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.

Proposition 4.32. Let N be the final time used in Algorithm 4.1, and let Θ = {0,1,...,m}.<br />

In additions to the assumptions on {θ k }, G θ , F , and δ made above, assume that the fault<br />

input f k (ϑ 0:k ) can be computed in O(1) time, for any k and ϑ 0:k . Then, the total running time<br />

<strong>of</strong> Algorithm 4.1 is O(N m+1 ).<br />

Pro<strong>of</strong>. Because θ is assumed to be a tractable Markov chain, the for all-loop over possible<br />

sequences ϑ 0:N executes O(N m ) times. Line 4 is a simple look-up and Line 6 is a single<br />

multiplication, so Lines 3–7 take O(1) time to compute. Since f k (ϑ 0:k ) can be computed<br />

in O(1) time, Lines 8–11 can be computed in O(1) time, as well. By assumption, the decision<br />

function δ is such that Line 12 can be computed in O(1) time. Clearly, the remaining<br />

computations (Line 13–19) can also be computed in O(1) time. Since each individual line<br />

takes O(1) time, we conclude that each iteration <strong>of</strong> the for-loop over k takes O(1) time.<br />

Therefore, the total running time <strong>of</strong> Algorithm 4.1 is O(N m+1 ).<br />

4.5.2 LTV Special Case Based on Component Failures<br />

In this section we present a special system structure, based on Sections 4.2.2 and 4.3.3, that<br />

permits a more straightforward implementation <strong>of</strong> Algorithm 4.1. Suppose that the system<br />

consists <strong>of</strong> L components that fail independently at random, and assume that system is<br />

only affected by additive faults. Hence, the combined dynamics <strong>of</strong> the system G θ and the<br />

residual generator F are given by<br />

η k+1 = A k η k + B u,k u k + B v,k v k + B f<br />

r k = C k η k + D u,k u k + D v,k v k + D f<br />

L∑ (<br />

ϕ j k − κj (θ 0:k ) ) ,<br />

j =1<br />

L∑ (<br />

ϕ j k − κj (θ 0:k ) ) ,<br />

where κ j (θ 0:k ) is the random time at which the j th component fails. Because θ 0:k only affects<br />

the system via the random failure times, specifying a particular parameter sequence ϑ 0:N is<br />

equivalent to specifying the corresponding failure times ˆκ j := κ j (ϑ 0:N ), for j = 1,2,...,L.<br />

Another important feature <strong>of</strong> this special case is the additive structure <strong>of</strong> the fault input.<br />

Since each ϕ j enters additively, the portion <strong>of</strong> the residual due to each ϕ j can be computed<br />

separately and then combined using the principle <strong>of</strong> superposition. Similarly, the portion<br />

<strong>of</strong> the residual due to the initial condition η 0 and the known input u 0:N can be computed<br />

separately. Because ϕ j has no effect until the j th component fails (i.e., ϕ j (k − ˆκ j ) = 0,<br />

for k < ˆκ j ), we only need to compute the portion <strong>of</strong> the residual due to ϕ j for k ≥ ˆκ j .<br />

The procedure for computing the performance metrics for this special case is split<br />

into two parts: Algorithm 4.2 computes each portion <strong>of</strong> the residual, while Algorithm 4.3<br />

computes the performance metrics. Although Algorithm 4.2 applies to any system <strong>of</strong> L<br />

components, Algorithm 4.3 focuses on the case L = 2. This greatly simplifies the presentation<br />

j =1<br />

70

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

Saved successfully!

Ooh no, something went wrong!