2020 17th International Multi-Conference on Systems, Signals & Devices (SSD'20)

Fault Detection and Isolation for Continuous-time

Switched Linear Systems: A Set Membership
Approach
Haifa Ethabet Tarek Raı̈ssi
Research Laboratory MACS LR16ES22 Conservatoire National des Arts et Metiers (CNAM)
National Engineering School of Gabes (ENIG) Cedric - lab 292
2020 17th International Multi-Conference on Systems, Signals & Devices (SSD) | 978-1-7281-1080-6/20/$31.00 ©2020 IEEE | DOI: 10.1109/SSD49366.2020.9364097

University of Gabes 6029 Gabes, Tunisia Rue St-Martin, 75141 Paris Cedex 03, France
Email: [email protected] Email: [email protected]

Messaoud Amairi Mohamed Aoun

Research Laboratory MACS LR16ES22 Research Laboratory MACS LR16ES22
National Engineering School of Gabes (ENIG) National Engineering School of Gabes (ENIG)
University of Gabes 6029 Gabes, Tunisia University of Gabes 6029 Gabes, Tunisia
Email: [email protected] Email: [email protected]

Abstract—In this paper, the problem of Fault Detection and for uncertain and disturbed systems. To compute robust esti-
Isolation (FDI) is investigated for continuous-time switched linear mates, interval observers have been introduced in the literature
systems via a set-membership approach. Under the fulfillment of [16], [4], [20]. They assume that noises and disturbances are
the relative degree property by all the subsystems, the proposed
solution is based on the use of a bank of interval unknown unknown but bounded with a priori known bounds. This tech-
input observers. Under the assumption that disturbances and nique is developed to cope these uncertainties and disturbance
measurement noise are unknown but bounded with a priori and determine certain upper and lower estimates of the system
known bounds, cooperativity and stability conditions are given in at each time instant. Indeed, it provides guaranteed upper and
terms of Linear Matrix Inequalities (LMIs) with the fulfillment lower bounds of the state of the corresponding system. A
of an Average Dwell Time (ADT) constraints. Then, upper and
lower residuals are computed. A numerical example illustrating growing interest is carried out to fault detection and isolation
the validity of the method in fault detection and isolation is given. in a set-membership context [17], [18]. For the best of the
authors knowledge, the set-membership fault detection and
Index Terms—Fault detection and isolation, set-membership, isolation for switched linear systems is not fully considered
switched systems, continuous-time. yet.
I. INTRODUCTION The main contribution of this paper is to develop a
fault detection and isolation methodology for continuous-time
Being one of the most important classes of Hybrid Dy-
switched linear systems in an unknown but bounded error
namical Systems (HDS), switched systems have been widely
framework. The proposed technique consists in the use of a
studied in the last decade [8], [9]. They involve both con-
bank of interval unknown input observers. The disturbances
tinuous and discrete time dynamics. They are a family of
and measurement noise are unknown but bounded with a
some different subsystems orchestrated by a switching rule.
priori known bounds. The design of interval unknown input
Due to their theoretical and practical challenges, several works
observers is based on a decoupling strategy to make the state
addressed stability and control of switched systems such as [7],
partially affected by the unknown input. Then, upper and lower
[10], [11].
bounds of the state are estimated. Stability and cooperativity
Among the large variety of problems encountering switched
conditions are given in terms of Linear Matrix Inequalities
systems, Fault Detection and Isolation (FDI) has received
(LMIs) besides a fulfillment of an Average Dwell Time (ADT)
a growing attention. For instance, ensuring a high level of
constraint. Finally, upper and lower residuals are derived from
security and reliability is a crucial issue. FDI have been in-
different unknown input observers.
vestigated using several methods such as geometric approches
[12], observer based techniques [13], [14] and parity checks This paper is arranged as follows. In section 2, some
[15]. For instance, in [14] the fault detection and isolation is useful preliminaries and notations are presented. Section 3 is
ensured via a H∞ observer for linear switched system. devoted to the fault detection and isolation proposed technique.
Roughly speaking, measurements are usually subject to Simulation results are shown in section 4. Section 5 concludes
noises. Thus, the design of classical observers is quite difficult the paper.

