Probabilistic Performance Analysis of Fault Diagnosis Schemes
Probabilistic Performance Analysis of Fault Diagnosis Schemes
Probabilistic Performance Analysis of Fault Diagnosis Schemes
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
If the residual generator is partitioned as<br />
then the residual can be written as<br />
r = F 1 y + F 2 u<br />
]<br />
F =<br />
[F 1 F 2 ,<br />
= (F 1 G 1,ϑ + F 2 )u + F 1 G 2,ϑ v + F 1 G 3,ϑ w + F 1 G 4,ϑ f (ϑ)<br />
= (F 1 G 1,ϑ + F 2 )(u ◦ + ũ) + F 1 G 2,ϑ v + F 1 G 3,ϑ w + F 1 G 4,ϑ<br />
(<br />
f ◦ (ϑ) + ˜ f ) .<br />
Divide the residual into the sum <strong>of</strong> its nominal, uncertain, and random parts as follows:<br />
where<br />
r = r nom + r unc + r rnd ,<br />
r nom = (F 1 G 1,ϑ + F 2 )u ◦ + F 1 G 4,ϑ f ◦ (ϑ),<br />
r unc = (F 1 G 1,ϑ + F 2 )ũ + F 1 G 3,ϑ w + F 1 G 4,ϑ ˜ f ,<br />
r rnd = F 1 G 2,ϑ v.<br />
Since v is zero-mean by assumption, the conditional mean <strong>of</strong> the residual at time k is<br />
and the conditional variance at time k is<br />
ˆr k = E(r k | θ 0:k = ϑ 0:k ) = r nom<br />
k<br />
+ r unc<br />
k<br />
,<br />
Σ k = E ( (r k − ˆr k ) 2) ( (r ) )<br />
rnd 2<br />
= E .<br />
Note that Assumption 1 holds because the variance Σ is not affected by any <strong>of</strong> the uncertain<br />
signals ũ, w, or ˜ f . However, Assumption 2 only holds if the operator G 2,ϑ does not depend<br />
on the fault parameter ϑ. That is,<br />
]<br />
G ϑ =<br />
[G 1,ϑ G 2 G 3,ϑ G 4,ϑ .<br />
A convenient choice is to take G 2 = I , which corresponds to additive measurement noise<br />
injected between the plant G ϑ and the residual generator F .<br />
k<br />
87