Author’s Accepted Manuscript

Actuator and sensor faults estimation based on

proportional integral observer for TS fuzzy model

T. Youssef, M. Chadli, H.R. Karimi, R. Wang

www.elsevier.com/locate/jfranklin

PII: S0016-0032(16)30334-9
DOI: http://dx.doi.org/10.1016/j.jfranklin.2016.09.020
Reference: FI2731
To appear in: Journal of the Franklin Institute
Received date: 24 October 2015
Revised date: 25 August 2016
Accepted date: 15 September 2016
Cite this article as: T. Youssef, M. Chadli, H.R. Karimi and R. Wang, Actuator
and sensor faults estimation based on proportional integral observer for TS fuzzy
m o d e l , Journal of the Franklin Institute,
http://dx.doi.org/10.1016/j.jfranklin.2016.09.020
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
 Actuator and sensor faults estimation based on proportional
integral observer for TS fuzzy model
T. Youssef1, M. Chadli2*, H.R. Karimi3, R. Wang4
1
University of Boumerdès, Boumerdes 35000, Algeria (e-mail: tewfikyoussef@umbb.dz).
2
University of Picardie Jules Verne, MIS (EA 4029)
33 rue Saint- Leu, 80039 Amiens, France (e-mail: mchadli@u-picardie.fr)
3
Department of Mechanical Engineering, Politecnico di Milano, 20156 Milan, Italy.
4
School of Mechanical Engineering, Southeast University, Nanjing 211189, P.R. China.

Abstract

This paper presents a novel method to address a Proportional Integral observer design for the actuator and sensor faults
estimation based on Takagi-Sugeno fuzzy model with unmeasurable premise variables. The faults are assumed as time-
varying signals whose kth time derivatives are bounded. Using Lyapunov stability theory and L2 performance analysis,
sufficient design conditions are developed for simultaneous estimation of states and time-varying actuator and sensor faults.
The Proportional Integral observer gains are computed by solving the proposed conditions under Linear Matrix Inequalities
constraints. A simulation example is provided to illustrate the effectiveness of the proposed approach.