978-1-7281-1080-6/20/$31.00 ©2020 IEEE 279

Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.
 II. PRELIMINARIES Throughout this paper, z1 , z2 and d are assumed to be
The sets of real and natural numbers are denoted by R and bounded.
N respectively. Denote the sequence of integers {1, ..., N} by The assumptions below are required in the sequel.
1, N. |·| denotes the elementwise absolute value of a vector Assumption 1: The measurement noise and the state distur-
x ∈ Rn resp. a matrix A ∈ Rn×m . E p denotes a (p × 1) vector bance are assumed to be unknown but bounded with a priori
whose elements are equal to 1. I is the identity matrix of known bounds such that
proper dimension. For a matrix P = PT , the relation P ≺ 0 −w ≤ w (t) ≤ w, ∀t ≥ 0 (5)
(P ≻ 0) means that the matrix P ∈ Rn×n is negative (positive)
definite. Denote by x and x the lower and upper bounds of −v ≤ v (t) ≤ v, ∀t ≥ 0 (6)
a variable x such that x ≤ x ≤ x. The relation ≤ should be p
where w ∈ Rn+and v ∈ R+ .
interpreted elementwise for vectors as well as for matrices, Assumption 1 means that the state disturbance w and the
i.e. A = (ai, j ) ∈ R p×m and B = (bi, j ) ∈ R p×m such that A ≥ B measurement noise v belong to the intervals [−w, w] and
if and only if, ai, j ≥ bi, j for all i ∈ 1, p, j ∈ 1, m. For a matrix [−v, v] respectively. The values of w and v are not available
A ∈ Rm×n , define A+ = max {0, A} and A− = A+ − A. for measurement but their bounds w and v are a priori known.
Assumption 2:
Lemma 1: [1] Let x ∈ Rn be a vector satisfying x ≤ x ≤ x̄
and A ∈ Rm×n be a constant matrix, then rank (Cq Dq ) = rank (Dq ) = δ , ∀q ∈ I , δ ≤ p (7)
Assumption 2 states the classical condition for the existence
A+ x − A− x̄ ≤ Ax ≤ A+ x̄ − A−x (1)
of unknown input observer.
If A is satisfying the relation A ≤ A ≤ A, then
A. Design of interval unknown input observer
+ + + − − + − −
A x − A x − A x + A x ≤ Ax ≤ Unknown input observers are known to be efficient residual
+ + + − − + − −
A x −A x −A x +A x . generators for fault detection and isolation. Through state and
output transformation matrices, the state is partially affected
Lemma 2: [2] Let µ > 0 be a positive scalar and P ∈ Rn be by the unknown input. Then, upper and lower bounds of the
a symmetric positive definite matrix, then state are estimated.
1 T Let f (t) = 0. Then, the system (4) is rewritten as:
2xT y ≤ x Px + µ yT P−1 y x, y ∈ Rn . (2)
µ

ẋ (t) = Aq x (t) + Bq u (t) + Dq d (t) + w (t)
(8)
Lemma 3: [3] Consider the system described by: ym (t) = Cq x (t) + v (t)

ẋ (t) = Ax (t) + u (t) , x (0) = x0 (3) The proposed technique is based on a decoupling method
that makes the state partially affected by the unknown input.
The system (3) is said to be cooperative if A is a Metzler matrix First, non singular state and output transformation matrices
and u (t) ≥ 0. For any initial condition x0 ≥ 0 the solution of respectively Tq and Uq are calculated. Then, an Interval
(3) satisfies x (t) ≥ 0, ∀t ≥ 0. Unknown Input Observer (IUIO) is designed for the state of
the continuous-time Switched Linear System (SLS) without
III. FAULT DETECTION AND ISOLATION FOR unknown input [19].
CONTINUOUS-TIME SWITCHED LINEAR SYSTEMS Under Assumption 2, there exist non singular state and
In this section, a fault detection and isolation technique is output transformation matrices Tq and Uq respectively defined
developed for continuous-time switched linear systems. The as follows:
key point of this approach is to design a bank of interval un-  

