
the observer and the controller design can be avoided by
0
introducing them independently. Thus, sufficient condi-
tions are presented in terms of LMIs. The effectiveness of
-0.2 x4 (t) the proposed method has been proven by simulation on a
on of x 3 (t) estimation of x 4 (t) truck-trailer system.
-0.4
5 30 0 5 10 15 20 25 30
Time(s) REFERENCES

Fig. 5. System states and their estimations under FTC law. [1] S. de Oca, V. Puig, M. Witczak, and Ł. Dziekan, “Fault-
tolerant control strategy for actuator faults using LPV tech-
niques: Application to a two degree of freedom helicopter,”
results, we can conclude that the adaptive fuzzy observer- International Journal of Applied Mathematics and Com-
based FTC used in this paper can rapidly recover the per- puter Science, vol. 22, no. 1, pp. 161-171, 2012.
formance and the stability of the closed-loop descriptor [2] X. Li and F. Zhu, “Simultaneous time-varying actuator and
system despite the presence of actuator, sensor faults and sensor fault reconstruction based on PI observer for LPV
external disturbances. systems,” International Journal of Adaptive Control and
Signal Processing, vol. 29, no. 9, 1086-1098, 2015. [click]
5. CONCLUSION [3] X. Li and F. Zhu, “Simultaneous actuator and sensor
fault estimation for descriptor LPV system based on H∞
In this paper, we have proposed a strategy of fault es- reduced-order observer,” Optimal Control Applications
timation and fault tolerant control for T-S descriptor sys- and Methods, vol. 37, no. 6, pp. 1122-1138, 2016.
10 Dhouha Kharrat, Hamdi Gassara, Ahmed El Hajjaji, and Mohamed Chaabane

[4] Q. Shen, B. Jiang, and P. Shi, “Active fault-tolerant con- [17] H. Li, F. You, F. Wang, and S. Guan, “Robust fast adap-
trol against actuator fault and performance analysis of the tive fault estimation and tolerant control for TS fuzzy sys-
effect of time delay due to fault diagnosis,” International tems with interval time-varying delay,” International Jour-
Journal of Control, Automation and Systems, vol. 15, no. nal of Systems Science, vol. 48, no. 8, pp. 1708-1730, 2017.
2, pp. 537-546, 2017. [click] [click]
[5] J. Yang, F. Zhu, X. Wang, and X. Bu, “Robust sliding-mode [18] X. Li, D. Lu, G. Zeng, J. Liu, and W. Zhang, “Integrated
observer-based sensor fault estimation, actuator fault de- fault estimation and non-fragile fault tolerant control de-
tection and isolation for uncertain nonlinear systems,” In- sign for uncertain Takagi-Sugeno fuzzy systems with actu-
ternational Journal of Control, Automation and Systems, ator fault and sensor fault,” IET Control Theory & Appli-
vol. 13, no. 5, pp. 1037-1046, 2015. [click] cations, 2017.
[6] J. Zhang, A. K. Swain, and S. K. Nguang, Robust [19] F. You and M. Li, “Fault tolerant control for TS fuzzy sys-
Observer-Based Fault Diagnosis for Nonlinear Systems tems with simultaneous actuator and sensor faults,” Proc.
Using MATLAB, Advances in Industrial Control, 2016. of Control And Decision Conference (CCDC), 2017 29th
Chinese, pp. 7354-7359, IEEE, 2017.
[7] S. H. S. Ziyabari and M. A. Shoorehdeli, “Robust fault di-
agnosis scheme in a class of nonlinear system based on [20] H. Gassara, A. El Hajjaji, and M. Chaabane, “Adaptive
UIO and fuzzy residual,” International Journal of Control, fault tolerant control design for Takagi-Sugeno fuzzy sys-
Automation and Systems, pp. 1-10, 2017. tems with interval time-varying delay,” Optimal Control
Applications And Methods, vol. 35, pp. 609-625, 2014.
[8] J, Han, H. Zhangn, Y. Wang, and X. Liu, “Robust [click]
state/fault estimation and fault tolerant control for T-S
fuzzy systems with sensor and actuator faults,” Journal of [21] K. Zhang, B. Jiang, and V.t Cocquempot, “Adaptive
the Franklin Institute, vol. 353, pp. 615-641, 2016. observer-based fast fault estimation,” International Jour-
nal of Control, vol. 6, no. 3, pp. 320-326, 2008.
[9] M. Liu, X. Cao, and P. Shi, “Fuzzy-model-based fault-
tolerant design for nonlinear stochastic systems against si- [22] Z. Gao and S. X. Ding, “Actuator fault robust estimation
multaneous sensor and actuator faults,” IEEE Transactions and fault-tolerant control for a class of nonlinear descriptor
on Fuzzy Systems, vol. 21, no. 5, pp. 789-799, 2013. [click] systems,” Automatica, vol. 43, no. 5, pp. 912-920, 2007.

