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.
for all t ∈ T , the autocorrelation function <strong>of</strong> x is defined as<br />
R x (s, t) := E(x s x T t ),<br />
for all s, t ∈ T , and the autocovariance function <strong>of</strong> x is defined as<br />
C x (s, t) := E( (xs<br />
− m x (s) )( x t − m x (t) ) T ) ,<br />
for all s, t ∈ T . The random process {x t } is said to be strictly stationary if<br />
p(x t1 , x t2 ,..., x tm ) = p(x t1 +τ, x t2 +τ,..., x tm +τ)<br />
for all finite sets <strong>of</strong> indices t 1 , t 2 ,..., t m ∈ T , where m ∈ N, and all τ ≥ 0. The random process<br />
{x t } is said to be wide-sense stationary (wss) if for some constant ¯m ∈ R n ,<br />
m x (t) = ¯m,<br />
for all t ∈ T , and for any τ ∈ T ,<br />
R x (s + τ, s) = R x (t + τ, t),<br />
for all s, t ∈ T . If {x t } is wss, then R x only depends on the difference between its arguments<br />
and we may write R x (s + τ, s) = R x (τ), for all s,τ ∈ T . Given a wss process {x t }, the power<br />
spectral density <strong>of</strong> x is defined as<br />
∫<br />
S x (ξ) := F(R x )(ξ) =<br />
e −2πiξτ R x (τ) dτ,<br />
where F is the Fourier transform operator.<br />
2.2.6 Common Probability Distributions<br />
1. A Gaussian random variable x : Ω → R n with mean µ ∈ R n and variance Σ ∈ R n×n , such<br />
that Σ ≻ 0, is defined by the pdf<br />
p x (s) :=<br />
1<br />
<br />
(−<br />
(2π) n |Σ| exp 1 )<br />
2 (s − µ)T Σ −1 (s − µ) .<br />
This distribution is denoted x ∼ N (µ,Σ). If we define z := Σ − 1/2<br />
(x − µ), then z ∼ N (0, I ),<br />
which is known as the standard Gaussian distribution. If z is scalar, then the cdf <strong>of</strong> z can<br />
be written as<br />
P z (c) = 1 2<br />
( ( c<br />
1 + erf 2<br />
)),<br />
9