Observer based Fault Tolerant Control for Takagi-Sugeno Nonlinear

Descriptor systems
H. HAMDI†, M. RODRIGUES‡ Ch. MECHMECHE† and N. BENHADJ BRAIEK†

Abstract— In this paper, an observer-based Fault Tolerant Probably the most signficant one is the integration between
control (FTC) study is proposed for nonlinear descriptor the FDD part and the FTC part. The majority of approaches
systems approximated by Takagi-Sugeno representation. A in the literature are focused on sytems modeled by many
control laws is designed in order to compensate the actuator
faults and allows the system states to track a reference states classes of nonlinear ordinary systems [1], [9]. Nonlinear de-
corresponding to the output of the original descriptor system scriptor systems are not also studied, then only a considerable
in the fault free case. The design of such a control law requires amount of results have been established in the framework of
the knowledge of the faults, for this aim, a Proportional Multi- linear descriptor systems. Very often the dynamics of real
Integral Observer (PMIO) is presented to achieve this task. The physical systems can not be represented accurately enough
robust stability of the system with the fault tolerant control law
is analyzed with Lyapunov theory. Sufficient stability conditions by linear dynamical models so that nonlinear models have to
and the gains of the FTC are obtained in terms of linear matrix be used. This necessitates the development of techniques for
inequalities (LMIs). A numerical example is used to illustrate FTCS design that can explicitly deal with nonlinearities in
the efficiency of the studied method. the mathematical representation of the system. Nonlinearities
are, in fact, very often encountered in the representations
I. INTRODUCTION
of complex safety critical controlled systems. During the
Reliability and safety of physical process have always last two decades, fuzzy technique has been widely used
been a major concern for industrial manufacturers. There is in nonlinear system modeling, especially for systems with
an absolute necessity to identify early unexpected changes incomplete plant information. Fuzzy logic systems serve well
in the system before they lead to a complete breakdown. as universal approximators [14]. The well-known Takagi-
Fault Detection and Isolation (FDI) and Fault Tolerant Con- Sugeno (T-S) fuzzy model [7] is a popular and convenient
trol (FTC) for linear and nonlinear ordinary systems have tool in functional approximations. Accordingly, the stabi-
already been addressed in a large number of papers [11], lization problem for systems in T-S fuzzy model has been
[6]. FDI comprises fault detection, fault isolation and even studied. Recently, a wider class of fuzzy systems described
fault identification. FTCS are generally divided into two by the descriptor form is considered in [12], where the
classes: passive and active. Passive FTCS are based on model is in the extended T-S fuzzy model. It is known
robust controller design techniques and aim at synthesizing that a descriptor model describes a practical system better
one (robust) controller that makes the closed-loop system than a standard dynamic model [4]. The descriptor system
insensitive to certain faults. This approach requires no online describes a wider class of systems including physical models
detection of the faults, and is therefore computationally more and nondynamic constraints. In [13], a fuzzy model in the
attractive. Its applicability, however, is very restricted due to descriptor form is introduced, and stability and stabilization
its serious disadvantages. problems for the system are addressed.
As opposed by the passive methods, the active approach Motivated by the above discussion, in this paper, a Pro-
