Академический Документы
Профессиональный Документы
Культура Документы
Abstract: In this paper, the approximation of nonlinear systems using unscented Kalman
filter (UKF) is discussed, and the conditions for the convergence of the UKF are derived.
The detection of faults from residuals generated by the UKF is presented. As fault
detection often reduced to detecting irregularities in the residuals, such as the mean, the
local approach, a powerful statistical technique to detect such changes, is used to detect
fault from the residuals generated from the UKF. The properties of the proposed method
are also presented. To illustrate the performance of the proposed method, it is applied to
detect faults in the attitude sensors of a satellite. Copyright © 2005 IFAC
∑w
degree of freedom to "fine tune" the higher order
moments of the approximation, and is usually set to a γ i (k ) = h( χ i (k | k − 1)) , yˆ (k ) = i γ i (k )
m
i =0
small positive value. The sample mean and
covariance of χ are (Wan and Merwe, 2000): the covariance and the cross correlation matrix by,
2L
∑ wic [γ i (k ) − yˆ (k )][γ i (k ) − yˆ (k )]T + R (k )
2L 2L
Pεε =
χ = ∑
i =0
wi χ i ; Pχ = ∑i =0
wi ( χ i − χ )( χ i − χ )
T
(3) i =0
2L
⎧ 1 Pxy = ∑ wic [ χ i (k | k − 1) − xˆ (k | k − 1)][γ i (k ) − yˆ (k )]T
⎪⎪wi = 1 − 2 i=0 i =0
where ⎨ a
1 And the predicted state is computed using the
⎪ wi = i = 1,L,2 L classical Kalman filter,
⎪⎩ 2 La 2
⎧⎪ xˆ (k ) = xˆ (k | k − 1) + Pxy Pεε−1[ y(k ) − yˆ (k )]
Let γi = f(χi) ∈ RL, for i = 0, 1, …, 2L. The mean and ⎨ (9)
−1 T
covariance for y can be approximated by the sample ⎪⎩P(k ) = P(k | k − 1) − Pxy Pεε Pxy
mean and covariance of γ, as given below.
2L 2L Step 4 Repeat steps 1 to 3 for the next sample.
γ = ∑ wi γ i ; Pγ = ∑ wi (γ i − γ )(γ i − γ )T (4)
i =0 i =0 Since the mean and covariance of x(k) are accurate
The properties of the UT are (Julier, et al., 2000): up to second order, and the same also applied to the
computed mean and covariance of y(k), the UKF can
predict with a second-order accuracy, but without the where U = I + H T α (k ) Pεε−1Pxy
T
+ Pxy Pεε−1α (k ) H +
need to compute the Jacobian or Hessian matrix. In
contrast, the state vector computed by the EKF is H T α (k ) Pεε−1Pxy
T
Pxy Pεε−1α (k ) H , and λmax (⋅) is the
only a first-order approximation of the nonlinear
maximum eigenvalue of the matrix F, then
system, and hence can only achieve the first-order
accuracy. Further, the computational load of UKF is lim ~
x (k ) = 0 (20)
k →∞
only in the same order as that for the EKF.
