ZHANG et al.: ANALYSIS AND DESIGN OF ROBUST H
∞
FAULT ESTIMATION OBSERVER 1227
is solvable with respect to matrix of compatible dimension
if and only if the following inequalities hold:
N
T
N
< 0, N
T
N
< 0(6)
where N
and N
, respectively, are any basis of the nullspace
of and .
III. M
AIN RESULTS
A. Analysis of Finite-Frequency Fault Estimation Observer
For the T-S fuzzy model (3), we design fuzzy fault estima-
tion observer as follows:
⎧
⎪
⎪
⎨
⎪
⎪
⎩
ˆx(k + 1) = A(h)ˆx(k) + B(h)u(k) + E(h)
ˆ
f (k)
− L(h)(ˆy(k) − y(k))
ˆy(k) = C(h)ˆx(k)
ˆ
f (k + 1) =
ˆ
f (k) − F(h)(ˆy(k) − y
(k))
(7)
where ˆx(k) ∈ R
n
is the observer state vector, ˆy(k) ∈ R
p
is the
observer output vector,
ˆ
f (k) ∈ R
r
is an estimate of f (k) and
L(h) ∈ R
n×p
, F(h) ∈ R
r×p
are fuzzy observer gain matrices
to be determined.
Denote the following error vectors:
e
x
(k) =ˆx(k) − x(k), e
y
(k) =ˆy(k) − y(k)
e
f
(k) =
ˆ
f (k) − f (k)
then it gets
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
e
x
(k + 1) =
A(h) − L(h)C(h)
e
x
(k) + E(h)e
f
(k)
+
L(h)D
2
(h) − D
1
(h)
d(k)
e
y
(k) = C(h)e
x
(k) − D
2
(h)d(k)
e
f
(k + 1) = e
f
(k) − F(h)C(h)e
x
(k) + F(h)D
2
(h)d(k)
−f (k)
(8)
where f (k) = f (k + 1) − f (k).
Furthermore, for the error dynamics (8), the augmented
system can be as follows:
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
e
x
(k + 1)
e
f
(k + 1)
=
A(h) − L(h)C(h) E(h)
−F(h)C(h) I
r
×
e
x
(k)
e
f
(k)
+
L(h)D
2
(h) − D
1
(h)
F(h)D
2
(h)
d(k)
−
0
n×r
I
r
f (k)
e
f
(k) =
0
r×n
I
r
e
x
(k)
e
f
(k)
.
(9)
Denote new vectors and variables as follows:
¯e(k) =
e
x
(k)
e
f
(k)
,ν(k) =
d(k)
f (k)
¯
A(h) =
A(h) E(h)
0
r×n
I
r
,
¯
L(h) =
L(h)
F(h)
¯
C(h) =
C(h) 0
p×r
,
¯
D
1
(h) =
D
1
(h)
0
r×d
¯
I
r
=
0
n×r
I
r
then we have
⎧
⎨
⎩
¯e(k + 1) =
¯
A(h) −
¯
L(h)
¯
C(h)
¯e(k)
+
¯
L(h)D
2
(h) −
¯
D
1
(h)
d(k) −
¯
I
r
f (k)
e
f
(k) =
¯
I
T
r
¯e(k).
(10)
Next, Theorem 1 presents a fuzzy fault estimation observer
design with finite-frequency specifications for T-S fuzzy
models.