1. Introduction

The required performances at different operating conditions due to the constantly increasing of the
productivity of modern industrial systems are become more difficult to achieve, especially when the
unexpected faults occur. Therefore, whatever the origin of faults i.e., actuator, sensor or plant, the system
behavior may change in most cases. This undesirable behavior may degrade the performances and lead to the
system instability. In order to guarantee the safety and improve the performances of such systems, the
techniques of fault estimation, fault diagnostic and fault tolerant control developed by the scientific community
have received a growing interest, in particular the developed approaches based on state observers 1-2 for
nonlinear systems. The technique of Fault Detection and Isolation (FDI) based on the analytical redundancy
principle is one of the most commonly used approaches in the area of fault diagnosis.
The fault diagnosis methods can include data-driven or knowledge-based and model-based techniques.
Recently, a two-part survey paper is comprehensively reviewed on fault diagnosis approaches and their
applications. The fault diagnosis with data-driven techniques is reviewed in the first-part survey [3] and the
fault diagnosis with model-based techniques in the second-part survey [4]. Likewise, the data-based techniques
with recent developments focused on the complicated industrial applications are provided in an overview form
in [5]. This paper is destined to review and summarize the recent achievements of data-based techniques for the
modern industrial applications like process monitoring and control. Thus, it provides a referee for further study
on the related topics both from academic and practical points of view. The model-based diagnosis of nonlinear
systems with state observers is a direct application of state estimation, which the simultaneous estimation of
states and actuator and sensor faults will be discussed in this article.
Firstly, the FDI approaches have been developed for linear or linearized models 6, and then more recently,
they have been used to realize the diagnostic task 7. The linear model represents the system behavior only
near one operating point. Indeed, when fault occurs the operating point changes, therefore the linear model is
not representative. Thus, to take account of an operating wide range of nonlinear systems, we have proposed
our approach based on fuzzy model of Takagi-Sugeno (TS), which is a nonlinear representation. This approach
can approximate the nonlinear behavior of various dynamical systems by means of fuzzy sets and fuzzy IF-
THEN rules 8. Then, the overall behavior is achieved by an interpolation of linear models using nonlinear
membership functions satisfying the convex sum property 9, and each linear model defines a local behavior of
the nonlinear system. TS fuzzy models constitute a powerful tool of analysis and synthesis for nonlinear
systems since the control theory of linear systems can be applied. Membership functions can be constructed by
two premise variables types, measurable for inputs and/or outputs, and unmeasurable for state variables. When
the state variables are used, the obtained TS fuzzy models can approximate a large class of nonlinear systems
10. The stability analysis and stabilization of TS fuzzy models for design of observers and control laws are
addressed in 9 and 10. Based on the nonlinear state observers with measurable premise variables, the works
developed in 11-16 are applied for unknown inputs estimation and faults diagnosis. Indeed, in 11 the gain-
scheduled worst-case control on nonlinear stochastic systems subject to actuator saturation and unknown
information is studied, and in 12 the unknown inputs observers is designed in finite frequency domain for
fuzzy fault detection. The fault estimation observer is developed in [13] for discrete-time Takagi-Sugeno fuzzy
systems based on piecewise Lyapunov functions. The authors in [14] designed a robust fault detection observer
by using H‒/H formulation for discrete-time TS fuzzy system affected by sensor faults and unknown bounded
disturbances, whereas a chaotic system reconstruction is studied in [15]. A robust fault estimation approach and
its application in vehicle lateral dynamic systems are developed in 16. The authors in [17] deal with the H∞
filtering problem for a class of nonlinear networked systems subject simultaneously to randomly occurring
distributed state delays, nonlinearities, sensor saturation and missing measurements via unreliable
communication channels. For interval type-2 (IT2) TS fuzzy systems, a new observer-based sensor fault
detection design scheme is proposed in [18] for IT2 fuzzy discrete-time systems. The IT2 fuzzy parameter
uncertainties in the membership functions are handled by the lower and upper membership functions with
weighting functions. In the case of the PI observer design, the faults diagnosis is proposed in [19] for TS fuzzy
model with unmeasurable premise variables. The faults estimation and fault-tolerant control of actuators and/or
sensors is addressed in 20-26. The authors in [20] proposed a sensorless vector control that combines two
discrete-time sliding mode observers to estimate the rotor speed and position of permanent magnet synchronous
machines. In [21], the authors designed a fuzzy-model-based fault-tolerant for nonlinear stochastic systems
against simultaneous sensor and actuator faults. Real-time implementation of fault-tolerant control systems
with performance optimization is presented in 22. A fault estimation and tolerant control is studied in [23] for
fuzzy stochastic systems, and in [24] for stochastic systems with both input and output disturbances. Integrated
fault estimation and accommodation design is proposed in [25] for discrete-time Takagi-Sugeno fuzzy systems
with actuator faults. Based on a sliding mode observer, a sensorless and sensor-fault-resilient control of a
networked DC motor system with an optimal integral-square-error proportional-integral (PI) controller is