T1q T2q

known input observers. Then, upper and lower residual signals, Tq = Γq Dq = (9)
T3q T4q
derived from each unknown input observer, are computed.  
Consider the following continuous-time switched linear Uq = Cq Dq Qq (10)
system described by: where Tq ∈ Rn×n , Γq ∈ Rn×(n−δ ) and Qq ∈ R p×(p−δ ) .
Denote by Uq−1 and Tq−1 the inverse of matrices Uq and Tq

ẋ (t) = Aq x (t) + Bqu (t) + Dq d (t) + w (t) + Hq f (t)
(4) respectively such that
ym (t) = Cq x (t) + v (t)
 
where x ∈ Rn , u ∈ Rm , ym ∈ R p , d ∈ Rδ , w ∈ Rn , v ∈ R p , and −1 U1q
Uq = (11)
f ∈ Rn are the state vector, the input, the output, the unknown U2q
 ∗ 
input, the disturbance, the measurement noise and the additive Dq
fault respectively. q = σ (t) = σk , k = 1, 2, ..., is a piecewise Tq−1 = (12)
D+q
constant function that takes its values in an index set I = 1, N.
It is assumed to be known. N is the number of the possible with Uq ∈ Rδ ×p , U2q ∈ R(p−δ )×p , D∗q are matrices such that

T
subsystems represented by specific constant matrices Aq , Bq , D∗q Dq = 0 and D∗q 6= 0. D+q are the left pseudo-inverse of
Cq , Dq , and Hq of appropriate dimensions. matrices Dq .

280
Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.
 Using the decoupling method, one can obtain: Then (14) is Input to State Stable and z1 , z1 are bounded

ż1 = Ǎ1q z1 + B̃1q u + γqym + w̃1q − γq v provided that the ADT τa is chosen sufficiently large
(13)
ỹm = Čq z1 + U2qv ln c1
τa > (20)
where c3
2
−1
c1 = sup M 1/2 T1q+1 T1q M −1/2


Ǎ1q = Ã1q − Ã2qU1qCq Γq , γq = Ã2qU1q , (21)
q∈I
Čq = U2qCq Γq , ỹm = U2q ym
λmin (Jq )
Remark 1: For brevity of presentation, we will use the c3 = inf . (22)
q∈I λmax (M)
unknown input interval observer proposed in [19]. Thus, for
more details about the decoupling method one can referee to Proof 1: The proof of the above theorem is detailed in [19].
[19]. The next step consists in computing upper and lower bounds
The following assumption guarantees the existence of an for the whole state in the original basis “x”. They are given
observer structure for the system (13). Mainly, this assumption as follows:
is a classical condition in the field
of observer design theory.  
+ 
+
 x1 = T1q + −T2qU1qCq Γq z1
Assumption 3: The pairs Ǎ1q , Čq are detectable for all q ∈ 

 
− 
+
−
 − T1q + −T2qU1qCq Γq z1 + T2qU1q ym

I. 


Once the unknown input is decoupling, an interval unknown  − T2qU1q − ym + −T2qU1q + v + −T2qU1q − v






input observer is designed for the system (13). Inspired by the (23)
proposed observer in [19], consider two conventional point  
+

+ 
x1 = T1q + −T2qU1qCq Γq z1


observers expressed as follows: 

 
 −

− 
+
− T1q + −T2qU1qCq Γq z1 + T2qU1q ym
 + 
ẑ˙1 = hǍ1q − LqČq ẑ+ −1


1 + B̃ 1q u + Pq Pq w̃1q
 

 