to the design of FTCS is based on controller redesign, portional Multi-Integral Observer (PMIO) is developed. The
or selection/mixing of predesigned controllers. This tech- proposed PMIO is dedicated to the design of a fault tolerant
nique usually requires a fault detection and diagnosis (FDD) control strategy for a class of nonlinear descriptor sys-
scheme that has the task to detect and localize the faults that tems described by Takagi-Sugeno models with unmeasurable
eventually occur in the system. The FDD part uses input- premise variables.
output measurement from the system to detect and localize This paper is organized as follows: in Section II the fuzzy
the faults. The estimated faults are subsequently passed to a Takagi-Sugeno structure of nonlinear descriptor systems is
reconfiguration mechanism (RM) that changes the parame- introduced. In Section III, we study the structure and the
ters and/or the structure of the controller in order to achieve design of the proposed Proportional Multi-Integral Observer.
an accept able post-fault system performance. There are a Fault tolerant control by state feedback is tackled in Section
number of important issues when designing active FTCS. IV,Finally, and before concluding, an illustrative example is
considered in Section V.
†H. Hamdi, Ch. Mechmeche and N. BenHadj Braiek are with
Laboratory for Advanced Systems (LSA), Polytechnic school of Tunisia
(e-mail: hammmdihabib@yahoo.fr, chokri.mechmeche@esstt.rnu.tn,
naceur.benhadj@ept.rnu.tn).
‡M. Rodrigues is with Laboratory Automation and Engineering Pro-
cess (LAGEP), CNRS UMR 5007,University Lyon1, France (e-mail:
mickael.rodrigues@univ-lyon1.fr).
 II. SYSTEM DESCRIPTION AND PROBLEM rewritten as an external additive signal: u(t) + f (t) where
STATEMENT f (t) = −γ u(t) with
A. Takagi-Sugeno fuzzy Model γ , diag[γ 1 , γ 2 , . . . , γ p ], 0 ≤ γ k ≤ 1 such that
 k
Consider the following general form of nonlinear descrip-  γ = 1 → a total failure of the kth actuator k ∈ [1, . . . , p]
tor systems: γ k = 0 → the kth healthy actuator
{ 
E ẋ(t) = A(x(t))x(t) + B(x(t))u(t)
(1)
y(t) = C(x(t))x(t) The goal of this paper is to seek a control law to ensure
the closed-loop stability of the system (4) as well as the
where x ∈ Rn , u ∈ Rp
(p ≤ n) and y ∈ Rm
represent respec-
actuator fault detection and isolation. This goal can be well
tively the singular state, the control input and the output
accomplished by introducing the following control law:
vectors. A(x(t)), B(x(t)) and C(x(t)) are nonlinear matrices
functions. For simplicity, we can always consider that h
u f (t) = ∑ hi (ξ (t)(Γi (x(t) − x̂(t)) − fˆ(t) + u(t)) (5)
E ∈ Rn×n is a constant matrix,i t may not have full rank. i=1
Fuzzy descriptor system is defined by extending the T-S
where x̂(t) and fˆ(t) are respectively the state and the fault
fuzzy ordinary model. The fuzzy descriptor model of Takagi
estimations and Γi ∈ Rn×p the feedback gains to be found.
Sugeno is then described by the following fuzzy IF-THEN
Before starting the FTC design, some useful basic assump-
rules;
tion for descriptor systems are given as follows [8] and [4]:
IF ξ1 (t) is M1i and ... and ξ p (t) is M1p , T HEN • A1. rank(Ci Bi ) = rank(Bi ) = p, ∀i = 1, . . . , h,
{ • A2. The triple matrix (E, Ai ,Ci ) is R-observable,
E ẋ(t) = Ai x(t) + Bi u(t)
(2) for all i = 1, . . . , h, i.e.,
y(t) = Ci x(t), i = 1...h [ ]
sE − Ai
Ai ∈ Rn×n , Bi ∈ Rn×p , and Ci ∈ Rm×n are time invariant rank = n, ∀s ∈ C . (6)
Ci
matrices of appropriate dimensions. r is the number of IF
T HEN rules, and Mi j are the fuzzy sets. ξ1 (t)...ξ2 (t) are where C denotes the complex plane.
premise variable. We set ξ (t) = [ξ1 (t)...ξ p (t)]. Then the • A3. The triple matrix (E, Ai ,Ci ) is Impulse-
descriptor equation is defined as follows; observable, for all i = 1, . . . , h, i.e.,
 
