Neural Networks: An Overview - Neural Networks: Methodology and Applications - page 56

Fig. 1.36. The eddy current generation and detection system
contains a “signature” of the defects. Since there are different categories of
defects, which may be more or less detrimental to the operation of the sys-
tem, classifying the defects is generally desirable. In the present case, it is
also important to be able to discriminate between real defects and normal
phenomena that are also detected by the eddy current technique, such as the
presence of a weld joint or of a switch (the position of the latter is known,
which makes discrimination easier).
In the present application, the system that generates and detects eddy
currents is mounted below the carriage, a few tens of millimeters above the
rail, as shown on Fig. 1.36.
As usual, the choice of the descriptors of the signal is crucial for the ef-
ficiency of discrimination. In the present case, a relatively small number of
features, derived from Fourier components of the signal, are usually su cient,
provided they are chosen appropriately. Feature selection was performed by
the “probe feature method,” which is described in Chap. 2 [Oukhellou 1998].
1.4.4 An Application in Forecasting: The Estimation of the
Probability of Election to the French Parliament
After the elections to the French parliament, all candidates must make an of-
ficial statement of the amount of the expenses incurred during the campaign,
and of the breakdown of those expenses. Making use of the data pertaining to
the 1993 elections, it was possible to assess the probability of being elected as
a function of the expenses and of their breakdown. This is a two-class classi-
fication problem, and neural networks provide an estimate of the probability