considered in [26], while network-induced delays and packet dropouts are taken into account.
In the framework of a class of systems in the network, for example, the control design problem is studied in
[27] for networked control systems (NCSs) and in [28] for linear multi-agent systems (MASs). The L2 state
feedback networked controller design problem is presented in [27] for event-triggered NCSs with quantizations
and time-varying delay. In order to reduce the communication load in the network, a new model of NCSs that
involves the network conditions, state and event-triggered communication strategy is proposed. The event-
triggered consensus problem is studied in [28] for MASs with general linear dynamics. Based on centralized
and distributed observer two novel event-triggered output feedback control schemes are proposed. In the case
of fuzzy control of the nonlinear time delay systems with unmodeled dynamics, an adaptive fuzzy control
scheme is developed in [29] for a class of single-input single-output strict-feedback nonlinear systems in order
to deal with phenomena like unmodeled dynamics, nonlinear uncertainties, unknown time delays and dynamic
disturbances. By combining the advances of the hyperbolic tangent function with adaptive fuzzy backstepping
technique and Lyapunov-Krasovskii functional approach, the proposed controller contained only one adaptive
parameter.
Most of the above mentioned works consider the case of measurable premise variables except for the
reference [19] which deals with unmeasurable premise variables. Now, in the case of unmeasurable premise
variables, the works developed in 30-35 lead the state estimation and sensor or actuator faults based on
observers. The robust fault tolerant tracking controller problem is addressed in [30] for uncertain TS fuzzy
continuous systems subject to unknown inputs and constant faults. The proposed control scheme is based on a
fuzzy PI observer to estimate constant faults by using the descriptor redundancy property. In addition, this
strategy is formulated in terms of bilinear matrix inequality. The authors in [30] lead the unmeasurable premise
variables as a disturbed system. In the context of the polynomial unknown inputs, the simultaneous estimation
of the state and the unknown inputs are presented in 31 for TS systems by considering the unmeasurable
premise variables as a Lipchitz constraint. The authors in 32 developed the robust observer design for
unknown inputs to continuous and discrete TS models subjected to disturbances. In the framework of a class of
nonlinear NCSs, for example, the observer-based controller design problem is investigated in [34] for nonlinear
NCSs subject to random measurement loss, parameter uncertainties and unmeasurable premise variables. The
considered nonlinear NCSs are described by IT2 TS fuzzy model which the uncertainties are handled via lower
and upper membership functions and relative weighting coefficients. A novel data compensation and regulation
strategy is adopted, which facilitates maintaining the system stability and improving the closed-loop system
performance. The unknown inputs PI observer for TS fuzzy models is developed in 35, the authors deal with
the estimation of identical unknown inputs which affect simultaneously the dynamics and output of the system.
In these works, the stability conditions for the state observers design are derived by using the Lyapunov
stability theory and the L2 performance analysis, and formulated in terms of linear matrix inequalities (LMIs)
constraints 36.
There are few works, in literature, on observer-based fault estimation for nonlinear systems described TS
fuzzy models with unmeasurable premise variables, which motivates our current study in this paper. Indeed, in
the framework of the unmeasurable premise variables, we are particularly interested to design a proportional
integral observer, because of their simultaneous estimations of states and faults. Based on the unknown inputs
PI observer for TS fuzzy models, the time-varying actuator and sensor faults estimation are presented in this
study. The faults signals act as the unknown inputs on the system whose kth time derivatives are assumed to be
norm-bounded. The main contribution of this paper can be summarized as follows: 1) a new PI observer design
for the estimation problem of both actuator and sensor faults with unmeasurable premise variables by
introducing an additional parameter which compensates the effect due to these variables and faults; 2) the
proposed PI observer will allow simultaneous reconstruction of the time-varying actuator and sensor faults
directly without considering a FDI scheme; and 3) based on the Lyapunov stability theory and L2 performance
analysis, sufficient conditions are derived to design the faults estimation PI observer gains under LMI
formulation.
The rest of the paper includes the following sections. In Section 2, we describe the TS fuzzy model with
unmeasurable premise variables. This TS model is subject to time-varying actuator and sensor faults, which can
affect the dynamics and output of the system, respectively. In Section 3, the description of the PI observer is
considered. In Section 4, the design of the proposed faults estimation PI observer is presented. Finally, in
Section 5, a simulation example is given to illustrate the effectiveness of our approach for estimation of both
states and actuator or sensor faults.
Notations: Throughout this paper, ‖ ‖ denotes the Euclidean norm. Let a symmetric matrix , ( )
denotes the positive-definite matrix (negative-definite matrix, respectively). and denote the transpose
and the inverse matrix, respectively. denote the nonlinear functions verifying the convex sum property
and , where denotes the number of fuzzy rules (or linear models). Let be a square matrix, such
that denotes the Hermitian of the matrix , defined by : . denotes the identity matrix
with dimension . Symbol denotes the transposed elements in the symmetric matrix . The symbol
represents the sign function.