 E Ai
 h [ ]

 E ẋ(t) = ∑ hi (ξ (t)) Ai x(t) + Bi u(t) rank  0 E  = n + rank(E) (7)
i=1 (3) 0 Ci


h
 y(t) = ∑ hi (ξ (t))Ci x(t)
i=1 • A4. The fault f (t) is assumed to be a bounded
time varying signal with null sth derivative i.e.
where
∥ f (t)∥ ≤ α1 and f˙(t) ≤ α2 and 0 ≤ α1 , α2 < ∞.
βi (ξ (t)) p
• A5. Only partial actuator faults are considered,
hi (ξ (t)) =
∑hi=1 βi (ξ (t))
, βi (ξ (t)) = ∏ Mi j (ξ j )
i.e.,γ k ∈ [0 1[.
j=1

and Mi j (−) are the membership functions of the fuzzy In the following section, we propose to design a PMIO
sets of Mi j . We assume that βi (ξ (t)) ≥ 0, i = 1...h and to estimate the system state vector and the fault signal
h
∑ βi (ξ (t)) > 0, ∀t hence the weighting function hi (ξ (t)) simultaneously for TS fuzzy descriptor model. After an
i=1 efficient fault tolerant control scheme by using the estimated
satisfy the properties of the sum convex.
{ states and faults is developed.
h
∑ hi (ξ (t)) = 1 0 ≤ hi (ξ (t)) ≤ 1 III. TS FUZZY PMIO DESIGN
i=1
By using the same idea in T-S fuzzy descriptor model, a
B. Problem statement fuzzy Proportional Multi-Integral Observer uses a number of
Under actuator faults, the system (3) can be rewritten in local linear time-invariant observers. Each local observer is
the following form: associated with each fuzzy rule given below:

 h [ ] Rule i IF ξi (t) is M1i T hen

 E ẋ(t) = ∑ hi (ξ (t)) Ai x(t) + Bi (u(t) + f (t)) 
i=1 (4) 
 ż(t) = Ni z(t) + Gi u(t) + Li y(t + Hi fˆs (t))


h 

 y(t) = ∑ hi (ξ (t))Ci x(t) 
 x̂(t) = z(t) + M2 y(t)


 f˙ˆs (t) = Φsi (y(t) − ŷ(t)) + fˆs−1 (t)
i=1

where f (t) is an actuator fault. It can be represented by .. (8)


 .
an additive or a multiplicative external signal [9]. These 


 ˙fˆ (t) = Φ (y(t) − ŷ(t)) + fˆ (t)
malfunctions of an actuator faults can be represented by 
 2 2i 1

a faulty control input u f (t) = (I p − γ )u(t) which can be f˙ˆ (t) = Φ (y(t) − ŷ(t))
1 1i
where x̂ ∈ Rn , z ∈ Rn and fˆ ∈ Rm are respectively the If the following conditions hold true ∀ i = 1, ..., h:
estimated state vector, the state vector of the observer and the
M1 Ai − LiC − Ni M1 E = 0 (18a)
estimated unknown input. Φi are the integral gains matrices.
Ni , Gi , Li , Hi , Φi and M2 are the unknown parameters of the M1 E + M2C = In (18b)
local PMIO which we have to design. M1 Bi − Gi = 0 (18c)
The global state estimation is a fuzzy combination of each Li G − Ni T2 G − T1 Ri = 0 (18d)
local observer outputs. The overall PMIO dynamics will then
M1 Bi − Hi = 0 (18e)
be a weighted sum of individual linear PMIO as follows:
 then, the estimation error dynamic (14) becomes:


h
ż(t) = ∑ hi (ξ (t))(Ni z(t) + Gi u(t) + Li y(t) + Hi fˆs (t))


h


 i=1 ė(t) = ∑ hi (ξ (t))[Ni e(t) + Hi es (t)] (19)

 x̂(t) = z(t) + M2 y(t)


i=1

 h

 f˙ˆs (t) = ∑ hi (ξ (t))Φsi (y(t) − ŷ(t)) + fˆs−1 (t) or from equation (18), the above equation is equivalents to:
 i=1
(9) h
..


 . ė(t) = ∑ hi (ξ (t))[(M1 Ai + KiC)e(t) + Hi es (t)] (20)




h i=1

 f˙ˆ2 (t) = ∑ hi (ξ (t))Φ2i (y(t) − ŷ(t)) + fˆ1 (t)

 where


i=1


h
 f˙ˆ1 (t) = ∑ hi (ξ (t))Φ1i (y(t) − ŷ(t)) Ki = Ni M2 − Li (21)
i=1
The equations (15)-(17) and (20) can be rewritten in the
where the weights hi (ξ (t)), (i = 1, ..., h) are the same as the following augmented form:
weights functions used in TS descriptor model (3). fˆj (t), h ( )
j = 1, ..., s are the estimation of the (s − 1) first derivatives ē˙(t) = ∑ hi (ξ (t)) Āi + K̄iC̄ ē(t) (22)
of the fault f (t). i=1
The state and the fault estimation errors are given by: where
 
