J.

Automation & Systems Engineering 5-3 (2011): 112-129

Improvement of the Performances of a

JASE
Journal of Automation
Multiobserver for the Reconstruction of State and & Systems Engineering
Estimation of Unknown Inputs of a Discrete
Multimodel Using Piecewise Quadratic Functions

Sami ZEMMELa,*, Mongi BESBESb,*, Elyes MAHERZIb,*

Abstract- This paper deals with a piecewise quadratic approach very powerful for the synthesis of
a multiobserver able to estimate the states and unknown inputs of a discrete nonlinear system. The
stabilization of the error observation is guaranteed by the use of the second Lyapunov approach.
The stability conditions obtained are expressed in term of Linear Matrices Inequalities (LMI) and
are got from the use of the Lyapunov quadratic functions in a first study and from the Lyapunov
piecewise quadratic functions in a second study which appears less conservative and less
constraining than the first. Numerical examples are also presented in this paper.
Keywords: Discrete multimodel, unknown inputs multiobserver, quadratic stabilization,
piecewise quadratic stabilization, unknown input estimation

1. Introduction
The state observation of the nonlinear dynamic systems constitutes a very vast search
field having many applications. This is justified by the fact that the state estimation is an
important stage, even indispensable, for the synthesis of command laws, for the diagnosis
or for the supervision of the industrial process. The basic idea is the substitution of the
materials sensor by using non linear observers not only for economical reasons but also to
come over on many technological difficulties, particularly when the adequate sensor isn’t
obtainable. So, the use of this idea for several years has been the replacement of the
material sensors by state observers, which allow the rebuilding of intern information (states,
unknown inputs, and unknown parameters) of a system using its model, known inputs and
measured outputs.
Among the solutions brought to the problem of the state estimation and output in the
presence of unknown inputs we have two approaches of construction of multiobserver that
emerged. The first one supposes a priori knowledge of information about these not
measurable inputs; in particular, the filter of Kalman [24] who allows to rebuild the state of
the system in the presence of measurement noise which is defined like unknown inputs, by
using a priori of statistical knowledge on these noises. The second approach proceeds
either by estimation of the unknown inputs or by their complete elimination from the
equations of the system [1, 2, 17, 19]. The observers with unknown inputs have attracted
the attention of many researchers that of Dassanake et al [29] who used unknown input
observers to detect and isolate sensor faults in a turbofan engine.

* Corresponding author: Sami ZEMMEL

a
Faculté des Sciences de Tunis, département de physique et d’électronique,
b
Ecole Supérieure de Technologie et d’Informatique,
*
Laboratoire Réseaux et Machines Electriques, Institut National des Sciences Appliquées et Technologies,
Tunisia
E-Mail address: mongi.besbes@gmail.comDepartment of mechanical engineering, ,National Institute of
Technology- Rourkela,

Copyright © JASE 2011 on-line: jase.esrgroups.org

 J. Automation & Systems Engineering 5-3 (2011): 112-129

The crucial problem, in the synthesis of the observers with unknown inputs, is the
convergence of the estimation error towards zero. Several works used the second method of
Lyapunov and their quadratic functions for the stabilization of the estimation error in the
case of the linear systems [10, 34] and in the case of the nonlinear systems [6]. However,
this method generates very conservative conditions of stability of the observer in particular
for certain classes of nonlinear systems such as the systems with parametric uncertainties
[16], the saturated systems and linear systems piecewise, which includes no information
about space state partition [20]. Whereas, the using of the no quadratic Lyapunov functions
like the polyquadratic functions and the piecewise quadratic functions, allows the reduce of
the conservatism of the quadratic method and leads to after less pessimistic results for the
stabilization and the control of the systems [5, 31]. For this reason, it is interesting to use
the piecewise quadratic Lyapunov functions for the stabilization of the estimation error in
the cases of a nonlinear multiobserver with unknown inputs which is used for the
diagnosis and the supervision of a nonlinear systems represented by multimodels [4, 21, 23,
29].
The paper is organized as follows. Section 2 gives a general structure of the considered
multimodel is presented. In Section 3, the considered structure of unknown multiobserver is
given and the main results are presented. Two approaches are presented for the
convergence of estimation error are given as a set of LMI with additional equality
constraints and rank constraints. We end the paper by an illustrative example which shows
the performance of the piecewise quadratic approach and a conclusion.

Notation: Throughout the paper, the following useful notation is used: (X)T denotes the
−1
transpose of the matrix X, (Y ) denotes the inverse of the matrix Y, (Z)¯ denotes the
pseudo inverse of the matrix Z and I denotes the n × n identity matrix.

2. Estimation of state and unknown inputs of a multimodel

The formalization of the estimation problem of state and unknown inputs rests on
multimodel approach. The considered system comprises known inputs and unknown inputs.
The reconstruction of the non measured system states and its unknown inputs is realized on
the base of information generated by a nonlinear observer called to unknown inputs widely
widespread for the state estimation of systems.

2.1. General structure of the multimodel

The model of system, supposed known, is taken in the following form [3]:
⎧ M
⎪ x(k + 1) = ∑ μi ( ξ (k ) ) ( Ai x(k ) + Bi u (k ) + Ri u(k ) )
⎨ i =1 (1)
⎪ y (k ) = Cx (k ) + Fu (k )
⎩
with:
⎧M
⎪ ∑ μi ( ξ (k ) ) = 1
⎨i =1 ∀i ∈ {1, ...., M }
⎪0 ≤ μ ( ξ ( k ) ) ≤ 1
⎩ i
where x(k) ∈ Rn is the state vector, u(k) ∈ Rm is the vector of the known inputs, u (k ) ∈ Rq is
the vector of unknown inputs and y(k) ∈ Rp represent the vector of measurable outputs.
For the ith local model Ai ∈ Rn×n is the state matrix, Bi ∈ Rn×m is the matrix of input, Ri ∈
Rn×q is the matrix of influence of the unknown inputs on the state x(t), F ∈ Rp×q is the
matrix of influence of the unknown inputs on the output y(k) with rank(F)=q and C ∈ Rp×n
is the matrix of output. Finally, ξ(k) represents the vector of decision depending on the

113
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

input and/or the measurable state variables. At every moment, µi(ξ(k)) indicate the degree
of activation of each local model for the global model. Choosing the number M of local
models of this multimodel can be intuitively achieved taking into account the number of
modes functioning observed of the system [11, 32]. However, determining the matrices Ai ,
Bi ,C, Ri and F needs the use of adapted estimation techniques parametric or techniques of
linearization [33].
In the following the considered problem concerns both the reconstruction of state variables
x(k) and the unknown inputs u (k ) , using only the available namely known inputs u(k)
and the measured outputs y(k).