−
+
−
− T2qU1q ym − −T2qU1q v − −T2qU1q v
  i
−1 P+ γ + y − γ − y − γ +y − γ −y




 +P q q q m q m
− Pq q m q m
 +Pq−1 Pq γq v + Lq ỹm + Pq−1 Pq Lq U2q v
  
+ 
+

x = T + −T U C Γ z1

 2

 3q 4q 1q q q
(14) 
−

−

+
 − T3q + −T4qU1qCq Γq z1 + T4qU1q ym

 ẑ˙−

− −1 − P


= hǍ1q − + +


 1  L qČq ẑ1 B̃1q u Pq q w̃ 1qi



−
+
−
 − T4qU1q ym + −T4qU1q v + −T4qU1q v
 
−1 P+ γ + y − γ − y − γ +y − γ −y

+P − P
 
q q q q m q q m q

 m m

 −P−1 P γ v + L ỹ − P−1 P L U v
 (24)
q q q m q q 2q
q q

+ 
+

 x2 = T3q + −T4qU1qCq Γq z1


where

 
− 
+
ẑ+ −1 P+ z − P− z −


10 = Pq − T3q + −T4qU1qCq Γq z1 + T4qU1q ym

q 10 q 10


(15)

ẑ− −1
Pq+ z10 − Pq− z10

10 = Pq


−
+
−
− T4qU1q ym − −T4qU1q v − −T4qU1q v
 
z10    
with z0 = , z10 ∈ R(n−δ ) , z20 ∈ Rδ , x1 x1
z20 where x (t) = and x (t) = .
x2 x2
(
+
−
z0 = Tq−1 x0 − Tq−1 x0 B. Interval unknown input observers bank
+
− (16)
z0 = Tq−1 x0 − Tq−1 x0


In this section, a fault isolation methodology based on
The following theorem gives the conditions for achieving unknown input observers bank in a set-membership context is
partially the desired design goal. used. There exist different types of fault isolation scheme in
Theorem 1: [19] Let Assumptions 1-3 be satisfied and x0 ≤ the fault diagnosis literature. The two commonly used schemes
x0 ≤ x0 . If there exist non singular transformation matrices are the Dedicated Scheme (DS) proposed by Clark [5] and
Pq ∈ R(n−δ )×(n−δ ) such that Pq Ǎ1q − LqČq Pq−1 are Metzler,


the Generalized Scheme (GS) proposed by Frank [6]. In both
then an interval estimation for the state of the system (13) is scheme, the number of observers constituting the observers
described by: bank is equal to the number of faults N f under consideration.
(
+
− For the DS, the ith residual is sensitive and robust only to the
z1 (t) = Pq−1 Pq ẑ+ 1 (t) − Pq−1 Pq ẑ− 1 (t) ith fault, i = 1, N f , however decoupled from all other faults.
−1 + − −1 −
(17)
Pq ẑ+



z1 (t) = Pq Pq ẑ1 (t) − Pq 1 (t) Whereas in the GS, the ith residual is designed to be sensitive
such that to all faults but the ith one.
z1 (t) ≤ z1 (t) ≤ z1 (t) , ∀t ≥ 0 (18) The residual evaluation in a set-membership context is
ensured via the following belonging test: if the zero signal is
Also, if there exist symmetric positive definite matrices M and enclosed by upper and lower bounds of the residual signal, it is
Jq and positive scalars µq such that a fault-free case. Otherwise, a fault is occurred. For instance,
9 in the GS, for each subsystem, if the zero signal is enclosed by
ǍT1q M + M Ǎ1q − ČqT Wq T − WqČq + M  −Jq (19) upper and lower bounds of the ith residual and not enclosed
µq

281
Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.
 by upper and lower bounds of all the remaining residuals, then 2) Calculate the matrices Tqi and Uqi such that
a decision on the occurrence of the ith fault is made.
U1q i
 
Upper and lower residuals signals are defined as the gap −1 i
Uq = (28)
between measured outputs and estimated outputs resulting U2q i
" #
from the proposed observer. D∗q i
−1 i
From Lemma 1, the upper and lower bounds of the esti- Tq = i (29)
D+q
mated output are then computed as follows:
y = Cq+ x − Cq− x + v 3) Compute observer gains Liq , M i and Jqi by resolving the

(25)
y = Cq+ x − Cq− x − v LMIs defined by (19).
Upper and lower residuals are then denoted by: IV. SIMULATION RESULTS

r = y − ym Consider the continuous-time switched linear system de-
(26) scribed by:
r = y − ym