Lemma 1. For any matrices and with appropriate dimensions, the following property holds [37]:
for any

2. Description of the TS Fuzzy Model

A nonlinear system can be approximated by a TS fuzzy model with unmeasurable premise variables subject
to faults which affect both actuator and sensor as follows:

̇ ∑ ( )
{

where represents the state vector, is the input vector, and are the
actuators and sensors faults vectors, respectively, is measurement noise vector, and
represents the output vector. are the state matrices, are the input matrices, is the
output matrix, and are the faults matrices, and is the disturbance matrix.
The represent the membership functions which depend on the unmeasurable premise, state of the
system. These functions satisfy the convex sum property:

∑
{

In order to simultaneously estimate the actuator and sensor faults, we transform the system (1) by using a
new state that is a filtered version of 38, defined by ̇ , where is a
stable matrix. The augmented system can be represented as follows:
 ̇ ∑ ( ̅ ̅ ̅ ) ̅
{
̅

̅
where [ ] and [ ] with ̅ , and

̅ [ ], ̅ [ ], ̅ [ ], ̅ [ ], ̅ [ ].

In the system given by (3), the sensor faults appear as actuator faults and the matrices ̅ are full column rank.

Hypothesis 1. The faults are assumed as time-varying signals whose kth time derivatives are bounded by
. The following notations are used:
̇
̇

̇
{

Remark 1. This assumption allows considering a wide range of actuator and sensor faults [16].

3. Description of the PI observer

The considered PI observer provides simultaneously the states and actuator and sensor faults in presence of
unmeasurable premise variables [35], described as follows:

̂̇ ∑ ( ̂) ( ̅ ̂ ̅ ̅ ̂ ( ̂ ))

̂ ̅̂

̂̇ ∑ ( ̂) ( ̂ ) ̂

̂̇ ∑ ( ̂) ( ̂ ) ̂
{

̅
where , and represent the proportional and integral gains, respectively,
which are to be designed. The variables , and are introduced in order to compensate the effect
due to unmeasurable premise variables.

Based on Hypothesis 1, the augmented forms of the TS model (3) and the PI observer (5) are given,
respectively, by:

̅̇ ∑ ̅ ̅
{
̅ ̅
and
̅̂̇ ∑ ( ̅̂ ) ( ̅̂ (̅ ̅̂ ))
{
̅̂ ̅̂

where
 ̂
̂
̅ , ̅̂ ̂ ,
[ ] [̂ ] [ ]

with ̅ , , ̅ and

̅ ̅ ̅̂ , ̅ ̅ ̅̂

̅ ̅ ̅

, , ,

[ ] [ ] [ ] [ ]

[ ] , [ ̅ ]

The dynamics of the augmented state estimation error is represented in the following form:

̇̅ ∑ ̅̂ ̅ ̅ ̅

where
̅ , [ ], ̅ [ ]

̅
[ ], [ ]

∑ ̅ , ∑ ̅ , ̅ ̅ ̅̂

Remark 2. According to the convex sum property (2) of membership functions we can state ̅ , note
that ∑ ̅ are not convex sums, and the variables matrices and are norm bounded, and the following
conditions are satisfied:

‖ ‖ , , ‖ ‖ ,

where and are the Euclidian norms of matrices and , respectively.

4. PI observer design

In this section, the PI observer design is developed to estimate the time-varying actuator and sensor faults.
The computing of gains is achieved by solving the convergence conditions of the augmented state
estimation error towards zero under LMI constraints given in Theorem 1.

Theorem 1. The system (9) is asymptotically stable and the L2 performance is guaranteed with an attenuation
level , if there exist a matrix , matrices and the positive scalars , and such that for all
the following minimization problem holds:

[ ]
where the parameters , , and are given in (8c), (8d) and (11), respectively.

Then, the gains of the PI observer (5) are computed by:

and
| ̅ |
{ ̅̂ ̅̂
̅ ̅ | ̅ |
̅ ̅ ̅ ̅

with variables ̅̂ , , ̅ and the parameter are given in (8a), (8b) and (11), respectively, with ( ) and
( ). and are positive scalars arbitrarily fixed.

Proof. By choosing , the considered Lyapunov quadratic function has the form ̅ ̅ . The
conditions (12) guarantee the asymptotic stability of the error dynamics (9). The proof is given in Appendix. □

The PI observer gains derived from the resolution of the constraint (12c) allow simultaneously estimating the
states and actuator and sensor faults. In the following section, a simulation example is given in order to validate
the proposed approach.

5. Simulation example

The hydraulic system (see Fig. 1) is composed by three tanks of same cross section which communicate by
means of cylindrical pipes of identical cross section , and a water reservoir which supplies the tanks 1 and 2
through two pumps and , respectively. The inputs and are the flow rates of pumps. The water is
recovered by the leak valves of tanks, thus the hydraulic system operates in a closed circuit [12].

Pump 1 u1 u2 Pump 2
S

x1 Sp
x3
13 32 x2 20

q13 q32
Tank 1 Tank 3 Tank 2 q20

Fig. 1. Technological scheme of three-tank hydraulic system.

The three-tank levels , and are governed by the constraint . The mass
conservation law leads to the following dynamic equations:

The flows are expressed according to Torricelli’s law by:

( ) √ | |

where is the flow from the ith tank to jth tank.