2.2. General structure of the multiobserver

This paragraph clarifies the construction of the observer. This last shows an analytical
form resulting from the aggregation of the local linear observers [14, 26]. This one is
particularly adapted for studying the stability and the convergence property of the state
reconstruction error. The numerical aspects related to the determination of the gains of the
local observers will be also treated. The form of the multiobserver that is able to estimate
the state vector and the unknown input of the discrete multimodel (1) has the following
structure [28]:
⎧ M
⎨ i =1
( )
⎪ z (k + 1) = ∑ μi ( ξ (k ) ) Ni z (k ) + Gi1u (k ) + Gi 2 + Li y (k )
(2)
⎪ x (k ) = z (k ) − Ey (k )
⎩
It is about a general structure of observer, it is not only, where Ni ∈ Rn×n, Gi1 ∈ Rn×m,
Gi2 ∈ Rn, Li ∈ Rn×p and E are the gains matrices of the ith local observer with unknown input.
The variable z(k) is an intermediate variable allowing to deduce the estimated value from

the state x (k ) .
Obviously, the observer uses only the known variables u(k) and y(k), u (k ) being not
measured. The whole of these matrices must be given with a high degree of accuracy from
a numerical point of view in order to guarantee the convergence of the estimated state by
the observer towards the real state. For that, let us define the state estimation error:

e( k ) = x ( k ) − x ( k ) (3)

Starting from this definition and by using the expression of x (k ) given by the equation (2),
the expression of the error becomes:
e( k ) = Qx ( k ) − z ( k ) + EFu ( k ) (4)
with:
Q = I + EC (5)
Then, one expresses the temporal evolution of the state error in order to analyse its
convergence towards zero. Thus, at time (k + 1), the state estimation error is expressed:
M
i =1
(
e ( k + 1) = ∑ μi ( ξ ( k ) ) Q ( Ai x ( k ) + Bi u ( k ) + Ri u ( k ) ) − N i z ( k )
(6)
)
− Gi1u ( k ) − Gi 2 − Li y ( k ) + EFu ( k + 1)
Replacing y(k), z(k) and e(k) by their respective expressions given by (1), (2) and (3), the
state error takes the form , the state estimation error is written:

114
 J. Automation & Systems Engineering 5-3 (2011): 112-129

M 
e ( k + 1)= ∑ μi ( ξ ( k ) ) ⎡⎣ N i ( x ( k ) − x ( k ))
i =1
+ (QAi − N i Q − Li C ) x ( k ) + (QBi − Gi1 )u ( k ) (7)
+ ( QRi − N i EF − Li F ) u ( k ) − Gi 2 ⎤⎦ + EFu ( k + 1)
We assure that:
K i = N i E + Li (8)
Finally :
M
( ( )
e ( k + 1) = ∑ μi ( ξ ( k ) ) N i e( k ) + ( QAi − N i − K i C ) x ( k ) + QBi − Gi1 u ( k ) +
i =1 (9)
( QRi − K i F ) u (k ) − Gi 2 ) + EFu (k + 1)
If the following conditions are satisfied:
Ni = QAi − Ki C (10a)
Gi1 = QBi (10b)
QRi = Ki F (10c)
Gi 2 = 0 (10d)
EF = 0 (10e)
then the problem of the stabilization of the dynamic error of observation is reduced to the
M
stabilization ∑ μi (ξ ( k ) )N i e( k ) as the following expression indicates:
i =1
M
e(k + 1) = ∑ μi (ξ (k )) Ni e(k ) (11)
i =1
The stability conditions of the multiobserver (2) will emerge starting from the conditions
from stabilization of the observation error (11). It is important to note that the stability of
the local matrices Ni (i=1,…,M), does not guarantee the global stability of the
M
matrix N = ∑ μi (ξ ( k ) )N i .
i =1

Remark 1: The stabilization of the estimation error (11) can be made by three methods:
ƒ Quadratic method: This approach is developed in the paragraph (2.4.1) which uses
a function of Lyapunov of the form (17) and this last can exist if the inequality
(18) having a solution.
ƒ Piecewise quadratic method: This approach is developed in the paragraph (2.4.2)
which uses a function of Lyapunov of the form (31) and this last can exist if the
inequality (33) having a solution.
ƒ Polyquadratic method: It is another approach which uses a function of Lyapunov
of the following form:
⎛M ⎞
V ( e(k ) ) = e(k )T ⎜ ∑ µi (ξ (k ) ) Pi ⎟ e(k ) , Pi > 0 (12)
⎝ i =1 ⎠
where (Pi, i = 1,…,M) are symmetric positive definite matrices.
This last can exist if the following inequality having a solution [9]:
⎡ P ( Pj Ni )T ⎤
⎢ i ⎥ > 0, ∀(i, j ) ∈ {1,..., M } (13)
⎢⎣ Pj Ni Pj ⎥
⎦

115
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

In this paper we will develop the advantage of the piecewise quadratic approach which
offers a relieving compared to the quadratic approach proposed by Chadli et al. [17] (see
example 2).

2.3. Unknown input estimation

In the system (2), the unknown input appears with the matrix of influence Φ (k ) .
Let us define:
⎛M ⎞
∑ μ (ξ ( k )) Ri ⎟
Φ ( k ) = ⎜ i =1 i (14)
⎜ ⎟
⎝ F ⎠
To estimate the unknown input, it is necessary that the matrix Ф(k) is of full column rank
−
and its pseudo inverse Φ (k ) exists, otherwise this method is not applicable:

( )
−1
Φ − (k ) = Φ T (k )Φ (k ) Φ (k )T (15)
The unknown input can then be deduced in the following way:
⎛ M  ⎞
 − ⎜ x (k + 1) − ∑ μi (ξ (k )) ( Ai x (k ) + Bi u (k ) ) ⎟
u (k ) = Φ (k ) ⎜ i =1 ⎟ (16)
⎜  ⎟
⎝ y (k ) − Cx (k ) ⎠
(
We choose Φ (k ) of full column rank, it is to be able to reverse the matrix Φ T (k )Φ (k ) . )

Remark 2: The asymptotic convergence of u (k ) towards u (k ) is satisfied, if the
stabilization of the observation error (11) of the multiobserver (2) and the condition on the
rank of the matrix Ф(k) are verified.