Fig.1 represents the generalized scheme of fault isolation. ẋ (t) = Aq x (t) + Bqu (t) + Dq d (t) + w (t) + Hq f (t)
(30)
ym (t) = Cq x (t) + v (t)
u1 Continuous-time y(t)
ui where
um switched linear system  
−1.9 0 0 0
 0 −2.2 0 0 
A1 =  ,
 0 0 −1 0 
0 0 0 −2
r1  
Interval unknown −1.8 0 0 0
input observer 1 r1  0 −1.5 0 0 
A2 =  ,
 0 0 −1 0 
ri 0 0 0 −1.3
Interval unknown
ri
input observer i
 
1 0 0 0
 0 1 0 0 
rN f B1 = B2 = 
 0
,
0 1 0 
Interval unknown
rN f 0 0 0 0
input observer N f
 
Fig. 1. Generalized scheme of fault isolation.
  0.2
1 0 0 0  0 
C1 =  0 1 0 0  , D1 = 
 0.3  ,

Table .1 represents the general case of the signature matrix
0 0 1 0
of the GS of fault isolation with zero diagonal and 1 off 0.5
diagonal. It is worth noting that 0 and 1 mean respectively  
  0.3
that the residual is insensitive and sensitive to a fault fi . −1 0 0 0.1  0.2 
C2 =  0 1 0 0  , D2 =  0.4  ,

TABLE I 0 0 −1 0
S IGNATURE MATRIX FOR THE GENERALIZED SCHEME 0.2
     
❍
❍ fi 1 0 0 0 0 0
f1 f2 ··· fN f −1 fN f
ri ❍❍  0 0   1 0   0.2 0 
r1 0 1 ··· 1 1 H1 = 
 0
 , H2 = 
 0 0  , H3 =  −0.8 1.1
  .
r2 1 0 ··· 1 1
0  
. .. 0 0 0 0 0 0
.. .
rN f −1 1 1 ··· 0 1 w (t) and v (t) are uniformly distributed bounded
T signals
× 10−3

rN f 1 1 ··· 1 0 such that −w ≤ w (t) ≤ w with w = 5 5 5 5
 T
and −v ≤ v (t) ≤ v with v = 0.03 0.03 .
Remark 2: In this paper, the sensitivity to a fault fi is Note that the switched linear system is affected by the
ensured through the belonging test already mentioned above. following additive faults:
The design of the interval unknown input observers bank is 
elaborated as follows: 1 10s ≤ t ≤ 13s
f1 =
1) Verify the existence condition for each observer for all 0 else
q∈I

1 20s ≤ t ≤ 23s
rank Cq Dq i = rank Diq



(27) f2 =
0 else

282
Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.
   
1 45s ≤ t ≤ 50s −2.3649 1.6892 0
f3 =
0 else U23 =  0.9459 −0.6757 −1 
−0.5812 −0.8137 0
The matrices Liq , Mqi and Jqi are computed verifying that Bi
is a two-column matrix obtained after deleting the ith column 3) Compute observer gains Liq , Mqi and Jqi by resolving the
of B. LMIs defined by (19). A feasible solution of this LMI for
Now, the observer bank design is constructed as follows: each Interval Unknown Input Observer (IUIO) is given
1) The existence condition is satisfied for each observer for by:
all q ∈ I • Parameter of IUIO 1:
2) The matrices Tqi and Uqi are calculated:    
−0.4883 0.2070
  L11 = , L12 = ,
0 00.2 1 −0.5157 −0.5783
 0.5145 0.8575 0 0 
T11 = 
 
 0.7353 −0.4412 0.3 0 

1 87.7547 20.9413
M = ,
−0.4412 0.2647 0.5 0 20.9413 97.6352
   
88.5831 −2.1377 79.3275 −11.3853
J11 = , J21 =
 
0 0 3.3333
−2.1377 86.3028 −11.3853 62.2495
U11 =  1 0 −0.6667 
0 1 0 • Parameter of IUIO 2:
 