 The dynamic equations of hydraulic system can be expressed as:

( √| | )

( √ √| | )

( √| | √| |)
{

with
( ) √
√
( ) √

According to the constraint and the constants m/s2, m2, m2,

we obtain the values and .

5.1 TS fuzzy model representation

The analytical knowledge of nonlinear model (15) of three-tank hydraulic system allows us to obtain, with
the use of the sectors nonlinearity approach detailed in [9], the TS fuzzy model form given by (20) for not
unique choice of the following nonlinear continuous terms:

√| | √ √| |
, ,

The dynamical model (15) can be written in the following state-space representation:

̇ [ ] [ ]

where [ ] , [ ] and , , are three premise

variables which depend on state vector . Each premise variable is bounded in a compact state space
following the variation intervals [ ], [ ] and [ ]. Moreover, each
premise can be described by two partition functions ( ) and ( ) such that for all :

( ) ( ) ( ) ( ) ( )
where
( ) ( )

Then, from equation (18), the membership functions can be obtained by:

( )
( )
( ) ( )

( )
( )
( ) ( )

The combination of functions (19) leads to the construction of the TS fuzzy model (20) according to eight
local linear models. Then, the TS fuzzy model is obtained by an interpolation of these local
models with eight nonlinear functions described by the partition functions (19) as follows:

, , ,

, , ,
 The dynamic model (15) is exactly represented by the following TS fuzzy model:

̇ ∑ ( )
{

with

[ ], [ ],

[ ], [ ],

[ ], [ ],

[ ], [ ],

[ ] , [ ].

Hypothesis 2. We assume that the time-varying actuator and sensor faults fulfill second-derivative norm-
bounded by considering the unmeasurable state under a measurement noise on the outputs.

The considered inputs are given by Fig. 2 and measurement noise is a centered noise depicted in
Fig. 3.
-5 -4
x 10 x 10
7
u1 wy
6 u2 5
5

4
0
3

2
-5
1

0
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

Fig. 2. Three-tank system inputs (m3/s).

-5 -4
x 10 x 10
7
u1 wy
6 u2 5
5

4
0
3

2
-5
1

0
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

Fig. 3. Centered measurement noise (m).

 In the following subsection we present the simulation results of estimation of states, actuator and sensor
faults according to Hypothesis 2.

5.2 Actuator and sensor faults estimation

The TS fuzzy model of the three-tank system with unmeasurable decision variables subject to time-varying
additive actuator and sensor faults, and measurement noise can be represented in the following form:

̇ ∑ ( )
{

The TS model (21) can take the form (1) with the following notations:

where and are the actuator and sensor faults, and [ ] , [ ], [ ].

Solving LMIs constraints (12) of Theorem 1 leads to the PI observer gains given in Table 1. The simulation
results are carried out with level attenuation , constants , , ,
, and initial conditions ̂ [ ] , [ ] and [ ] . The
considered actuator and sensor faults appear at and as shown by Fig. 4 and Fig. 5,
respectively.

Table 1
PI observer gains.

1 [ ] [ ]

[ ]

2 [ ] [ ]

[ ]

3 [ ] [ ]

[ ]

4 [ ] [ ]

[ ]

5 [ ] [ ]

[ ]
 6 [ ] [ ]

[ ]

7 [ ] [ ]

[ ]

8 [ ] [ ]

[ ]

Actuator and sensor faults and their estimates are given by Fig. 4 and Fig. 5, respectively.
-5
x 10 -5
15 x 10
15
10
fa 5
10 0
f^a -5
4500 5000 5500
fa and f^a
5

-5
0 5000 10000 15000
t (s)

Fig. 4. Actuator fault and its estimate ̂ (m3/s).

-5
x 10
15 0.2

fa fs
0.15
10 f^a f^s

0.1
5
0.05

0
0

-5 -0.05
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)
Fig. 5. Sensor fault and its estimate ̂ (m).
-5
x 10
15
fa
States and outputs estimation
f^
are represented by Fig. 6 and Fig. 7, respectively.
10 a

-5
4400 4600 4800 5000 5200 5400 5600 5800
t (s)
 1 2
0.1
0.3 x ^1 x ^2

0.2
0.05
0.1

0 0
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

0.4 0.8
x1
0.7 x2 0.2
0.3 0.6
x3
0.6
x3 x ^1 0.15
0.2 0.50.4
x ^3 x ^2
0.4
xy^3 0.1
1
0.1 0.30.2 y2
y ^1
0.2 y ^2
0.05
0 0
0 5000 10000 15000 0.1 0 5000 10000 15000
t (s) t (s)
0 0
Fig. 6. States 0
, , 5000 10000
and their estimates ̂ ̂
, 15000 ,̂ (m).0 5000 10000 15000
t (s) t (s)

0.8
y1 y2 y ^1 y ^2

0.6

0.4

0.2

0
0 5000 10000 15000
t (s)

Fig. 7. Outputs , and their estimates ̂ ,̂ (m).