Theorem 1: Given the region D(α, τ ) and H
∞
perfor-
mance indexes γ
1
,γ
2
, the eigenvalues of
¯
A(h) −
¯
L(h)
¯
C(h)
locate in the specified region D(α, τ ) and the error dynam-
ics (10) satisfies H
∞
performances e
f
(k)
2
<γ
1
d(k)
2
and e
f
(k)
2
<γ
2
f (k)
2
, if there exist symmetric posi-
tive definite matrices
¯
P
1
(h),
¯
Q
1
,
¯
Q
2
∈ R
(n+r)×(n+r)
, symmetric
matrices
¯
P
2
(h),
¯
P
3
(h) ∈ R
(n+r)×(n+r)
, and matrices
¯
S(h) ∈
R
(n+r)×(n+r)
,
¯
Y(h) ∈ R
(n+r)×p
satisfying
¯
P
1
(h
+
) −
¯
S(h) −
¯
S
T
(h)
∗
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h) − α
¯
S(h)
− τ
2
¯
P
1
(h)
< 0 (11)
⎡
⎢
⎢
⎣
φ
d11
φ
d12
¯
Y(h)D
2
(h) −
¯
S(h)
¯
D
1
(h) 0
∗ φ
d22
0
¯
I
r
∗∗ −γ
1
I
d
0
∗∗ ∗ −γ
1
I
r
⎤
⎥
⎥
⎦
< 0
for the low-frequency disturbance |ϑ
d
|≤ϑ
dl
(12a)
⎡
⎢
⎢
⎣
ϕ
d11
ϕ
d12
¯
Y(h)D
2
(h) −
¯
S(h)
¯
D
1
(h) 0
ϕ
d21
ϕ
d22
0
¯
I
r
∗∗ −γ
1
I
d
v0
∗∗ ∗ −γ
1
I
r
⎤
⎥
⎥
⎦
< 0
for the middle-frequency disturbance ϑ
d1
≤ ϑ
d
≤ ϑ
d2
(12b)
⎡
⎢
⎢
⎣
ψ
d11
ψ
d12
¯
Y(h)D
2
(h) −
¯
S(h)
¯
D
1
(h) 0
∗ ψ
d22
0
¯
I
r
∗∗ −γ
1
I
d
0
∗∗ ∗ −γ
1
I
r
⎤
⎥
⎥
⎦
< 0
for the high-frequency disturbance |ϑ
d
|≥ϑ
dh
(12c)
where φ
d11
= ϕ
d11
= ψ
d11
=
¯
P
2
(h
+
) −
¯
S(h) −
¯
S
T
(h), φ
d12
=
¯
Q
1
+
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h), φ
d22
=−2 cos(ϑ
dl
)
¯
Q
1
−
¯
P
2
(h), ϕ
d12
= e
jϑ
dc
¯
Q
1
+
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h), ϕ
d21
=
e
−jϑ
c
¯
Q
1
+
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h)
T
,ϕ
d22
=−2 cos(ϑ
dw
)
¯
Q
1
−
¯
P
2
(h), ψ
d12
=−
¯
Q
1
+
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h), ψ
d22
=
2 cos(ϑ
dh
)
¯
Q
1
−
¯
P
2
(h)
⎡
⎢
⎢
⎣
φ
f 11
φ
f 12
−
¯
S(h)
¯
I
r
0
∗ φ
f 22
0
¯
I
r
∗∗−γ
2
I
r
0
∗∗ ∗ −γ
2
I
r
⎤
⎥
⎥
⎦
< 0
for the low-frequency fault |ϑ
f
|≤ϑ
fl
(13a)
⎡
⎢
⎢
⎣
ϕ
f 11
ϕ
f 12
−
¯
S(h)
¯
I
r
0
ϕ
f 21
ϕ
f 22
0
¯
I
r
∗∗−γ
2
I
r
0
∗∗ ∗ −γ
2
I
r
⎤
⎥
⎥
⎦
< 0
for the middle-frequency fault ϑ
f 1
≤ ϑ
f
≤ ϑ
f 2
(13b)
⎡
⎢
⎢
⎣
ψ
f 11
ψ
f 12
−
¯
S(h)
¯
I
r
0
∗ ψ
f 22
0
¯
I
r
∗∗−γ
2
I
r
0
∗∗ ∗−γ
2
I
r
⎤
⎥
⎥
⎦
< 0
for the high-frequency fault |ϑ
f
|≥ϑ
fh
(13c)
where φ
f 11
= ϕ
f 11
= ψ
f 11
=
¯
P
3
(h
+
) −
¯
S(h) −
¯
S
T
(h), φ
f 12
=
¯
Q
2
+
¯
S(h)
¯
A(h) −
¯
Y(h)
¯
C(h), φ
f 22
=−2 cos(ϑ
fl
)
¯
Q
2
−
¯
P
3
(h),