Proof: Let V (k ) = ~ x (k )T ~
x (k ) . From (9), (11), (17)
2.3 Convergence analysis of the UKF and (18),
The convergence analysis of the UKF is derived V (k ) − V (k − 1) = ~
x (k )T ~
x (k ) − ~
x (k − 1)T ~
x (k − 1)
using an approach similar to that of the EKF = [ x (k | k − 1) + P P e(k )] [ x (k | k − 1) + P P −1e(k )] −
~ −1
xy εε
T ~
xy εε
(Boutayeb, et al., 1997). Denote the error of the ~
x (k − 1) T ~
x (k − 1)
estimated state by
~ =~
x ( k − 1) [ F T β (k )Uβ (k ) F − I ]~
T
x (k − 1)
(21)
x (k ) = x(k ) − xˆ (k ) (10)
From the Rayleigh-Ritz theorem (Yu and Shi, 2004),
and the prediction error of the state by for any vector z ≠ 0,
~
x (k | k − 1) = x(k ) − xˆ (k | k − 1) (11) x≠ 0
(
λ max (U ) = max z T Uz ( z T z ) )
Assuming that w(k) is neglectable, expanding x(k)
−1
x ≠0
(
λ min ( F −T F ) = min z T F −T F −1 z ( z T z ) )
given by (5) by a Taylor Series about xˆ (k − 1) If assumption (19) hold, then
gives,
x( k ) = f ( xˆ (k − 1)) + f ′( xˆ ( k − 1)) ~
x ( k − 1)
12
⎛ z T F −T F −1 z ⎞ (
⎡ min z T F −T F −1 z ( z T z ) )⎤⎥ 12
⎜ ⎟ ≥ ⎢ z ≠0
1
+f ′′( xˆ (k − 1)) ~
x (k − 1) 2 + L (12)
⎜
⎝ z T Uz ⎟
⎠ ⎣⎢ z ≠0
(
⎢ max z T Uz ( z T z ) ) ⎥
⎦⎥
2 12
Similarly, xˆ (k | k − 1) can be expressed as, ⎡ λ ( F −T F −1 ) ⎤
= ⎢ min ⎥ > λ max ( β (k )) ≥ β j (k ) (22)
xˆ (k | k − 1) = f ( xˆ (k − 1)) +
1
f ′′( xˆ (k − 1)) P(k − 1) ⎣⎢ λ max (U ) ⎦⎥
2 where the subscript j denotes the jth component of the
+ … (13) diagonal matrix β(k). From (22), the following
hence, ~ x (k | k − 1) can be approximated by, inequality is obtained:
~
x (k | k − 1) ≈ F~ x (k − 1) (14) zT [ β j (k )Uβ j (k ) − F −T F −1]z < 0 (23)
where F = f ′( xˆ (k − 1)). Assuming that v(k) is small, As z ≠ 0, hence
expanding y(k ) and yˆ (k ) about xˆ (k | k − 1) gives, β (k )Uβ (k ) − F −T F −1 < 0 (24)
y (k ) = h( xˆ (k | k − 1)) + h′( xˆ (k | k − 1)) ~
x (k | k − 1) For ~x (k − 1) ≠ 0 , it follows that
1 ~
x (k − 1)T [ F T β (k )Uβ (k ) F − I ]~
x (k − 1) < 0
+ h′′( xˆ ( k | k − 1)) ~
x ( k | k − 1) 2 + L (25)
2
1
From (25) and (21), V (k ) − V (k − 1) < 0 , V(k) is a
yˆ (k ) = h( xˆ (k | k − 1)) + h′′( xˆ (k | k − 1)) P (k | k − 1) + L
2 decreasing sequence, and hence lim V ( k ) = 0 . It
k →∞
Similarly, ε (k ) = y(k ) − yˆ (k ) can be approximated
follows that lim ~x ( k ) = 0 .
by, k →∞
ε (k ) ≈ H~x (k | k − 1) (16) The unknown diagonal matrices α(k) and β(k) are
where H = h′( xˆ (k | k − 1)). In general, ε(k) is not introduced to evaluate the UT of the state variables
that propagates through the nonlinear model. If the
identically zero, as it is a second order approximation
magnitude of the eigenvalue of β(k) is sufficiently
of ~x (k | k − 1). Hence, (14) and (16) are modified as, small, the convergence of the UKF is ensured. If the
~
x (k | k − 1) = β (k ) F~x (k − 1) (17) magnitude of αik are small enough, the convergence
~
ε (k ) = α (k ) Hx (k | k − 1) (18) of the UKF may be improved in the sense that the
where α(k) = diag(α1(k), α2(k), …, αN(k)) and domain of λmax(β(k)) will be enlarged. Indeed, the
β (k ) = diag ( β1 (k ), β 2 (k ), L , β L (k )) are unknown sufficient conditions (19) mean that if the error
introduced by the UT is small enough, V(k) is a
diagonal matrices. The sufficient condition for the decreasing sequence. As α(k) and β(k) are unknown
convergence of the UKF is given below. factors, sigma points should be chosen properly to
decrease the error of the UT so that (20) is fulfilled.
Theorem 1: Assuming F is a nonsingular matrix, and
β(k) satisfies the following condition:
12
⎡ λ min ( F −T F −1 ) ⎤ 3. FAULT DETECTION BY LOCAL APPROACH
λ max ( β (k )) < ⎢ ⎥ (19)
⎣⎢ λ max (U ) ⎦⎥ The measurement equation (5) can be rewritten as,
y (k ) = h ( x(k )) + ψ (k ) + v(k ) (26)
where h ( x(k )) is a measurement model, and where N is a large positive integer, the subscript i
denotes the ith component of vector.