The estimation errors of actuator and sensor faults, states and outputs are given by Fig. 8, Fig. 9, Fig. 10 and
Fig. 11, respectively. The simulation results show a good simultaneous estimation of states, actuator and sensor
faults.
-5
x 10
5
ef
a

-5
0 5000 10000 15000
t (s)

Fig. 8. Error between actuator fault and its estimate (m3/s).

-5
x 10
5 0.01
ef ef
a s

0.005

0 0

-0.005

-5 -0.01
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

Fig. 9. Error between sensor fault and its estimate (m).

 0 0

-0.005 -0.005

-0.01 -0.01
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

0.01 0.01
ex ex
3 1
0.01 0.01
ex
0.005 0.005 e2y ey
1 2
ex
3
0.005 0.005
0 0

0 0
-0.005 -0.005

-0.01 -0.005
-0.01 -0.005
0 5000 10000 15000 0 5000 10000 15000
t (s) t (s)

-0.01 -0.01
0 Errors between
Fig. 10. 5000 states and 10000 15000
their estimates (m). 0 5000 10000 15000
t (s) t (s)

0.01

ey
1
0.005
ey
2

-0.005

-0.01
0 5000 10000 15000
t (s)

Fig. 11. Errors between outputs and their estimates (m).

6. Conclusion

In this work, a proportional integral observer was designed to simultaneous estimation of states, and time-
varying sensor and actuator faults for TS fuzzy model. The considered TS fuzzy model was subject to
unmeasurable premise variables and measurement noise. In order to take into account a wide range of faults,
the actuator and sensor faults are assumed to be time-varying signals with bounded kth time derivatives. Based
on Lyapunov stability theory and L2 performance analysis, sufficient conditions of the proposed PI observer are
derived in LMIs formulation. A simulation example was provided to illustrate the effectiveness and feasibility
of the proposed approach. Indeed, the obtained results show the good simultaneous estimation of states, and
actuator and sensor faults.

Appendix. Proof of Theorem 1

Taking into account (10a) and (10b), the time-derivative of quadratic Lyapunov function ̅ ̅
where leads:

̇ ∑ ̅̂ ( ̅ ( ̅ ̅) ̅ ̅ ̅ ) ̅ ̅ ̅ ̅

On grounds of simplicity, the time variable is omitted.

Then, based on (8b) and according to Lemma 1 with the notation (11), we obtain:

̇ ∑ ( ̅̂ ) ( ̅ ( ̅ ̅) ̅ ̅ ̅ ) ( ̅ ̅ ̅̂ ̅̂ ̅̂ ̅̂ ̅ ̅)

̅ ̅ ̅ ̅ ̅

Using again Lemma 1, the inequality (A.2) becomes:

 ̇ ∑ ( ̅̂ ) ( ̅ ( ̅ ̅ ) ̅ ̅ ̅ )

̅̂ ̅̂ ̅

with
, , ( ), ( ).

By substituting the expression of z (12d) in (A.3), we get:

̅̂ ̅̂
̅ ̅ ̅ ̅ ̅
̅ ̅ ̅ ̅
̅̂ ̅̂

with ̅ ̅ and ̅ ̅ .

Thus, the inequality (A.3) becomes:

̇ ∑ ( ̅̂ ) ( ̅ ( ̅ ̅ ) ̅ ̅ ̅ )

The system (9) is stable and the L2 performance is guaranteed with an attenuation level if the following
condition holds [7]:

̇ ̅ ̅

The condition (A.6) is satisfied if the following matrix inequalities hold:

[ ]

with ̅ ̅ .

From the condition (A.7), the Schur complement on with the variables change ̅ and
lead to linear matrix inequalities (12a).