  M1 Ai Hi 0 ··· 0 0
e(t) = x(t) − x̂(t) (10) e(t)  0 0 Is ··· 0 0 
 
e j (t) = f (s− j)
(t) − fˆj (t), j = 1, 2, ..., s (11)  es (t)   .. .. .. .. .. .. 
   . . . . . . 
ē(t) =  .  , Āi =  
 ..   0 0 0 ··· Is 0 
we assume that f (s) = 0.  
el (t)  0 0 0 ··· 0 Is 
For subsequently it is assumed that the matrix C is constant.
0 0 0 ··· 0 0
Under Assumption A3, there exists nonsingular matrices
 
M1 ∈ Rn×n and M2 ∈ Rn×m such that: Ki
 −Φsi 
  [ ]
M1 E + M2C = In (12)  .. 
K̄i =  .  , C̄ = C 0 ··· ··· 0 0
 
 −Φ2i 
The dynamic estimation error is then described by:
−Φ1i
ė(t) = ż(t) − M1 E ẋ(t) (13) The system dynamics [Āi + K̄iC̄] can be stabilized by select-
ing the gain K̄i thanks to the detectability of each pair (Āi ; C̄),
Then, the dynamics of the state and the fault estimation errors ∀ i = 1, ...h.
are given as the following form: In the sequel and before considering the stability of the
estimation error dynamics (22), it is shown how to find
h matrices M1 and M2 such that constraint (18b) is satisfied.
ė(t) = ∑ hi (ξ (t))[Ni e(t) + (M1 Ai − LiC − Ni M1 E)x(t) For that, rewrite (18b) as follows:
i=1 [ ]
+(M1 Bi − Gi )u(t) + (M1 Bi − Hi ) f (t) + Hi es (t)] (14) [ ] E [ ]
M1 M2 = In (23)
h C
ės (t) = − ∑ hi (ξ (t))ΦsiCe(t) + es−1 (t) (15) [ ]
i=1
A solution M1 M2 exists [ if [4]
] :
E
.. rank =n (24)
. C
h
ė2 (t) = − ∑ hi (ξ (t))Φ2iCe(t) + e1 (t) (16) Then, a particular solution of (23) using the pseudo inverse
i=1 matrix denoted by (·)+ is given by:
h [ ]+
[ ]
ė1 (t) = − ∑ hi (ξ (t))Φ1iCe(t) (17) M1 M2 =
E
(25)
i=1 C
The proposed observer give the possibility to estimate a large V. SIMULATION EXAMPLE
class of faults, because of its multi-integral structure which
may change according to the class of faults. Thereafter, the To illustrate the performances of the proposed approach,
outputs of this observer will be used in fault tolerant control let us consider the following nonlinear dynamic system:

x1 (t) = x2 (t)
x2(t) = −2x1 (t) − 3x2 (t) + x4 (t) − x43 (t) + 2u(t)
IV. FTC FOR TS DESCRIPTOR SYSTEMS
0 = x1 (t) + x2 (t) − 2x3 (t)
0 = −x1 (t) − x2 (t) + x43 (t) − 5x4 (t) (29)
By using the PMIO (8) and the proposed active fault toler-
y1 (t) = x1 (t) + x3 (t)
ant control (5), the objective is to determine the parameters
y2 (t) = x2 (t) + x4 (t)
of the used observer and the gains Γi in order to minimize the
y3 (t) = x4 (t)
impact of actuator faults on the TS descriptor model output.
The system with the fault f (t) ∈ Rn f is described by the To carry out the proposed PMI observer design, system (29)
following T-S model with measurable premise variables: is rewritten as:
 E ẋ(t) = A(x(t))x(t) + Bu(t)
 h [ ] (30)
E ẋ(t) = ∑ hi (ξ (t)) Ai x(t) + Bi u f (t) + Bi f (t) y(t) = Cx(t)
(26)
 i=1
y(t) = Cx(t) where
   
The goal is to design the control law u f (t) such that the 0 1 0 0 0
 −2 −3 0 (1 − x42 (t))   2 
system state x f (t) converges toward the reference state x(t). A(x(t)) = 
 1
, B = 
 