[10] K. Zhang, B. Jiang, V. Cocquempot, and H. Zhang, “A [23] Q. Jia, W. Chen, Y. Zhang, and H. Li, “Fault reconstruction
framework of robust fault estimation observer design for and fault-tolerant control via learning observers in Takagi-
continuous-time/discrete-time systems,” Optimal Control Sugeno fuzzy descriptor systems with time delays,” IEEE
Applications and Methods, vol. 34, no. 4, pp. 442-457, Transactions on Industrial Electronics, vol. 62, no. 6, pp.
2013. [click] 3885-3895, 2015.

[11] X.-J. Li and G.-H. Yang, “Robust adaptive fault-tolerant [24] D. Kharrat, H. Gassara, M. Chaabane, and A. El-Hajjaji,
control for uncertain linear systems with actuator failures,” “Fault tolerant control based on adaptive observer for
IET control theory & applications, vol. 6, no. 10, pp. 1544- Takagi-Sugeno fuzzy descriptor systems,” Proc. of 16th
1551, 2012. [click] International Conference on Sciences and Techniques of
Automatic Control and Computer Engineering (STA), pp.
[12] M. Sami and R. J. Patton, “Active fault tolerant control for 273-278, IEEE, 2015.
nonlinear systems with simultaneous actuator and sensor
faults,” International Journal of Control, Automation and [25] D. Kharrat, H. Gassara, A. El Hajjaji, and M. Chaabane,
Systems, vol. 11, no. 6, pp. 1149-1161, 2013. “Adaptive fuzzy observer-based fault-tolerant control for
Takagi-Sugeno descriptor nonlinear systems with time de-
[13] X.-H. Chang, Z.-M. Li, and J. H. Park, “Fuzzy generalized lay,” Circuits, Systems, and Signal Processing, pp. 1-20,
H2 filtering for nonlinear discrete-time systems with mea- 2017.
surement quantization,” IEEE Transactions on Systems,
Man, and Cybernetics: Systems, 2017. [26] T. Taniguchi, K. Tanaka, and H. O. Wang, “Fuzzy descrip-
tor systems and nonlinear model following control,” IEEE
[14] X.-H. Chang, J. H. Park, and P. Shi, “Fuzzy resilient Transactions on Fuzzy Systems, vol. 8, no. 4, pp. 442-452,
energy-to-peak filtering for continuous-time nonlinear sys- 2000. [click]
tems,” IEEE Transactions on Fuzzy Systems, 2016.
[27] Q.-X. Jia, Y.-C. Zhang, W. Chen, and H.-Y. Li, “Fault
[15] H. Gassara, A. El Hajjaji, M. Kchaou, and M. Chaabane, reconstruction in descriptor linear parameter-varying sys-
“Observer based (Q, V, R)-α -dissipative control for TS tems via polytopic unknown-input proportional-integral
fuzzy descriptor systems with time delay,” Journal of the observers,” Optimal Control Applications and Methods,
Franklin Institute, vol. 351, no. 1, pp. 187-206, 2014. vol. 36, no. 6, pp. 873-888, 2015.
[16] D. Ichalal, B. Marx, J. Ragot, S. Mammar, and D. Maquin, [28] Y. Ma and Y. Yan, “Observer-based H∞ guaranteed cost
“Sensor fault tolerant control of nonlinear Takagi-Sugeno control for uncertain singular time-delay systems with in-
systems, application to vehicle lateral dynamics,” Interna- put saturation,” International Journal of Control, Automa-
tional Journal of Robust and Nonlinear Control, vol. 26, tion and Systems, vol. 14, no. 5, pp. 1254-1261, 2016.
no. 7, pp. 1376-1394, 2016. [click] [click]
 Adaptive Observer and Fault Tolerant Control for Takagi-Sugeno Descriptor Nonlinear Systems with Sensor and ... 11