References
[1] Ron J. Patton, Paul M. Frank, R. N. Clark, Issues of Fault Diagnosis for Dynamic Systems, London: Springer-Verlag, 2000.
[2] M. Blanke, M. Kinnaert, J. Lunze, M. Staroswiecki, Diagnosis and fault-tolerant control, 2nd ed. Germany: Springer, 2006.
[3] Gao Z., C. Cecati, S. X. Ding, A Survey of Fault Diagnosis and Fault-Tolerant Techniques–Part I: Fault Diagnosis with Model-Based
and Signal-Based Approaches, IEEE Transactions on Industrial Electronics 62 (6) (2015) 3757–3767.
[4] Gao Z., C. Cecati, S. X. Ding, A Survey of Fault Diagnosis and Fault-Tolerant Techniques–Part II: Fault Diagnosis with Knowledge-
Based and Hybrid/Active Approaches, IEEE Transactions on Industrial Electronics 62 (6) (2045) 3768–3774.
[5] S. Yin, X. Li, H. Gao, O. Kaynak, Data-Based Techniques Focused on Modern Industry: An Overview, IEEE Transactions on
Industrial Electronics 62 (1) (2015) 657–667.
[6] R. Patton, P. Frank, R. Clark, Fault Diagnosis in Dynamic Systems: Theory and Application, Prentice Hall international, 1989.
[7] S. X. Ding, Model-Based Fault Diagnosis Techniques Design Schemes, Algorithms and Tools, Springer-Verlag, 2008.
[8] T. Takagi, M. Sugeno, Fuzzy Identification of Systems and Its Applications to Modeling and Control, IEEE Transactions on Systems,
Man and Cybernetics SMC–15 (1) (1985) 116–132.
[9] K. Tanaka, H. O. Wang, Fuzzy Control Systems Design and Analysis : A LMI Approach, John Wiley and Sons, 2001.
[10] J. Yoneyama, H filtering for fuzzy systems with immeasurable premise variables: an uncertain system approach, Fuzzy Sets and
Systems 160 (12)(2009) 1738–1748.
[11] P. Shi, Y. Yin, F. Liu, Gain-scheduled worst case control on nonlinear stochastic systems subject to actuator saturation and unknown
information, J. of Optimization Theory and Applications156 (3) (2013) 844-858.
[12] A Chibani, M Chadli, P Shi, NB Braiek. Fuzzy Fault Detection Filter Design for T-S Fuzzy Systems in Finite Frequency Domain
IEEE Trans. on Fuzzy Systems (2016), doi: 10.1109/TFUZZ.2016.259392
[13] K. Zhang, B. Jiang, P. Shi, Fault estimation observer design for discrete-time Takagi-Sugeno fuzzy systems based on piecewise
Lyapunov functions, IEEE Trans on Fuzzy Systems20 (1) (2012)192-200.
[14] M. Chadli, A. Abdo, S. X. Ding, H/H∞ fault detection filter design for discrete-time Takagi–Sugeno fuzzy system, Automatica 49
(7) (2013) 1996–2005.
15 M. Chadli, I. Zelinka, T. Youssef. Unknown inputs observer design for fuzzy systems with application to chaotic system
reconstruction. Computers & Mathematics with Applications, 66 (2) (2013), 147-154.
[16] Z. Gao, S. X. Ding, Y. Ma, Robust fault estimation approach and its application in vehicle lateral dynamic systems, Optimal Control
Applications and Methods28 (2007) 143–156.
[17] H. Yan, F. Qian, F. Yang, H. Shi, H∞ filtering for nonlinear networked systems with randomly occurring distributed delays, missing
measurements and sensor saturation, Information Sciences (2015) 1–11, doi:10.1016/j.ins.2015.09.027.
[18] H. Li, Y. Gao, P. Shi, H. K. Lam, Observer-based Fault Detection for Nonlinear Systems with Sensor Fault and Limited
Communication Capacity, IEEE Trans. on Automatic Control (99) (2015) 1–6, doi:10.1109/TAC.2015.2503566.
[19] T. Youssef, H.R. Karimi, M. Chadli, Faults Diagnosis based on Proportional Integral Observer for TS Fuzzy Model with
Unmeasurable Premise Variable, IEEE International Conference on Fuzzy Systems (2014)2119-2124.
20 T. Bernardes, V. Foletto Montagner, H.A. Grundling, H. Pinheiro, Discrete-Time Sliding Mode Observer for Sensorless Vector
