B. Benoist et al., EXPERT-SYSTEM FOR THE CHARACTERIZATION OF DEFECT SIGNALS IN STEAM-GENERATOR TUBES, Engineering applications of artificial intelligence, 8(3), 1995, pp. 309-318
The automatic analysis, by computer, of defect signals in steam genera
tor tubes, based on the eddy current multifrequency technique, is freq
uently inefficient due to ''pilgrim noise''. The first step is to use
a method that allows the noise to be eliminated: adaptive interpolatio
n. Thanks to this method, which ensures reliable data on each channel,
the analysis can be realised by taking account of the data correspond
ing to each basic or mixed channel. By correlating these diverse data,
the signals can be grouped according to two types of defects: single
defects, and multiple defects (several at the same place). The second
step is to use an expert system which allows a reliable diagnosis for
whatever family the defect belongs to. According to this classificatio
n, analysis is continued and results in the characterization of the de
fect. The expert system has already been developed using the general-p
urpose application expert-system shell SUPER, which is briefly describ
ed. The knowledge base (SOCRATE) is described in detail. The first res
ults obtained with signals corresponding to real defects, that have be
en recorded in different places, are presented and discussed. The expe
rt system is shown to be efficient in all the cases, even with signals
obtained in very noisy environments.