To characterize flaws caused by intergranular attack (IGA) in steam generator tubes, the Babcock & Wilcox (B&W) Owners’ Group sponsored a program to study different techniques to determine the depth of flaws based on eddy current data obtained from Bobbin coil probes. Techniques based on multiple regression analysis, neural network regression, and artificial intelligence have been reported elsewhere. This report summarizes the results based on application of a probabilistic neural network (PNN) to classify the flaws into groups of different depths. As a starter, only two classes were selected: Class 1 contained flaws with actual maximum depths below 40% through-wall; Class 2 contained flaws with depths above 40% through-wall. Classification was based on the Bobbin coil peak voltage and the corresponding phase angle data at the three different excitation frequencies of 600, 400, and 200 kHz. The results were compared with results from destructive examination. The study showed that best results were obtained when only the peak voltages were used to train the neural network. Based on this approach, the network classified the flaws correctly 72% of the time. Two methods of improving the network performance are proposed.

1.
“PWR Steam Generator Examination Guidelines: Rev. 5. Vol. 1,” EPRI Report TR-107569-V1R5, 1997.
2.
Dodd, C. V., Deeds, W. E., and McClung, R. W., 1990, “Improved Eddy-Current Inspection for Steam Generator Tubing,” U.S. NRC Report NUREG/CR-5478.
3.
Popa
,
R. C.
, and
Miya
,
D.
,
1998
, “
Approximate Inverse Mapping in ECT Based on Aperture Shifting and Neural Network Regression
,”
Nucl. Eng. Des.
,
17
, pp.
209
221
.
4.
Upadhyaya
,
B. R.
, et al.
1999
, “
Development of a Diagnostic Expert System for Eddy Current Data Analysis using Applied Artificial Intelligence Methods
,”
Nucl. Eng. Des.
,
193
, pp.
1
11
.
5.
Masters, T., 1995, Advanced Algorithms for Neural Networks, Wiley, New York, NY.
You do not currently have access to this content.