−0.3482 −0.6963 −0.3482 0.3
   
0.1051 0.6839
 0.9204 −0.1593 −0.0796 0.2  L21 = , L22 = ,
T21 =   −0.1264 −0.5686
 −0.1593 0.6815 −0.1593 0.4   
−0.0796 −0.1593 0.9204 0.2 2 230.0848 17.6699
M = ,
  17.6699 178.0850
0 1 −2    
209.7778 −6.1221 187.3843 3.7086
U21 =  −1 −0.28 0.56  2
J1 = 2
, J2 =
−6.1221 216.4899 3.7086 205.4194
00.8944 0.4472
  • Parameter of IUIO 3:
−0.4867 −0.8111 0.2 0
0 0 01
   
T12 = 
  −0.4379 −0.1678
 0.8212 −0.2980 0.3 0 
 L31 = , L32 = ,
0.0210 0.2117
−0.2980 0.5033 0.5 0  
277.2083 −3.1465
  M3 = ,
1.5385 0 2.3077 −3.1465 228.5401
U12 =  0 1 0     
3 281.5239 3.1610 1 281.25 −3.1096
−0.8321 0 0.5547 J1 = , J2 =
3.1610 289.0861 −3.1096 291.4486
 
−0.3482 −0.6963 −0.3482 0.3 To ensure the cooperativity property, different transformation
 0.9204 −0.1593 −0.0796 0.2 
T22 =   matrices P1i and P2i are used. Then, interval observers as in
 −0.1593 0.6815 −0.1593 0.4 
(14) can be designed. Transformation matrices P1i and P2i are
−0.0796 −0.1593 0.9204 0.2 i
computed such that Pqi Ǎq − LqČq Pq−1 are Metzler for all



−1.1745 0 −1.6779
 q∈I.
U22 =  0.2349 1 0.3356  The evolution of residual signals derived from IUIO 1, 2
−0.8192 0 0.5735 and 3 plotted in Fig.2, Fig.3 and Fig.4 respectively. Solid lines
  present the zero signal and blue and red dashed lines are the
−0.3168 −0.8728 0.2 0 upper and lower bounds of the residual signals respectively.
 −0.9400 0.3412 0 0  The switching between the two subsystems is governed by
T13 =  
 0 0 0.3000 1  the switching signal verifying the ADT constraint as plotted
0.1267 0.3491 0.5000 0 in Fig.5.
  For the first fault f1 (Fig.2), the residual signals are sensible
5 0 0 to f2 and f3 . Then, the fault f1 is completely isolated.
U13 =  −1.5 0 1  For the second fault f2 (Fig.3), the residual signals are
0−1 0 sensible to f1 and f3 . Then, the fault f2 is completely isolated.

−0.3482 −0.6963 −0.3482 0.3
 For the third fault f3 (Fig.4), the residual signals are sensible
 0.9204 −0.1593 −0.0796 0.2  to f1 and f2 . Then, the fault f3 is completely isolated.
T23 = 
 −0.1593 0.6815 −0.1593 0.4 


−0.0796 −0.1593 0.9204 0.2

283
Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.
 1
V. CONCLUSION
0
In this work, a set-membership fault isolation technique is
-1
0 5 10 15 20 25 30 35 40 45 50
proposed for continuous-time switched systems. The distur-
0.5
bances and measurement noise are assumed to be unknown
Residual signals

0 but bounded with a priori known bounds. The key point of

-0.5
0 5 10 15 20 25 30 35 40 45 50
the idea consists in using a bank of interval unknown input
1
observers in the generalized scheme. The design of interval
0 unknown input observer is ensured by a decoupling method
-1
0 5 10 15 20 25 30 35 40 45 50
to make the state partially affected by the unknown input.
Time
Cooperativity and stability conditions are given in terms of
LMIs with the fulfillment of an ADT constraint.The validity of
Fig. 2. Residuals signals of IUIO 1.
the developed technique has been illustrated through numerical
example.