2.4. Global convergence of the multiobserver

For the stabilization of the observation error (11), we propose two approaches based on
the use of quadratic Lyapunov function in the first step and the piecewise quadratic
Lyapunov function in second step.

2.4.1. Global convergence of the multiobserver by the quadratic approach

In this part, the stabilization of the dynamic error (11) is based on the use of a quadratic
Lyapunov function of the form:
V (e(k )) = e(k )T Pe(k ) , P >0 (17)
From this function (15) we can study the stability of the observation error (11).
M
Theorem 1 [13] : The system of form e(k + 1) = ∑ μi (ξ (k )) Ni e(k ) is globally
i =1
asymptotically stable
if ∃ P > 0 such as ∀i ∈ {1,..., M } NiT PNi − P <0 (18)
■
Thus, constraints (18) and (10) allow the complete synthesis of multiobserver (2) for the
multimodel (1).
Theorem 2 [17]: The state estimation error between the multimodel (1) and the unknown
input multiobserver (2) converges globally asymptotically towards zero if there exists
matrices P > 0 , S and Wi such that the following conditions hold ∀i ∈ {1, ...,M} :

116
 J. Automation & Systems Engineering 5-3 (2011): 112-129

⎛ P ( PAi + SCAi − Wi C )T ⎞⎟ > 0

⎜ (19a)
⎜ PA + SCA − W C P ⎟
⎝ i i i ⎠
( P + SC ) Ri = Wi F (19b)
SF = 0 (19c)
Multiobserver (2) is then completely defined by
E = P −1 S (20a)
−1
Gi1 = ( I + P SC ) Bi (20b)
Gi 2 = 0 (20c)
N i = ( I + P −1SC ) Ai − P −1Wi C (20d)
−1
Li = P Wi − Ni E (20e)

Proof: Using equation (18), the hypothesis P>0 and using the Schur complement, one can
easily deduce that
⎛ P ( PNi )T ⎞
⎜ ⎟>0 (21)
⎜ PN P ⎟⎠
⎝ i

After using equations (10a) and (10c), the preceding inequality becomes
⎛ P ( PAi + PECAi − PKi C )T ⎞
⎜⎜ ⎟⎟ > 0 (22)
⎝ PAi + PECAi − PKi C P ⎠
However, this last inequality (22) is a bilinear matrix inequality BMI in synthesis
variables P, E and Ki. In order to convert these conditions into an LMI formulation, we
consider the following changes of variables:
Wi = PK i (23)
S = PE (24)
After using the new variables of equations (23) and (24), inequality (22) becomes
⎛ P ( PAi + SCAi − Wi C )T ⎞
⎜⎜ ⎟⎟ > 0 (25)
⎝ PAi + SCAi − Wi C P ⎠
The two equalities constraints (19b) and (19c) are obtained by pre-multiplying the
last two constraints (10e) and (10g) by P > 0 with the change of variable (23) and (24):
⎧ PQRi = PK i F ⎧( P + SC ) Ri = Wi F
⎨ ⇒ ⎨ ∀i ∈ {1, ...,M}
⎩ PEF = 0 ⎩ SF = 0
Therefore classical numerical tools may be used to solve LMI problem (19a) subject to
linear equality constraints (19b) and (19c). After having solved this problem, the different
gains matrices defining the multiobserver (2) Ni, Li, Gi1, Gi2 and E can be deduced from the
knowledge of P, S and Wi as given in equations (20). This completes the proof.
■
2.4.1.1. Determination of the multiobserver gain matrices
To determine the multiobserver (2) gain matrices by quadratic approach, we propose to
follow the steps of the following algorithm:
Step 1: Determination of the matrices P, S and Wi ∀i ∈ {1, ...,M}.
We solve the Linear Matrix Inequalities (19a) in synthesis variables P, S and Wi subject to
linear equality constraints (19b) and (19c). This problem can be solved by LMITOOL [15].
Step 2: Determination of the gains matrices Ni, Li, Gi1, Gi2 and E ∀i ∈ {1, ..., M } .

117
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

After the knowledge of the matrices P, S, and Wi, we determine the other gains matrices of
equations (20) defining the multiobserver (2).

Remark 3: If F is a scalar, we can deduce the following relations ∀i ∈ {1, ...,M}:

S =0 (26a)
E=0 (26b)
Q=I (26c)
Gi1 = Bi (26d)
Gi 2 = 0 (26e)
N i = Ai − P −1Wi C (26f)
−1
Li = P Wi (26g)
−1
Ri = P Wi F (26h)

2.4.1.2. Simulation Examples:

In this section, two examples are presented in order illustrate the utilisation limits of the
quadratic approach. The first is an academic example and the second describes the
conservatism of this approach.

2.4.1.2.1. An academic example:

Consider the following multiple model:
⎧ 2
⎪ x(k + 1) = ∑ µi (ξ (k )) ( Ai x(k ) + Bi u (k ) + Ri u (k ) )
⎨ i =1 (27)
⎪ y (k ) = Cx(k ) + Fu (k )
⎩
with:
⎛ x1 ⎞
⎜ ⎟ ⎛y ⎞
x = ⎜ x2 ⎟ and y = ⎜ 1 ⎟
⎜x ⎟ ⎝ y2 ⎠
⎝ 3⎠
The activation functions are the following form:
⎧ 1 − tanh(u (k ))
⎪ μ1 (u (k )) =
⎨ 2 (28)
⎪ μ 2 (u (k )) = 1 − μ1 (u (k ))
⎩
The numerical values of the matrices are the following:
⎛ −0.3 0.2 0.3 ⎞ ⎛ 0.6 −0.5 0.2 ⎞ ⎛ 1 ⎞ ⎛ 1 ⎞
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
A1 = ⎜ 0.3 0 0.5 ⎟ , A2 = ⎜ −0.4 0.1 0.3 ⎟ , B1 = ⎜ 1 ⎟ , B2 = ⎜ 1.25 ⎟
⎜ 0.2 0.1 0.6 ⎟ ⎜ 0.1 0.2 0.3 ⎟⎠ ⎜ 0.5 ⎟ ⎜ 0.5 ⎟
⎝ ⎠ ⎝ ⎝ ⎠ ⎝ ⎠
⎛ 39.86 ⎞ ⎛ 6.9 ⎞
⎛ 0 1 0⎞ ⎜ ⎟ ⎜ ⎟ ⎛ −1⎞
C = 10−2 ⎜ ⎟ , R1 = ⎜ − 23.30 ⎟ , R 2 = ⎜ 60.46 ⎟ , F = ⎜ ⎟
⎝ −1 1 1 ⎠ ⎜ 26.85 ⎟ ⎜ −1.51⎟ ⎝1⎠
⎝ ⎠ ⎝ ⎠