ψ (k ) = h( x(k )) − h ( x(k )) is the modelling error.
Step 3 At the kth sampling period, the cumulative sum
Consider the predicted observation yˆ (k ) obtained
of residuals is computed from (29) as given below.
from the UKF. Under normal operating condition, 1 k
the residual of the UKF is, Dim (k ) = ∑ (ε i (t ) − bi (0)) (34)
m t = k − m +1
ε (k ) = y (k ) − yˆ (k ) = h ( x(k )) + ψ (k ) + v(k ) − h ( xˆ (k ))
where k > m . Normalizing the cumulative sum of
= ψ ( k ) + φ ( k ) + v( k ) (27)
the residual by its variance gives,
where yˆ (k ) = h ( xˆ (k )) is the predicted observation Sim (k ) = [ Dim (k )]2 [ Pεεii (k − m)]−1 (35)
and φ (k ) = h ( x(k )) − h ( xˆ (k )) is the estimation error. where Pεεii is the ith diagonal element Pεε.
When there is a sensor fault, the residual becomes,
ε (k ) = y (k ) + b f − yˆ (k ) Step 4 If Sim (k ) ≤ λi , then there is no fault, but a
= b f + ψ (k ) + φ (k ) + v(k ) (28) fault otherwise. As S im (k ) is χ2-distributed, λi can be
where bf ≠ 0 is the output arising from the sensor obtained from χ2-table for a given confidence level.
fault. However, faults can only be detected if the
Step 5 Repeat step 3 and 4.
term is large compared with the modelling errors and
the system noise. For small faults, it is difficult to
3.2 Properties of the fault detection method
detect bf from ε(k).
If there is no fault, ε(k) is Gaussian distributed: N(0,
The local approach is now applied to the residuals Pεε). From (35), the expectation of ε(k) and the
generated by the UKF. In the local approach, the covariance matrix of Dm are respectively:
cumulative sum of the residual Dm is computed for a 1 k
window size of m samples (Wang and Chan, 2002), E(Dm ) = ∑ E (ε (t )) = 0 (36)
m t = k − m +1
1 m
Dm = ∑ ε (k ) 1 k
m k =1 Cov ( D m ) = ∑ Cov (ε (t )) = Pεε (37)
m t = k − m +1
1 ⎛ m m m ⎞
= ⎜ ∑ψ (k ) + ∑ φ (k ) + ∑ v(k ) ⎟ (29) where E(.) and Cov(.) are respectively the
⎜
m ⎝ k =1 ⎟
k =1 k =1 ⎠ expectation and the covariance. Hence Dm is also
Assuming the model is accurate and ψ(k) = 0, then Gaussian distributed: N(0, Pεε). If there is a fault, the
the residual Dm can now be approximated by distribution of ε(k) is: N(bf, Pεε), and the mean and
1 m 1 m covariance of Dm are:
Dm = ∑ φ (k ) + ∑ v (k ) (30)
E ( Dm ) = mb f , Cov ( D m ) = Pεε (38)
m k =1 m k =1
From Theorem 1, lim ( x(k ) − xˆ (k )) = 0 holds under The distribution of Dm is: N ( mb f , Pεε ) . The
k →∞
certain conditions. Assuming h(.) is a continuous miss-detection of the proposed fault detection
function, then scheme is given in the following theorem. This result
lim [h ( x(k )) − h ( xˆ (k ))] = lim φ (k ) = 0 (31) provides a guideline for choosing m and the
k →∞ k →∞ probability of the miss-detection. The argument t and
Consequently, if the sufficient condition (19) is the subscript i are ignored for simplicity.
satisfied and k is sufficiently large, Dm is Gaussian
distributed with zero mean. If there is a sensor fault, Theorem 2: Let λ be obtained for a given confidence
(30) becomes, level. A fault is detected, if S m = ( D m ) 2 Pεε−1 > λ .
