30.06.2013 Views

smart technologies for safety engineering

smart technologies for safety engineering

smart technologies for safety engineering

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

VDM-Based Health Monitoring 95<br />

locations, which supposedly contains all unknown locations of defects. In the most general<br />

case, the set of distortion locations includes all resistive elements of the circuit.<br />

3.5.4.1 Defect Identification in the Steady State<br />

In the steady-state cases, the problem can be solved by trans<strong>for</strong>ming Equation (108), in which<br />

modeled responses are substituted by the responses from reference points:<br />

D f ε 0 = f ref − f L<br />

(115)<br />

Obviously, to obtain a unique solution, the influence matrix D f , assembled <strong>for</strong> the selected<br />

reference points and distortion locations, needs to be square and nonsingular. This means that<br />

the number of independent reference responses cannot be lower than the assumed number<br />

of distortion locations. The mutual independence of responses is determined by distortion<br />

locations and Kirchhoff’s laws. In a circuit consisting of n elements and K nodes, there are at<br />

most (K-1) independent current Kirchhoff’s laws and (n − K + 1) voltage Kirchhoff’s laws.<br />

When all resistive elements are considered as possible defect locations, in order to obtain a<br />

nonsingular influence matrix, the set of reference points must include a current reference in<br />

every independent loop of the circuit and voltage reference in all but one node. With every<br />

element excluded from the set of distortion locations, the number of unknowns is reduced and<br />

one reference response becomes redundant.<br />

An alternative approach uses iterative methods based on gradient optimization. Although<br />

more computationally demanding, they enable constraints to be imposed. The evolution of<br />

optimized parameters can be actively controlled, ensuring more reliable results (when reference<br />

responses are perturbed by sloppy measurement or noise). In some cases a reduction in the<br />

number of necessary reference responses is possible. The proposed procedure, valid <strong>for</strong> both<br />

the DC and AC cases, is based on the steepest-descent method, with the objective function<br />

defined as the least square problem and distortions chosen as optimization variables.<br />

The vector of distance functions d describes differences between the modeled responses<br />

(defined by the actual state of distortions) and responses obtained from the reference points:<br />

d = f(ε 0 ) − f ref<br />

(116)<br />

Components of the gradient of distance functions with respect to distortions (complex derivatives<br />

in the AC case) are equal to entries of the influence matrix:<br />

The objective function g is defined as<br />

∇d = ∂di<br />

∂ε 0 j<br />

= D f<br />

ij<br />

(117)<br />

g = didi = (d) H d (118)<br />

where (·) H denotes the conjugate transpose. Function g is a real-valued function of real or<br />

complex arguments. With respect to the objective function, the direction of steepest descent h<br />

is defined as<br />

h = 2[∇d] H d (119)

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

Saved successfully!

Ooh no, something went wrong!