By using the quadratic approach and the resolution method presented at the paragraph
(2.4.1.1), the numerical values of all gain matrices are as follows:

118
 J. Automation & Systems Engineering 5-3 (2011): 112-129

⎛ 79.7 −0.26 3.41 ⎞ ⎛ −0.80 2.45 ⎞ ⎛ −2.06 −1.51⎞

⎜ ⎟ 3⎜ ⎟ 3⎜ ⎟
P = ⎜ −0.26 71.91 −7.94 ⎟ , W1 = 10 ⎜ 2.49 0.57 ⎟ , W2 = 10 ⎜ −1.92 2.45 ⎟
⎜ 3.41 −7.94 78.06 ⎟⎠ ⎜ −0.79 1.61 ⎟⎠ ⎜ 1.02 0.46 ⎟⎠
⎝ ⎝ ⎝
⎛ 18.92 18.92 ⎞ ⎛ 0.0024 0.0007 0.0032 ⎞ ⎛ 0.4040 −0.0402 0.3946 ⎞
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
S = ⎜ 20.74 20.74 ⎟ , N1 = ⎜ 0.4076 −0.0413 0.3996 ⎟ , N 2 = ⎜ −0.0512 0.0054 −0.0506 ⎟
⎜ 6.12 6.12 ⎟⎠ ⎜ 0.4057 −0.0405 0.3967 ⎟⎠ ⎜ 0.2022 −0.0200 0.1972 ⎟⎠
⎝ ⎝ ⎝

⎛ −9.7495 29.9794 ⎞ ⎛ −26.5981 −19.4174 ⎞ ⎛ 1.0035 ⎞ ⎛ 1.0047 ⎞

⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
L1 = ⎜ 33.7870 10.3039 ⎟ , L2 = ⎜ −25.5211 35.2852 ⎟ , G11 = ⎜ 1.0045 ⎟ , G21 = ⎜ 1.2560 ⎟
⎜ −6.4608 20.3391 ⎟ ⎜ 11.6825 10.2843 ⎟ ⎜ 0.5015 ⎟ ⎜ 0.5020 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠

Fig.1 represent the time evolution of known inputs u(k) and unknown inputs u (k ) . Fig.2

shows, on the same graph, the evolution of the unknown input u (k ) and its estimate u (k ) .
Fig.3 represent the state estimation error with the initial conditions x0 = (1 0 0)T and

x0 = (0 0 0)T . Excepted around the time origin, the unknown signal estimated perfectly
matches the true one.

u (k ) u (k )

Figs.1: input known u (k ) and input unknown u (k )

119
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…


Fig.2 : u (k ) and u (k )


Fig.3 : Estimation errors ei = xi − xi , i ∈ {1, 2,3}

120
 J. Automation & Systems Engineering 5-3 (2011): 112-129

2.4.1.2.2. Example 2 « Conservatism of the quadratic approach»

Let us consider the following discrete multimodel, composed of three local models and
comprising one single output and three states:
⎧ 2
⎪ x(k +1) = ∑ µi (ξ (k )) ( Ai x(k ) + Bi u (k ) + Ri u (k ) )
⎨ i =1 (29)
⎪ y (k ) = Cx(k ) + Fu (k )
⎩
In this example, the vector of variable decision ξ(k) is equal to the vector of known inputs
u(k). The numerical values of the matrices Ai, Bi, C and F are as follows:
⎛ 1 0⎞ ⎛ 0.1 −0.4 ⎞ ⎛ 1 ⎞ ⎛ 1 ⎞
A1 = ⎜ ⎟ , A2 = ⎜ ⎟ , B1 = ⎜ ⎟ , B2 = ⎜ ⎟ ,
⎝ 0.1 1 ⎠ ⎝ 0.7 0.2 ⎠ ⎝ 0.25 ⎠ ⎝ 0.5 ⎠
⎛ 1 ⎞ ⎛ 2 ⎞
R1 = ⎜ ⎟ , R2 = ⎜ ⎟ , C = ( 0 1) , F = 5
⎝ 0.5 ⎠ ⎝ 0.5 ⎠
The multiobserver able to estimate the multimodel (27) state is as follows:
⎧
( )
2
⎪ z (k +1) = ∑ µi (ξ (k )) Ni x(k ) + Gi1u (k ) + Gi 2 + Li y (k )
⎨ i =1 (30)
⎪ x (k ) = z (k ) − Ey (k )
⎩
The conditions of quadratic stabilization of theorem 2 fails to prove the stabilization of the
multiobserver (30), which shows that no quadratic function having the form (17) can exist.
The quadratic approach becomes more and more conservative in the following cases:
ƒ When the number of local models is very important, this is due the difficulty to
find one matrix P satisfying all (19) inequalities and equalities.
ƒ When the multimodel have local models saturated (the eigenvalues of matrices Ai
are closer to 1) like the local model number 1 (eigenvalues with A1 = {1, 1 }).

2.4.2. Global convergence of the multiobserver by the piecewise quadratic approach

In this part, the stabilization of the error dynamics (11) depended en the use of
piecewise quadratic Lyapunov function of the form:
(
V ( x(k ) ) = max V1 ( x(k ) ) , ...., Vi ( x(k ) ) , ...., Vn ( x(k ) ) ) (31)
witch:
Vi ( x(k ) ) = x (k )T Pi x(k ) , Pi > 0 , i ∈ {1,..., M } (32)
where (Pi, i = 1,…,M) are symmetric positive definite matrices.
The goal of this second approach is to cure the conservatism of the quadratic approach by
formulating new less constraining conditions of stabilisation of the observation error (11)
by using the following theorem:
Theorem 3 [9]: We suppose that there are symmetric positive definite matrices Pi and
scalars τijk ≥ 0 such as
M
NiT Pj Ni − Pj + ∑ τijk ( Pj − Pk ) < 0 , ∀ i, j , k ∈ {1,..., M } and j ≠ k (33)
k =1
then the observation error (11) is globally asymptotic stable.
■