1 ⎛⎜ ⎞
m m m The false alarm PF is independent of m, while the
Dm = ∑
m ⎜⎝ k =1
bf + ∑
φ (k ) + ∑ v(k ) ⎟
⎟
(32) miss-detection PM depends on m.
k =1 k =1 ⎠
As bf is non-zero, Dm is also non-zero. Proof: If there is no fault, the distribution of Dm is
N(0, Pεε), and the probability density function (pdf)
3.1 Fault detection method of Dm is:
The proposed fault detection scheme can be
implemented on-line as follows:
p( D m | H 0 ) =
1
2πPεε
(
exp − ( D m )2 (2 Pεε ) (39) )
Step 1 Select m, the window size for computing the Let the null hypothesis denoting no fault be H0. From
cumulative sum of residual. (35), S 2 = ( D m )2 Pεε−1 , the pdf of Sm is given by,
( )
Step 2 Compute the mean of the residual generated 2 Pεε
from the UKF. This is necessary, as ψ(k) is ignored p( S m | H 0 ) = exp − S m 2
in the above analysis. 2πPεε 2 Pεε S m
( )
N
∑ ε (k )
1 1
bi (0) = i (33) = exp − S m 2 (40)
N k =1 2πS m
The false alarm PF is defined by, ⎡ γ& ⎤ ⎡ γ ⎤ ⎡ ω Ix ⎤
∞ ⎢ & ⎥ ⎢ θ ⎥ ⎢ω + ω ⎥
PF = ∫λ ⎢θ ⎥
m m
p( S | H 0 )dS (41) ⎢ ⎥ ⎢ Iy 0⎥
⎢ ψ& ⎥ ⎢ ψ ⎥ ⎢ ω Iz ⎥
Since p(S |H0) is independent of m, PF is also
m
⎢& ⎥ ⎢ ⎥ ⎢ ⎥
independent of m. If there is a fault, the distribution ⎢ bIx ⎥ ⎡ A11 − I 3×3 − I 3×3 ⎤ ⎢ bIx ⎥ ⎢ 0 ⎥
⎢ b& ⎥ = ⎢0 0 0 ⎥⎢ ⎥ + ⎢
b 0 ⎥ +W
of Dm becomes N ( mb f , Pεε ) , and the pdf is: ⎢ Iy ⎥ ⎢ 3×3 3×3 3×3 ⎥ Iy
⎢ ⎥ ⎢ ⎥
⎢ b&Iz ⎥ ⎢⎣03×3 03×3 A33 ⎥⎦ ⎢ bIz ⎥ ⎢ 0 ⎥
⎡ ( D m − mb f ) 2 ⎤
1 ⎢ d& ⎥ ⎢d ⎥ ⎢ 0 ⎥
p( D m | H1) = exp ⎢− ⎥ (42) ⎢ Ix ⎥ ⎢ Ix ⎥ ⎢ ⎥
2πPεε ⎢
⎣
2 Pεε ⎥
⎦ ⎢d&Iy ⎥ ⎢d Iy ⎥ ⎢ 0 ⎥
⎢& ⎥ ⎢ ⎥ ⎢ ⎥
Let H1 be the hypothesis that there is a fault. Then ⎣ d Iz ⎦ ⎣ d Iz ⎦ ⎣ 0 ⎦
the pdf of Sm can be expressed as, ⎡ − ψS x0 + S y0 + γS z0 ⎤
p ( S m | H1 ) =
1 1
{exp[ − ( S m − m Pεε b f ) 2 ] ⎢ ⎥
2 ⎢ θS x0 − γS y0 + S z0 ⎥
2 2πS m
⎡ mγ ⎤ ⎢ 0 0 ⎥
⎢ m ⎥ ⎢ S x + ψS y − θS z ⎥
0
1
+ exp[− ( S m + m Pεε b f )2 ]} (43) ⎢ θ⎥ ⎢ 0 ⎥
⎢mψ ⎥ = ⎢ θS x − γS y + S z ⎥ + V
0 0
2 (46)
From (43), the miss-detection PM is given by, ⎢ ⎥ ⎢ S 0 + ψS 0 − θS 0 ⎥
λ ⎢ γh ⎥ ⎢ x y z ⎥
PM (m) = ∫0 p ( S m | H1)dS m ⎢ θ h ⎥ ⎢ − ψS x0 + S y0 + γS z0 ⎥
⎣ ⎦ ⎢ ⎥
= ∫0
λ 1 1
{exp[− ( x − m Pεε b f ) 2 ] ⎢ γ ⎥
2π 2 ⎢⎣ θ ⎥⎦
1 where
+ exp[− ( x + m Pεε b f ) 2 ]}dx
2 ⎡ 0 0 ω0 ⎤
( ) (
= Φ − m Pεε b f + λ − Φ − m Pεε b f ) A11 = ⎢ 0⎢ 0 0 ⎥⎥,
+ Φ ( m P b + λ )− Φ ( m P b )
⎢⎣− ω 0 0 0 ⎥⎦
εε f (44) εε f
⎡− 1 τ Ix 0 ⎤
exp(− x 2 2)dx . Therefore P
0
1 ⎢ 0 ⎥⎥
x
where Φ = ∫−∞ M A33 = ⎢ 0 − 1 τ Iy
2π
depends on m. ⎢⎣ 0 0 − 1 τ Iz ⎥⎦
I3×3 is the identity matrix, 03×3, the zero matrix, γ, θ
The relation between PM and m is shown in Fig. 2, and ψ are the roll, the pitch and the heading of
where the shaded part of the curve is PM. Clearly, if satellite, ω0 is orbit angle velocity, ωIx, ωIy, ωIz are
m is large, the miss-detection from (44) is small, the measurement from gyroscope, bIx, bIy, bIz, dIx, dIy,
lim PM (m) = 0 (45) dIz are the drifting errors of the gyroscope, τIx, τIy, τIz
m →∞
However, if m is large, a longer time is required are the first order Markov time constant, S x0 , S y0 , S z0
