08.11.2014 Views

Probabilistic Performance Analysis of Fault Diagnosis Schemes

Probabilistic Performance Analysis of Fault Diagnosis Schemes

Probabilistic Performance Analysis of Fault Diagnosis Schemes

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

where<br />

[<br />

T (β) T T (G 11,0 ) T T (z) + T (z) T T (G 11,0 )T (β) + T (z) T T (z) T (β) T<br />

J (β) :=<br />

( ) −1<br />

T (β)<br />

1<br />

].<br />

I − T (G<br />

γ 2 11,0 ) T T (G 11,0 )<br />

Note that the subscript N has been omitted from the operators T and T for clarity.<br />

Since u and f (0) are known, r unc is an affine function <strong>of</strong> β. Also, the signal z is fixed, so<br />

the function J (β) is linear in β, and the constraint J (β) ≽ 0 is a linear matrix inequality<br />

(lmi). Therefore, this optimization can be cast as a semidefinite program (sdp), which is a<br />

type <strong>of</strong> convex program that is readily solved with numerical optimization packages, such as<br />

SeDuMi [90].<br />

Case 2. Suppose that ∆ belongs to the set ∆ 2,ltv (γ) and assume that G 11,0 = 0 (i.e., ∆ does<br />

not experience feedback). Then, for k = 0,1,..., N , applying Theorem 5.9 yields the following<br />

optimization:<br />

ˆr ⋆ k = maximize<br />

β<br />

∣ r<br />

unc∣<br />

k<br />

subject to r unc = F 2 G 21,0 β + F 2 G 23,0 f (0) + (F 2 G 24,0 + F 1 )u<br />

α = G 13,0 f (0) +G 14,0 u<br />

‖τ l β‖ 2 ≤ γ‖τ l α‖ 2 ,<br />

l = 0,1,...,k.<br />

As in Case 1, r unc is affine in β. Since the k + 1 inequality constraints are quadratic in<br />

k<br />

β 0:N , this optimization problem is a socp. As previously mentioned, socps are readily solved<br />

with numerical optimization packages.<br />

Minimizing the Probability <strong>of</strong> Detection<br />

Let ϑ be a fault parameter sequence such that ϑ N ≠ 0, and let k f be the fault time, as defined<br />

in equation (5.2). Recall that the worst-case probability <strong>of</strong> detection is<br />

Pd ⋆ = 1 − max min P( )<br />

|r k | < ε k | θ 0:k = ϑ 0:k .<br />

∆∈P ∆ k f ≤k≤N<br />

By Proposition 5.2, the optimum values <strong>of</strong> ∆ and k are obtained by solving<br />

ˆµ ⋆ = min<br />

∆∈P ∆<br />

max<br />

k f ≤k≤N<br />

| ˆr k |<br />

<br />

Σk<br />

.<br />

97

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!