Thus, constraints (33) and (10) allow the complete synthesis of multiobserver (2) for the
multimodel (1) with unknown inputs.
Proposition 1: The state estimation error between the multimodel (1) and the unknown
input multiobserver (2) converges globally asymptotically towards zero, if all the pairs (Ai,

121
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

C) are observable and if there exist symmetric positive definite matrices P(i , j , k ) and
matrices Mj and Hji of appropriate dimensions such that the following conditions hold
∀(i, j , k ) ∈ {1,..., M } and j ≠ k :
⎛ M T⎞
⎜ Pj − ∑ τijk ( Pj − Pk ) ( Pj Ai + M j CAi − H ji C ) ⎟
⎜ k =1 ⎟>0 (34a)
⎜ Pj Ai + M j CAi − H ji C Pj ⎟
⎝ ⎠
rank(P H i ) = rank( P ) (34b)
( Pj + M j C ) Ri = H ji F (34c)
M jF = 0 (34d)
Multiobserver (2) is then completely defined by
E = Pj−1M j (35a)
Gi1 = ( I + EC ) Bi (35b)
Gi 2 = 0 (35c)
−
Ni = ( I + EC ) Ai − ( P H i )C (35d)
−
Li = ( P H i ) − N i E (35e)
with:
⎡ P1 ⎤ ⎡ H1i ⎤
⎢P ⎥ ⎢H ⎥
⎢ 2⎥ ⎢ 2i ⎥
P = ⎢ . ⎥ and H i = ⎢ . ⎥ , ∀ i ∈ {1,..., M } .
⎢ ⎥ ⎢ ⎥
⎢ . ⎥ ⎢ . ⎥
⎢ PM ⎥ ⎢ H Mi ⎥
⎣ ⎦ ⎣ ⎦
■

Proof:
The studied multiobserver is the result of M aggregated linear observer. The
observability of multimodel (1) is guaranteed if every local model is observable, in other
terms is (Ai, C) pairs are observable [25].
After the multiplication on the left and on the right by -1, the matrix inequality (33) can be
rewritten in the following form:
M
Pj − ∑ τijk ( Pj − Pk ) − NiT Pj Ni > 0, ∀ (i, j ) ∈ {1,..., M } and j ≠ k (36)
k =1
We multiply the matrix Pj of the last term of the matrix inequality (36) by ( Pj−1 Pj ) , we
obtain:
M
Pj − ∑ τijk ( Pj − Pk ) − NiT Pj Pj−1Pj Ni > 0, ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k (37)
k =1
T T T T T T
Knowing that ( AB ) = B A and ( A + B ) = A + B , constraint (37) can be
rewritten as follows:
M
Pj − ∑ τijk ( Pj − Pk ) − ( Pj Ni )T Pj−1 ( Pj Ni ) > 0 , ∀( i, j , k ) ∈ {1,..., M } and j ≠ k (38)
k =1

122
 J. Automation & Systems Engineering 5-3 (2011): 112-129

The use of the complement of Schur [27] allows expressing the constraints (38) in the
following equivalent forms:
⎛ M T⎞
⎜ Pj − ∑ τi jk ( Pj − Pk ) ( Pj Ni ) ⎟
⎜ k =1 ⎟ > 0 ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k (39)
⎜ Pj Ni Pj ⎟
⎝ ⎠
After using equations (5) and (10a) into inequality (39), we obtain:
⎛ M T⎞
⎜ Pj − ∑ τi jk ( Pj − Pk ) ( Pj Ai + Pj ECAi − Pj Ki C ) ⎟
⎜ k =1 ⎟ > 0 ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k
⎜ Pj Ai + Pj ECAi − Pj Ki C Pj ⎟
⎝ ⎠
(40)
However, this last inequality (40) is a bilinear matrix inequality BMI in synthesis
variables Pj, E, Ki and τ ijk . We consider the following changes of variables:
H ji = Pj K i (41)
M j = Pj E (42)
Using the new variables of equations (41) and (42), inequality (40) becomes
⎛ M T⎞
⎜ Pj − ∑ τi jk ( Pj − Pk ) ( Pj Ai + M j CAi − H ji C ) ⎟
⎜ k =1 ⎟ > 0 ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k
⎜ Pj Ai + M j CAi − H ji C Pj ⎟
⎝ ⎠
(43)
The tow equalities constraints (34c) and (34d) are obtained by pre-multiplying the last two
constraints (10c) and (10e) by Pj with the change of variable (41) and (42):
⎧⎪ Pj QRi = Pj Ki F ⎧⎪ Pj ( I + EC ) Ri = Pj Ki F ⎧⎪( Pj + M j C ) Ri = H ji F
⎨ ⇒ ⎨ ⇒ ⎨
⎩⎪ Pj EF = 0 ⎩⎪ M j F = 0 ⎩⎪ M j F = 0
∀i ∈ {1, ...,M}
The matrix inequalities (40) obtained is a BMI with Pj and Ki that we must linearize. The
technique chosen is based on the changes of variables (41).
With the expression of Hji (41) ∀(i, j , k ) ∈ {1,..., M } and j ≠ k , we can write the relations as
follows:
H11 = P1 K1 ⎫ ⎡ H11 ⎤ ⎡ P1 ⎤
H 21 = P2 K1 ⎪⎪ ⎢⎢ H 21 ⎥⎥ ⎢⎢ P2 ⎥⎥
⎪
. ⎬ ⇒ ⎢ . ⎥ = ⎢ . ⎥ K1 (44)
⎪ ⎢ ⎥ ⎢ ⎥
. . ⎥ ⎢ . ⎥
⎪ ⎢
H M 1 = PM K1 ⎪⎭ ⎣ H M 1 ⎦⎥ ⎣⎢ PM ⎦⎥
⎢
H12 = P1 K 2 ⎫ ⎡ H12 ⎤ ⎡ P1 ⎤
H 22 = P2 K 2 ⎪⎪ ⎢⎢ H 22 ⎥⎥ ⎢⎢ P2 ⎥⎥
⎪
. ⎬ ⇒ ⎢ . ⎥ = ⎢ . ⎥ K2 (45)
⎪ ⎢ ⎥ ⎢ ⎥
. . ⎥ ⎢ . ⎥
⎪ ⎢
H M 2 = PM K 2 ⎪⎭ ⎣ H M 2 ⎦⎥ ⎣⎢ PM ⎦⎥
⎢