before faults are detected (Wang and Chan, 2002). If are the projections of sun vector onto the coordinate
the false alarm and the miss-detection are chosen to
of the spacecraft, mγ, mθ, mψ are measurements of sun
be small, λ and m can be determined from (41) and
sensor, γh, θh are measurements of earth sensor, W
(44), as illustrated in the example presented below.
and V are zero mean Gaussian white noise.
When the fault occurs, the residuals ε1(k) and ε4(k) ACKNOLEDGEMENT
are shown in Fig. 3, showing a small step change in
the mean of ε1(k), for k > 300. For m = 6, S1m (k ) This work was supported in part by China Natural
Science Foundation (No. 60234010), and the
and S 4m (k )
are shown in Fig. 4. As only S1m (k )
is HKSAR RGC Grant (HKU 7050/02E).
greater than the threshold for k > 302, a fault is
detected in the component of the sun sensor, which
corresponds to S1m (k ) . This result agrees with the REFERENCES
properties of the fault detection method presented in
Boutayeb, M., H. Rafaralahy, and M. Darouach
section 3.2, illustrating the ability of the local
(1997). Convergence analysis of the extended
approach in detecting faults.
Kalman filter used as an observer for nonlinear
deterministic discrete-time systems. IEEE Trans.
Auto. Contr., 42, 581-586.
Del Gobbo, D., M. Napolitano, P. Famouri and M.
Innocenti (2001). Experimental application of
extended Kalman filtering for sensor validation.
IEEE Trans. Contr. Sys. Technology, 9, 376-380.
Hall, D. L. and J. Llinas (2001), Handbook of
Multisensor Data Fusion, CRC Press.
Julier, S., J. Uhlmann and H. F. Durrant-Whyte
(2000). A new method for the nonlinear
transformation of means and covariances in Filters
and Estimators. IEEE Trans. Auto. Contr. 45,
477-482.
Magnus Norgaard,Niels K. Poulsen, Ole Ravn(2000),
New developments in state estimation for
Fig. 3 ε1(k) and ε4(k) nonlinear systems, Automatic, 36, 1627-1638
Wan, E. A. and R. van der Merwe (2000). The
unscented Kalman filter for nonlinear estimation,
adaptive systems for signal processing.
Communication and Contr. Symposium, 153-158.
Wang, Y. and C. W. Chan (2002). Asymptotic local
approach in fault detection with faults modeled by
neurofuzzy networks. Proceedings 15th Triennial
World Congress, Barcelona, Spain.
Yang, J. (2002). Attitude determination, positioning
and fault diagnosis of GPS. Ph.D. Dissertation,
Beihang University.
Yu, S. X. and J. Shi (2004). Segmentation given
partial grouping constraints. IEEE Trans. on
Pattern Analy. and Mach. Intellig., 26, 173-183.
Zhang, Q., M. Basseville and A. Benveniste (1994).
Fig. 4 S1m (k ) and S 4m (k ) Early warning of slight changes in system.
Automatica, 30, 95-115.
Zhang, Q., M. Basseville and A. Benveniste (1998).
5. CONCLUSION Fault detection and isolation in nonlinear dynamic
systems: a combined input-output and local
In this paper, a fault detection scheme for nonlinear approach. Automatica, 34, 1359-1373.
systems is derived based on the UKF and the local