Control of Permanent Magnet Synchronous Machine, IEEE Transactions on Industrial Electronics61 (4)(2014) 1679–1691.
[21] M. Liu, X. Cao, P. Shi, Fuzzy-model-based fault-tolerant design for nonlinear stochastic systems against simultaneous sensor and
actuator faults, IEEE Trans on Fuzzy Systems 21 (5) (2013) 789-799.
22 S. Yin, H. Luo, S.X. Ding, Real-Time Implementation of Fault-Tolerant Control Systems With Performance Optimization, IEEE
Trans. on Industrial Electronics61 (5) (2014) 2402–2411.
[23] M. Liu, X. Cao, P. Shi, Fault estimation and tolerant control for fuzzy stochastic systems, IEEE Trans on Fuzzy Systems21 (2) (2013)
221-229.
[24] M. Liu, P. Shi, Sensor fault estimation and tolerant control for stochastic systems with simultaneous input and output disturbances,
Automatica 49 (5) (2013) 1242-1250.
[25] B. Jiang, K. Zhang, P. Shi, Integrated fault estimation and accommodation design for discrete-time Takagi-Sugeno fuzzy systems
with actuator faults, IEEE Trans on Fuzzy Systems19 (2) (2011) 291-304.
[26] A. Ahmadi, F.R. Salmasi, M. Noori-Manzar, T.A. Najafabadi, Speed Sensorless and Sensor-Fault Tolerant Optimal PI Regulator for
Networked DC Motor System With Unknown Time-Delay and Packet Dropout, IEEE Transactions on Industrial Electronics61 (2)
(2014) 708–717.
[27] H. Yan, S. Yan, H. Zhang, H. Shi, L2 control design of event-triggered networked control systems with quantizations, Journal of the
Franklin Institute 352 (1) (2015) 332–345.
[28] H. Zhang, G. Feng, H. Yan, Q. Chen, Observer-Based Output Feedback Event-Triggered Control for Consensus of Multi-Agent
Systems, IEEE Transactions on Industrial Electronics 61 (9) (2014) 4885–4894.
[29] S. Yin, P. Shi, H. Yang, Adaptive Fuzzy Control of Strict-Feedback Nonlinear Time-Delay Systems with Unmodeled Dynamics,
IEEE Transactions on Cybernetics 46 (8) (2016) 1926–1938.
[30] S. Aouaouda, M. Chadli, M.T. Khadir, T. Bouarar, Robust fault tolerant tracking controller design for unknown inputs T-S models
with unmeasurable premise variables. Journal of Process Control22 (5) (2012) 861–872.
[31] Z. Lendek, J. Lauber, T. M. Guerra, R. Babuska, B. De Schutter, Adaptive observers for TS fuzzy systems with unknown polynomial
inputs,Fuzzy Sets and Systems 161 (15) (2010) 2043–2065.
[32] M. Chadli, H. R. Karimi, Robust Observer Design for Unknown Inputs Takagi-Sugeno Models, IEEE Trans. on Fuzzy Systems 21
(1) (2013) 158–164, 2013.
33 H. Zhang, G. Zhang and J. Wang. Observer Design for LPV Systems with Uncertain Measurements on Scheduling Variables:
Application to an Electric Ground Vehicle. IEEE/ASME Trans. on Mechatronics 21(3) (2016), 1659-1670.
[34] H. Li, C. Wu, S. Yin, H. K. Lam, Observer-Based Fuzzy Control for Nonlinear Networked Systems under Unmeasurable Premise
Variables, IEEE Trans on Fuzzy Systems (99) (2015) 1–12, 10.1109/TFUZZ.2015.2505331.
[35] T.Youssef, M. Chadli, H. R. Karimi, M. Zelmat, Design of unknown inputs proportional integral observers for TS fuzzy models,
Neurocomputing 123 (2014) 156–165.
[36] S. Boyd, L. El Ghaoui, E. Feron, V. Balakrishnan, Linear Matrix Inequalities in System and Control Theory, SIAM, Studies In
Applied Mathematics, 1994.
[37] K. Zhou, P.P. Khargoneker, Robust stabilization of linear systems with norm-bounded time-varying uncertainty, Systems and Control
Letters 10 (1) (1988) 17–20.
[38] C. P. Tan, C. Edwards, Sliding mode observers for robust detection and reconstruction of actuator and sensor faults, Int. J. of Robust
and Nonlinear Control 13 (2003) 443–463.