123
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

H1M = P1 K M ⎫ ⎡ H1M ⎤ ⎡ P1 ⎤
H 2 M = P2 K M ⎪⎪ ⎢⎢ H 2 M ⎥⎥ ⎢⎢ P2 ⎥⎥
⎪ ⎢
. ⎬⇒ . ⎥ = ⎢ . ⎥ KM (46)
⎪ ⎢ ⎥ ⎢ ⎥
. . ⎥ ⎢ . ⎥
⎪ ⎢
H MM = PM K M ⎭⎪ ⎣⎢ H MM ⎦⎥ ⎣⎢ PM ⎦⎥
We pose:
⎛ P1 ⎞ ⎛ H1i ⎞
⎜ ⎟ ⎜ ⎟
P
⎜ 2 ⎟ ⎜ H 2i ⎟
P = ⎜ . ⎟ and H i = ⎜ . ⎟ ∀i ∈ {1,..., M }
⎜ ⎟ ⎜ ⎟
⎜ . ⎟ ⎜ . ⎟
⎜P ⎟ ⎜H ⎟
⎝ M⎠ ⎝ Mi ⎠
(47)
From the equations (44), (45), (46) and (47) we can deduce:
H i = PKi ⇒ Ki = P − H i ∀i ∈ {1,..., M } (48)
However the gains Ki of the equation (48) can exist only if the following condition of rank
is satisfied [8, 22] :
rank(P H i ) = rank( P ) ∀i ∈ {1,..., M } (49)
The conditions (34a) (34b), (34c) and (34d) of proposition 1 are verified.
Therefore classical numerical tools may be used to solve LMI problem (34a) subject to
linear equality constraints (34b), (34c) and (34d). After having solved this problem, the
different gains matrices defining the multiobserver (2) Ni, Li, Gi1, Gi2 and E can be deduced
from the knowledge of P(i,,j,k), Hji and Mi as given in equations (35). This completes the
proof of proposition 1.
■
Remark 4: The matrix inequality (34a) presents two not convex problems. The first
appeared in the bilinear of the matrix inequality (34a) in ( τ ijk , Pj ) and
( τ ijk , Pk ) ∀(i, j , k ) ∈ {1,..., M } and j ≠ k . The second is shown on the level of the
resolution of matrix inequality (34a) under the constraint of rank (34b).

2.4.2.1. Treatment of the not convex problems

Obtaining the solutions to the matrix inequality (34a) requires first of all the treatment
of not convexities of two problems presented previously. For that we give some methods
used by researchers to solve this kind of problems.
2.4.2.1.1. Linearization of matrix inequality Bilinear
The linearization of the bilinear matrix inequality (34a) in ( τ ijk , Pj ) and
( τ ijk , Pk ) ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k can be made by the application of an iterative
algorithm of initialization of the positive scalars τ ijk ≥ 0 or by their fixing [18].
2.4.2.1.2. Treatment of the constraint of rank
After having to linearize the matrix inequality (34a), it is necessary to solve the latter
with the constraint of rank (34b) and that is a not convex problem difficult to solve.
However, several authors proposed iterative algorithms to release this kind of problems like
[12, 22]. As an example, in [18] the author proposed an iterative algorithm of relaxation
which consists in solving a LMI with a constraint of rank.

124
 J. Automation & Systems Engineering 5-3 (2011): 112-129

2.4.2.2. Determination of the multiobserver gain matrices

To determine the multiobserver (2) gain matrices by piecewise quadratic approach, we
propose to follow the steps of the following algorithm:
Step 1: Determination of the matrices P(i , j ,k ) , H ji and M j , ∀(i, j , k ) ∈ {1,..., M } and
j≠k.
We solve the Linear Matrix Inequalities (32a) in synthesis variables P(i , j ,k ) , H ji and
M j subject to linear equality constraints (34b), (34c) and (34d). This problem can be solved
by LMITOOL [29].
Step 2: Determination of the gains matrices Ni, Li, Gi1, Gi2 and E , ∀(i, j , k ) ∈ {1,..., M } and
j≠k.
After the knowledge of the matrices P(i , j ,k ) , H ji and M j , we determine the other gains
matrices of equations (35) defining the multiobserver (2).
Remark 5: For a matrix F equal to a scalar, we can deduce the following relations,
∀(i, j , k ) ∈ {1,..., M } and j ≠ k .:
Mj =0 (50a)
E=0 (50b)
Q=I (50c)
Gi1 = Bi (50d)
Gi 2 = 0 (50e)
Ni = Ai − P − H i C (50f)
Li = ( P − H i ) = K i (50g)
−
Ri = P H i F (50h)
These relations are given after the resolution of the linear matrix inequality (34a) for
M j = 0 in synthesis variables P(i , j ,k ) and H ji .
⎛ M T⎞
⎜ Pj − ∑ τi jk ( Pj − Pk ) ( Pj Ai − H ji C ) ⎟
⎜ k =1 ⎟ > 0 ∀ (i, j , k ) ∈ {1,..., M } and j ≠ k (51)
⎜ Pj Ai − H ji C Pj ⎟
⎝ ⎠

Remark 6: It's obvious that the conditions of proposition 1 are less conservative than the
conditions depending on the use of a single Lyapunov function. The quadratic conditions
stabilization is considered like a particular case of (34a) by composing P(i , j , k ) = P0 it
follows that Pj − Pk = 0 and the condition of rank (34b) becomes trivial.

2.4.2.3. Example 3: Application of the piecewise quadratic approach

Let us consider the discrete multi-model of example 2; the conditions of stabilization of
proposition 1 prove the stability of the multiobserver (30), what shows that a piecewise
quadratic function of the form (31) exists.
By applying the method of resolution presented at paragraph (2.4.2.2), we showed global
convergence of the multiobserver (30).
The resolution of the conditions of proposition 1 leads to the following result:
With the choice of the parameters