In order to prove both the stability of the closed-loop system 1 −2 0 0 
and the convergence of the state and fault estimation errors −1 −1 0 (x4 (t) − 5)
2 0
and according to the equations (5), (8) and (26), the time  
1 0 0 0  
derivative of the augmented errors(22) become then: 1 0 1 0
 0 1 0 
0 
E =
 0 andC =  0 1 0 1 
h ( ) 0 0 0 
0 0 0 1
e˙¯a (t) = ∑ hi (ξ (t)) Ãi + K̃iC̃ ēa (t) (27) 0 0 0 0
i=1 .
System (30) can be exactly represented by the following
where
fuzzy model:
  
  M1 Ai (M1 + In )Bi 0 ··· 0 0
 4 [ ]
e(t)  0 0 Is ··· 0 0  E ẋ(t) = ∑ hi (ξ (t)) Ai x(t) + Bi u(t)
 es (t)    (31)
   .. .. .. .. .. ..   i=1
ēa (t) =  ..  , Ãi =  . . . . . .  y(t) = Cx(t)
 .   0 0 0 ··· Is 0 
   
el (t) 0 0 0 ··· 0 Is 0
0 0 0 ··· 0 0  1 
where Bi = B =   0 

  0
Ki Bi Γi
 −Φsi 0  [ ]    
  0 1 0 0 0 1 0 0
 .. ..  C 0 ··· ··· 0 0
K̃i =   , C̃ =  −2 −3 0 1   1 
 . .  In 0 ··· ··· 0 0 A1 =   , A =  −2 −3 0 
 −Φ2i 0   1 1 −2 0  2  1 1 −2 0 
−Φ1i 0 −1 −1 0 −5 −1 −1 0 −1
   
The synthesis of the gains K̃i of the PMI Observer and 0 1 0 0 0 1 0 0
 −2 −3 0 −3   −2 −3 0 −3 
those of the controller Γi are obtained by solving the LMIs A3 =  , A =  
given in the following theorem.  1 1 −2 0  4  1 1 −2 0 
−1 −1 0 −5 −1 −1 0 −1
Theorem 1: [5] The PMIO (8) is asymptotically stable, The membership functions are given by:
if there exist a symmetrical and definite positive matrix Q
and a matrices W̃i = QK̃i checking the following LMIs ((1−x42 (t))−1)((x42 (t)−5)+5)
h1 (x(t)) = −16 ,
((1−x42 (t))−1)(1+(x42 (t)−5))
T T h2 (x(t)) = ,
Ãi Q + QÃi + C̃T W̃i + W̃iC̃ < 0, ∀ i ∈ {1, ..., h} (28) 16
(3+(1−x42 (t)))((x42 (t)−5)+5)
 h3 (x(t)) = 16 ,
(3+(1−x42 (t)))(1+(x42 (t)−5))
h4 (x(t)) = −16
Now, following the design of PMIO and the FTC controller 10
x1(t) real x1(t) estimated
algorithm in the above sections, we consider that the fuzzy 8

model (31) is affected by a varying actuator fault as follows: 6


[ ]
4

 4
E ẋ(t) = ∑ hi (ξ (t)) Ai x(t) + B(u(t) + f (t)) 2

(32)
 i=1 0

y(t) = Cx(t) −2

u(t) = 10sin(0, 2π t) and f (t) = (0.8(t −20)−0.08(t −20)2 )ε (t)

−4

−6

where { −8

ε (t) = 1 f or 10 ≤ t ≤ 18
0 5 10 15 20 25 30
t(s)

ε (t) = 0 elsewhere
Fig. 1. Nonlinear real states x1 (t) and its estimated x̂1 (t)
A. state and fault estimation
According to the given procedure, we design the PMIO 6

based on theorem 1 via the Matlab LMI toolbox. Then we 5

x2(t) real x2(t) estimated

obtain the proportional gains matrices Ki and the integral 4

gains matrices Φi for i = 1, ..., h. 3

  2

−4.9357 −2.2850 5.1207 1

 −0.0917 −3.7974 −1.5133 

K1 =   0

 −2.9074 0.2169 −2.2369  −1

−2.2572 −13.3492 −19.3104 −2

  −3

−4.9357 −2.2850 5.1207 −4

0 5 10 15 20 25 30
 −0.0917 −3.7974 −1.5133  t(s)