1 R EFERENCES
0
[1] S. Chebotarev, D. Efimov,T. Raı̈ssi and A. Zolghadri, Interval observers
-1
0 5 10 15 20 25 30 35 40 45 50
for continuous-time LPV systems with L1/L2 performance, Automatica,
58, 2015, pp 82-89.
Residual signals

0.5

0
[2] B. Jiang, J. L. Wang and Y. C. Soh, An adaptive technique for
robust diagnosis of faults with independent effects on system outputs,
-0.5
0 5 10 15 20 25 30 35 40 45 50 International Journal of Control, 75(11), 2002, pp 792-802.
1
[3] J. L. Gouzé, A. Rapaport and M. Z. Hadj-Sadok, Interval observers for
0 uncertain biological systems, Ecological modelling, 133(1-2), 2000, pp
45-56.
-1
0 5 10 15 20 25 30 35 40 45 50 [4] H. Ethabet, D. Rabehi, D. Efimov, T. & Raı̈ssi, Interval estimation for
Time
continuous-time switched linear systems. Automatica, 90, (2018), pp
230-238.
Fig. 3. Residuals signals of IUIO 2. [5] R. N. Clark, Instrument fault detection, IEEE Transactions on Aerospace
and electronic systems, (3), (1978), pp 456-465.
[6] P. M. Frank, Fault diagnosis in dynamic systems using analytical
and knowledge-based redundancy: A survey and some new results,
Automatica, 26(3), (1990), pp 459-474.
2
[7] D. Liberzon, Switching in systems and control, Springer Science &
0
Business Media, (2012).
[8] M. Djemai, M. Defoort, Hybrid Dynamical Systems,Springer Interna-
-2
0 5 10 15 20 25 30 35 40 45 50 tional Publishing, (2015).
1
[9] M. S. Branicky, Multiple Lyapunov functions and other analysis tools for
Residual signals

0
switched and hybrid systems, IEEE Transactions on automatic control,
43(4), (1998), pp 475-482.
-1
0 5 10 15 20 25 30 35 40 45 50 [10] J. P. Hespanha, & A. S. Morse, Stability of switched systems with
2
average dwell-time, In Proceedings of the 38th IEEE Conference on
0
Decision and Control, Vol. 3, (1999), pp. 2655-2660.
[11] L. Hetel, Robust stability and control of switched linear systems,
-2
0 5 10 15 20 25 30 35 40 45 50 (Doctoral dissertation, TU Eindhoven), (2007).
Time
[12] A. Baniamerian, N. Meskin, & K. Khorasani, Geometric fault detec-
tion and isolation of two-dimensional (2D) systems,In IEEE American
Fig. 4. Residuals signals of IUIO 3. Control Conference (ACC), 2013, pp 3541-3548.
[13] H. Wang, & S. Daley, Actuator fault diagnosis: an adaptive observer-
based technique, IEEE transactions on Automatic Control, 41(7), (1996),
pp 073-1078.
[14] D. E. C. Belkhiat, N. Manamanni, N. Messai, & M. Djemai, Fault
2.5 detection & isolation for a class of hybrid systems: A dedicated switched
robust observer scheme, In 20th Mediterranean Conference on Control
& Automation (MED), 2012, pp 984-989.
2
[15] V. Cocquempot, T. El Mezyani, & M. Staroswiecki, Fault detection and
isolation for hybrid systems using structured parity residuals, In 5th
Switching signal

asian Control conference, Vol. 2, 2004, pp 1204-1212.

1.5
[16] T. Raı̈ssi, D. Efimov, & A. Zolghadri, Interval state estimation for a
class of nonlinear systems, IEEE Transactions on Automatic Control,
1
57(1), 2012, pp 260-265.
[17] D. Efimov, L. Fridman, T. Raı̈ssi, A. Zolghadri, and R. Seydou, Appli-
cation of interval observers and hosm differentiators for fault detection,
0.5
0 5 10 15 20 25 30 35 40 45 50
48th IFAC Symposium On Fault Detection, Supervision and Safety for
Time Technical Processes, Safeprocess, 2012.
[18] R. Seydou, T. Raı̈ssi, A. Zolghadri, D. Efimov, and C. Combastel, Robust
Fig. 5. Switching signal. fault diagnosis with interval continuous-time parity equations, In The
8th IFAC International Symposium on Fault Detection, Supervision and
Safety of Technical Processes, Mexico City, Mexico, 2012.

284
Authorized licensed use limited to: Beijing University of Chemical Technology. Downloaded on December 09,2024 at 04:35:38 UTC from IEEE Xplore. Restrictions apply.