125
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

τ 11 = 3, τ 12 = 1, τ 21 = 0.3, τ 22 = 0

we obtain four LMI in P1, P2,M1, M2 H11, H12, H21 and H22 which must be solved under the
following rank constraints:
⎡P ⎤ ⎡H ⎤ ⎡H ⎤
rank = ⎢ 1 ⎥ = rank ⎢ 11 ⎥ = rank ⎢ 12 ⎥ (52)
⎣ P2 ⎦ ⎣ H 21 ⎦ ⎣ H 22 ⎦
The resolution of the conditions of proposition 1 leads to the following result:
⎡5683.5055 908.3627 ⎤ ⎡9070.6173 3681.729 ⎤
P1 = ⎢ ⎥ , P2 = ⎢ ⎥
⎣ 908.3627 4189.8622 ⎦ ⎣ 3681.729 7455.7043⎦
M1 = M 2 = 0
⎡1182.1192 ⎤ ⎡1907.3198⎤ ⎡ 2364.2385⎤ ⎡3814.6396 ⎤
H11 = ⎢ ⎥ , H 21 = ⎢ ⎥ , H12 = ⎢ ⎥ , H 22 = ⎢ ⎥
⎣ 391.3313 ⎦ ⎣ 745.5706 ⎦ ⎣ 782.3313 ⎦ ⎣1491.1412 ⎦
We can to check that:
⎡P ⎤ ⎡H ⎤ ⎡H ⎤
rank = ⎢ 1 ⎥ = rank ⎢ 11 ⎥ = rank ⎢ 12 ⎥ = 2
⎣ P2 ⎦ ⎣ H 21 ⎦ ⎣ H 22 ⎦

After the knowledge of the matrices P1, P2,M1, M2 H11, H12, H21 and H22 we deduce the
other matrices from the profits defining the multiobserver (30):
⎡1 ⎤ ⎡1 ⎤
G11 = B1 = ⎢ ⎥ , G21 = B2 = ⎢ ⎥
⎣ 0.5⎦ ⎣ 0.5⎦
−
− ⎛ ⎡P ⎤ ⎞ ⎡ H ⎤ ⎡0.2 ⎤
K1 = P H1 = ⎜ ⎢ 1 ⎥ ⎟ ⎢ 11 ⎥ = ⎢ ⎥,
⎜ P ⎟ H 21
⎝⎣ 2⎦⎠ ⎣ ⎦ ⎣0.05⎦
−
− ⎛ ⎡P ⎤ ⎞ ⎡ H12 ⎤ ⎡0.4 ⎤
K2 = P H 2 = ⎜ ⎢ 1 ⎥ ⎟ ⎢H ⎥ = ⎢ ⎥
⎜ P ⎟
⎝⎣ 2⎦⎠ ⎣ 22 ⎦ ⎣0.1 ⎦
⎡ 0.8 0⎤ ⎡ −0.3 −0.4 ⎤
N1 = A1 − K1C = ⎢ ⎥ , N 2 = A2 − K 2C = ⎢ ⎥
⎣ 0.05 1⎦ ⎣ 0.6 0.2 ⎦
⎡0.2 ⎤ ⎡0.4 ⎤
With E=0 ⇒ Li = Ki L1 = K1 = ⎢ ⎥ and L2 = K 2 = ⎢ ⎥
⎣0.05⎦ ⎣0.1 ⎦
Figs.4 represent respectively the evolution of the activation functions, the inputs known
u(k) and unknown u (k ) . As for the Fig.5, they show the state estimation errors (xi(k)-

xi (k ) , i={1, 2}) as well as the unknown input u (k ) of the multimodel and their

estimation u (k ) . It is noted that the estimation quality is satisfactory except in the vicinity
of the time origin; that is due to the choice of the multiobserver (30) initial

values x0 = (1 1 0)T and x0 = (2 −1 0)T .

126
 J. Automation & Systems Engineering 5-3 (2011): 112-129

µ1 ( u ( k )) and µ 2 ( u ( k ))

u (k )
u (k )
Fig .4: Functions of activation and input known u(k) and unknown u (k ) .

 
x1 ( k ) − x1 ( k ) x 2 (k ) − x 2 (k )


u ( k ) et u ( k )
Fig. 5 : Results of the estimates.

127
 S Zemmel et al.: Improvement of the Performances of a Multiobserver for the Reconstruction of State and…

The simulation results presented by the Fig.5 show that the state estimation as well as the
unknown inputs of the multimodel are very satisfying except at the origin of time. This is
due to the choices of the initial conditions.

3. Conclusions and prospects

In this article, we presented two stabilization approaches of a multiobserver with
unknown inputs for a nonlinear system describes by a discrete multimodel with decision
variables measurable. The first approach is based on the use of the Lyapunov quadratic
functions, the conditions obtained of this approach for the convergence of the multi-
observer are often easy to obtain but they appear pessimistic. The second approach
suggested for the stabilization of the observation error, is based on the use of the Lyapunov
piecewise quadratic functions. The conditions obtained of the multiobserver convergence
are given in the form of Bilinear Matrices Inequalities BMI that we can linearize by the
technique of change of variables and easily solve them with the numerical tools traditional.
This second approach appears less conservative than the first.
Two illustrative examples are proposed to proof the efficiency and the utility of each of
the two approaches. The first example presented showed that the quadratic approach is
interesting from point of view of the practical implementation for the supervision and the
diagnosis of the industrial processes. The second example put emphases on the important
contribution of the piecewise quadratic approach compared to the quadratic approach for
the states estimation and unknown inputs of a nonlinear system represented by a discrete
multimodel.
The conditions obtained of the two approaches proposed, concern only the study of the
multiobserver stabilization with unknown inputs. The poles placement problem, thus the
state estimation and unknown inputs in cases where the decision variables are not
measurable, can constitute an interesting prospect for this work.

Acknowledgement:
This research was supported by the Laboratory Networks and Machines Electric of
National institute of Sciences Applied and Technologies of Tunis.