K2 = 
 −2.9074

0.2169 −2.2369 
Fig. 2. Nonlinear real states x2 (t) and its estimated x̂2 (t)
−2.2572 −13.3492 −19.3104
  10
x3(t) real x3(t) estimated
−4.6053 0.7978 −4.9443 8
 −0.0790 −1.0026 −2.8533 
K3 = 
 −3.0776 0.8820

1.8584 
6

−1.0092 39.7180 −4.6505 4

  2

−4.6053 0.7978 −4.9443

 −0.0790 −1.0026 −2.8533  0

K4 = 
 −3.0776 0.8820

1.8584  −2

−1.0092 39.7180 −4.6505 −4

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

Φ13 = [ −3.9737 −20.6827 −5.4375 ],

Φ12 = [ −3.7131 −46.9261 −10.5188 ] Fig. 3. Nonlinear real states x3 (t) and its estimated x̂3 (t)
Φ11 = −4.1868 −50.7179 −11.8780 , 1.5

[ ] x4(t) real x4(t) estimated

Φ23 = [ −3.9737 −20.6827 −5.4375 ], 1

Φ22 = [ −3.7131 −46.9261 −10.5188 ] 0.5

Φ21 = −4.1868 −50.7179 −11.8780 , 0

[ ]
Φ33 = [ −3.8532 −5.4641 −16.6861 ], −0.5

Φ32 = [ −4.7780 −44.8522 −37.0845 ] −1

Φ31 = −5.2937 −46.8821 −39.8762 , −1.5

[ ]
Φ43 = [ −3.8532 −5.4641 −16.6861 ], −2
0 5 10 15 20 25 30

Φ42 = [ −4.7780 −44.8522 −37.0845 ] t(s)

Φ41 = −5.2937 −46.8821 −39.8762 ,

Fig. 4. Nonlinear real states x4 (t) and its estimated x̂4 (t)
The state estimation and the fault estimation given by the
proposed PMIO are shown on the following figures.
The behavior of the PMIO is shown in the previous figures
(1) to (4). It is observed that the proposed PMIO rebuilds
the state by using the estimate of the fault presented in the
following figure. The estimation of the actuator fault figure
(5) is used to design the fault tolerant control.
 16
7
f(t) estimated f(t) real y2(t) without fault y2(t) with fault y2(t) with FTC
14
6

12
5

10
4

8
3
6
2
4
1
2
0
0
−1
−2
0 5 10 15 20 25 30
t(s) −2

−3

Fig. 5. Actuator fault f (t) and its estimated fˆ(t) −4

0 5 10 15 20 25 30
t(s)

B. Fault tolerant control of T-S descriptor systems Fig. 7. y2 (t) of the system: without fault, with fault and output with FTC
The control observer based control law given by the
equations (9) is designed by solving the LMI optimization 6

problem defined in the theorem 1. The feedback gains Γi are y (t) without fault
3
y (t) with fault
3
y (t) with FTC
3
5
given by:
[ ] 4

Γ1 = −0.4163 −3.7974 0.3246 2.2841 , 3

[ ] 1

Γ2 = −0.4163 −3.7974 0.3246 2.2841 , 0

−1

[ ] −2

Γ3 = −0.5789 −1.0026 0.4999 −1.8506 ,

−3

−4
0 5 10 15 20 25 30

[ ] t(s)

Γ3 = −0.5789 −1.0026 0.4999 −1.8506

The following figures illustrate a comparison between the Fig. 8. y3 (t) of the system: without fault, with fault and output with FTC
output of the reference model (without fault), the output of
the faulty system without FTC and finally the output with
FTC. The proposed observer is robust with respect to varying From These figures, the FTC scheme can accommodate ef-
actuator additive fault f (t). fectively the additive actuator fault. Here, the PMI Observer
shows very good results for the estimation of abrupt actuator
fault.