[29] B. Marx, D. Koenig, and J. Ragot, “Design of observers Hamdi Gassara received the Ph.D. in au-
for Takagi-Sugeno descriptor systems with unknown inputs tomatic control from the University of Pi-
and application to fault diagnosis,” IET Control Theory and cardie Jules Verne (UPJV), in 2011. Prior
Applications, vol. 1, no. 5, pp. 1487-1495, 2007. [click] to Ph.D., he received his Master degree
[30] K. Tanaka, H. Ohtake, and H. O. Wang, “A descriptor sys- from UPJV in 2008. His teaching expe-
tem approach to fuzzy control system design via fuzzy lya- rience started when he was Ph.D. student
punov functions,” IEEE Transactions on Fuzzy Systems, in UPJV France from 2008 to 2011. He is
vol. 15, pp. 333-341, 2007. [click] currently an Assistant Professor in Electri-
cal Department at National School of En-
[31] C. Lin, Q.-G. Wang, and T. H. Lee, “Stability and stabi- gineering of Sfax, Tunisia. His research focuses on analysis and
lization of a class of fuzzy time-delay descriptor systems,” control for fuzzy models with time delay, fault tolerant control,
IEEE Transactions on Fuzzy Systems, vol. 14, no. 4, pp. diagnosis, saturation, and polynomial fuzzy models.
542-551, 2006.
[32] S. Xu, P. Van Dooren, R. Stefan, and J. Lam, “Robust sta-
Ahmed El Hajjaji received the Ph.D. de-
bility and stabilization for singular systems with state delay
gree in automatic control and HDR degree
and parameter uncertainty,” IEEE Transactions on Auto-
from the University of Picardie Jules Verne
matic Control, vol. 47, no. 7, pp. 1122-1128, 2002. [click]
(UPJV), France in 1993 and 2000, respec-
[33] H. H. Choi, “LMI-based nonlinear fuzzy observer- tively. He is currently a full professor and
controller design for uncertain MIMO nonlinear systems,” head of Automatic control and Vehicle Re-
IEEE Transactions on Fuzzy Systems, vol. 15, no. 5, pp. search Group in MIS Lab (Modeling In-
956-971, 2007. [click] formation Systems Laboratory) of UPJV.
[34] B. Erkus and Y. J. Lee, “Linear matrix inequalities and mat- He has been the director of the Profes-
lab LMI toolbox,” University of Southern California Group sional Institute of Electrical Engineering and Industrial Comput-
Meeting Report, Los Angeles, California, 2004. ing from 2006 to 2012. Since 1994, he has published more than
350 Journal and conference papers in the areas of advanced fuzzy
[35] D. Koenig, “Observer design for unknown input nonlinear control, fault detection and diagnosis and fault tolerant control
descriptor systems via convex optimization,” IEEE Trans- and their applications to vehicle dynamics, engine control, power
actions on Automatic Control, vol. 51, pp. 1047-1052, systems, renewable energy conversion systems and to industrial
2006. processes. His research interests include fuzzy control, vehicle
[36] M. Alma and M. Darouach, “Adaptive observers design for dynamics, fault-tolerant control, neural networks, maglev sys-
a class of linear descriptor systems,” Automatica, vol. 50, tems, and renewable energy conversion systems.
no. 2, pp. 578 - 583, 2014.
[37] M. Darouach and M. Boutayeb, “Design of observers for Mohamed Chaabane was born in Sfax,
descriptor systems,” IEEE Transactions on Automatic Con- Tunisia, on August 26, 1961. He received
trol, vol. 40, no. 7, pp. 1323-1327, 1995. the Ph.D. degree in electrical engineer-
[38] K. Zhang, B. Jiang, and P. Shi, “A new approach to ing from the University of Nancy, Nancy,
observer-based fault-tolerant controller design for Takagi- France, in 1991. He is currently a Profes-
Sugeno fuzzy systems with state delay,” Circuits Syst Sig- sor with the National School of Engineer-
nal Process, vol. 28, pp. 679-697, 2009. [click] ing, University of Sfax, where he has been
a Researcher with the Laboratory of Sci-
ences and Techniques of Automatic Con-
trol and Computer Engineering (Lab-STA) since 1997. From
Dhouha Kharrat received her Master de-
1988 to 1992, he was an Associate Professor with the University
gree in automatic control from the Univer-
of Nancy, where he was a Researcher at Center of Automatic
sity of Poitiers, France, in 2014. She is
Control of Nancy (CRAN). The main research interests are in
currently working towards a Ph.D. degree
the filed of robust and optimal control, fault tolerant control, de-
at both University of Picardie Jules Verne
lay systems, descriptor systems, fuzzy logic systems and appli-
(UPJV), Amiens, France and National
cations of these techniques to fed-batch processes, asynchronous
School of Engineering of Sfax (ENIS),
machines, agriculture systems and renewable energy. Currently,
Sfax, Tunisia. Her research interests
he is an associate editor of the International Journal on Sciences
include fault tolerant control, diagnosis,
and Techniques of Automatic Control and Computer Engineer-
nonlinear control, analysis and design for T-S descriptor sys-
ing (www.sta-tn.com).
tems with time-delay and their application to practical engineer-
ing systems.