References
[1] A. Akhenak, M. Chadli, J. Ragot and D. Maquin, 2003. State estimation via Multiple observer with unknown
input. Application to the three tank system, 5th IFAC Symposium on Fault Detection Supervision and Safety
for Technical Processes, Safeprocess, pp. 245-251, Washington, USA.
[2] A. Akhenak, M. Chadli, J. Ragot, and D. Maquin. 2006. Unknown input observer based-approach -
application to secure communications, First Conference on Analysis and contol of Chaotic Systems, volume
1, part 1.
[3] A. Akhenak, M. Chadli, J. Ragot and D. Maquin, 2004. State estimation of uncertain multiple model with
unknown inputs, in:43rd IEEE Conference on Decision and Control,Atlantic, Paradise Island, Bahamas, vol.
4, pp. 3563-3568.
[4] A. Boukhris, G. Mourot and J. Ragot, 1999. Sensor fault diagnosis for urban sewer network management, In
Computers and Computational Engineering in Control. N. E. Mastorakis, Electrical and Computer
Engineering series, World Scientific Engineering Society.
[5] A. Jadbabaie, 1999. A reduction in conservatism in stability and L2 gain analysis of Takagi-Sugeno fuzzy
systems via linear matrix inequalities. In Proc. Of the IFAC, China. pp. 285-289.
[6] D. Marx, B. Dept, 2008. Unknown input observers for switched nonlinear discrete time descriptor systems,
Automatica Contol, IEEE Transactions on publication data.
[7] E. Maherzi, M. Besbes, M. Ellouze and R. Mhiri, 2007. Polyquadratic stabilization of a multiinputs
multimodel with quantified commands. INTERNATIONNAL JOURNAL OF MATHEMATICS AND
COMPUTERS IN SIMULATION. Issue 4, Vol. 1, pp. 344-349.
[8] H. Souley Ali, M. Zasadzinski, H. Rafaralahy, and M. Darouach, 2002. H1 functional unbiased filtering for
uncertain bilinear systems. In Proc. Triennal IFAC World Congress, Barcelona, Spain.
[9] V J. Daafouz and J. Bernussou, 2001. Parameter dependent lyapunov functions for discrete time systems with
time varying parametric uncertainties. Systems and Control Letters, 43/5:355 359.

128
 J. Automation & Systems Engineering 5-3 (2011): 112-129

[10] J. Moreno, 2001. Quasi-Unknown input observers for linear systems, IEEE Conference on Decision and
Control, pp. 732-737.
[11] K. Gasso, G. Mourot et J. Ragot, Structure identification in multiple model representation : elimination and
merging of local models, IEEE Conference on Decision and Control, Vol. 3, pp. 2992-2997, 2001.
[12] K. Tanaka and M. Sugeno. Stability analysis and design of fuzzy control systems. Uzzy Sets and Systems,
45(2) :135-156, 1992.
[13] K. Tanaka, T. Ikeda and Y. Y. He, 1998. Fuzzy regulators and fuzzy observers : relaxed stability conditions
and LMI-based design, IEEE Trans. Fuzzy Systems, Vol. 6 (1), pp. 250-256.
[14] 14. K. Seog-Joo, M. Young-Hyun and K. Soonman, 2006. Simultaneous stabilization by static output
feedback: a rank-constrained LMI approach. Proceedings of the 45th IEEE Conference on Decision & Control
Manchester Grand Hyatt Hotel San Diego, CA, USA.
[15] L. Vandenberghe, S. Boyd, 1996. Semidefinite programming, SIAM Review 38 (1) (1996) 49-95.
[16] M. Chadli, A. Elahajjaji, 2006. Observer-based robust fuzzy control of nonlinear systems with parametric
uncertainties – comment on, Fuzzy Sets and Systems Journal 157 (9) (2006) 1276-1281.
[17] M. Chadli, A. Akhenak, J. Ragot and D. Maquin, 2009. State and unknown input estimation for discrete time
multiple mode, Journal of the Franklin Institute 346 (2009) 593-610.
[18] M. Chadli, D. Maquin and J. Ragot, 2002. Nonquadratic stability of Takagi-Sugeno systems. In Proc. Of the
IEEE 41th CDC, Las Vegas, Nevada.
[19] M. darouach, M. Zasadzinski and S.J. Xu, 1994. Full-order observers for linear systems with unknown inputs,
IEEE Transactions on Automatic Contol 39 (3) (1994) 606-609.
[20] M. Johansson. Piecewise Linear Control Systems. Thesis of Doctorate, Lund Institue of Technology, 1999.
[21] P. Amann, J. M. Perronne, G. L. Gissenger and P. M. Frank, 2001. Identification of fuzzy relational models
for fault detection, Control Engineering Practice, Vol. 9 (5), pp. 555-562.
[22] P. Apkarian and H. D Tuan, 2000. Robust control via concacave minimisation : local and global algorithms.
IEEE transactions and Automatic Control, vol. 45. no. 2, pp. 299-305.
[23] R. Dixon, 2004. Observer-based FDIA : application to an electromechanical positioning system, Control
Engineering Practice, Vol. 12, pp. 1113-1125.
[24] R.E. Kalman, 1960. A new approach to linear filtering and prediction problems. Transactions of the ASME
Journal of Basic Engineering, 82 :35–45.
[25] R. E. Kalman and J. E. Betram, 1960. Control system analysis and design via the "second method" of
Lyapunov: Continuous-time system, ASME journal of Basic Engineering, Vol. 82, pp. 371-393.
[26] R. J. Patton, J. Chen and C. J. Lopez-Toribio, 1998. Fuzzy observer for nonlinear dynamic systems fault
diagnosis, IEEE Conference on Decision and Control, Vol. 1, pp. 84-89.
[27] S. Boyd, L. El Ghaoui, E. Feron and V. Balakrishnan, 1994. Linear Matrix Inequalities in System and Control
Theory. Philadelphia : SIAM.
[28] S.-K. Chang, W.-T. You and P.-L. Hsu, 1994. Design of general structured observers for linear systems with
unknown inputs, Journal of the Franklin Institute 334 (2) (1997) 1025-1030.
[29] S. K. Dassanake, G. L. Balas and J. Bokor, 2000. Using unknown input observers to detect and isolate sensor
faults in a turbofan engine, Digital Avionics Systems Conferences, Vol. 7, pp. 6E51-6E57.
[30] S. Zemmel, M. Besbes and E. Meherzi, 2008. Experimental identification of an industrial nonlinear process
(Case of a drying Blower). JTEA’2008, Hammamet, Tunis.
[31] T. A. Johansen, 2000. Computation of Lyapunov function for smooth nonliear systems using convex
optimization. Automatica, no. 36, pp. 1617-1626.
[32] T. A. Johansen and R. Babuska, 2003. Multiobjective Identification of Takagi-Sugeno Fuzzy Models, IEEE
Trans. on Fuzzy Systems, Vol. 11 (6), pp. 847-860.
[33] T.A. Johansen, R.Shorten and R. Murray-Smith, 2000. On the interpretation and identification dynamic
Takagi-Sugeno fuzzy models. IEEE Trans on Fuzzy Systems. Vol.8.n.3. pp.297-313.
[34] T. Floquet and J. Barbot, 2004. A sliding mode approach of unknown input observers for linear systems, in:
IEEE Conference on Decision and Contol, vol. 2, pp. 1724-1729.

129