15 y1(t) without fault y1(t) with fault y1(t) with FTC

VI. CONCLUSIONS
10
A combined state, faults estimation and fault tolerant
control method have been presented based on PMIO for
5
fuzzy descriptor systems affected by actuator faults. The
fault tolerant control requires the simultaneous estimations of
0 the state and faults, obtained by the proposed fuzzy PIMO.
This observer admits a greatest potential to estimate time
−5 varying faults with a good accuracy simultaneously with
the estimation of the state. Sufficient stability conditions
−10
are given in terms of LMI. Results have been illustrated in
0 5 10 15 20 25 30
t(s)
simulation.

Fig. 6. y1 (t) of the system: without fault, with fault and output with FTC R EFERENCES

[1] D. Ichalal, B. Marx, D. Maquin and J. Ragot, ”New fault tolerant

control strategy for nonlinear systems with multiple model approach”,
Conference on Control and Fault-Tolerant Systems (SysTol), pp 606 -
611, Oct. 2010.
 [2] D. Ichalal, B. Marx, J. Ragot and D. Maquin, ”Simultaneous state
and unknown inputs estimation with PI and PMI observers for Takagi
Sugeno model with unmeasurable premise variables ”, 17th Mediter-
ranean Conference on Control and Automation. MED ’09, pp 353-358,
2009.
[3] D. Koenig, ”Unknown Input Proportional Multiple-Integral Observer
Design for Linear Descriptor Systems: Application to State and Fault
Estimation”, IEEE Transactions on Automatic control, vol 50, no. 2,
pp 212-217, February 2005.
[4] H. Hamdi, M. Rodrigues, C. Mechmeche, D. Theilliol and N. BenHadj
Braiek, ”Fault detection and isolation in linear parameter-varying
descriptor systems via proportional integral observer”, International
journal of adaptive control and signal processing, Vol. 26, pp 224-
240, 2012.
[5] H. Hamdi, C. Mechmeche, M. Rodrigues and N. BenHadj Braiek,
”State and unknown inputs estimations for Multi-Model Descriptor
Systems”, 8th International Multi-Conference on Systems, Signals and
Devices, IEEE, March 22-25, 2011.
[6] H. Niemann, ”A fault tolerant control approach for descriptor sys-
tems”, Proc. The American Conf, pp 2230-2231, 2005.
[7] J. Yoneyama and A. Ichikawa, ”H∞ Control for Takagi-Sugeno Fuzzy
Descriptor Systems”, IEEE International Conference on Systems, Man,
and Cybernetics, IEEE SMC ’99 Conference Proceedings., vol. 3, pp
28-33, 1999.
[8] K. Zhang, B. Jiang, and V. Cocquempot, ”Adaptive Observer-based
Fast Fault Estimation”, International Journal of Control, Automation,
and Systems, vol. 6, no. 3, pp 320- 326, June 2008
[9] M. Rodrigues, D. Theilliol, S. Aberkane and D. Sauter, ”Fault Tolerant
Control Design For Polytopic LPV Systems”, International Journal of
Applied Mathematics and Computer Science, pp 27-37, Vol. 17 N. 1,
Number 1 / March 2007.
[10] M. Rodrigues, D. Theilliol and D. Sauter, ”Design of an Active Fault
Tolerant Control and Polytopic Unknown Input Observer for Systems
described by a Multi-Model Representation”, 44th IEEE Conference on
Decision and Control and European Control Conference ECC, Sevilla,
Spain, pp 6268-6273, December 2005.
[11] M. Blanke, M. Kinnaert, J. Lunze and M. Staroswiercki, ”Diagnosis
and fault-tolerant control”, Springer-Verlag, Berlin, 2003.
[12] R. J. Patton, J. Chen and C. J. Lopez-Toribio, ”Fuzzy Observers for
Non-linear Dynamic Systems Fault Diagnosis”, Proceedings of the
37th IEEE Conference on Decision and Control Tampa, Florida USA,
vol. 1, pp 84-98, December 1998.
[13] T. Taniguchi, K. Tanaka, and H. O. Wang, ”Fuzzy Descriptor Systems
and Nonlinear Model Following Control”, IEEE Transactions on Fuzzy
Systems, vol. 8, no. 4, August 2000.
[14] W. Shyong Yu, ”H∞ Tracking Performance Design for Fuzzy-Model-
Based Descriptor Systems Subject to Parameter Uncertainties”, JOUR-
NAL OF COMPUTERS, vol. 4, no. 7, July 2009.