Analysis of Defectoscopy Data to Be Used by Neural Classifier
At present a very perspective solution of indications classification in defectoscopy is neural network application. One of the fields is classification of indications into classes that are characterized by the signal shape, or by the signatures relating to the signal shape. Nondestructive defectoscopy of steam generator tubes of nuclear power plants by multifrequency eddy current method is the field in which the use of classifiers based on neural network architecture is very perspective.
The contribution concentrates on the choice of a suitable representation of indications for neural classifier represented by probabilistic neural network. Selected representations are compared using real records of steam generator tubes and also using artificial defects and imitations of construction elements.
KeywordsProbabilistic Neural Network Signal Shape Neural Network Architecture Destructive Testing Artificial Defect
Unable to display preview. Download preview PDF.
- C. Rajagopalan, Baldev Raj, P. Kalyanasundaram, “The Role of Articial Intelligence in Nondestructive Testing and Evaluation”, INSIGHT, Vol 38, No 2, February 1996.Google Scholar
- C. Charlton, “Investigation into the Suitability of a Neural Network Classifier for use in an Automated Tube Inspection System”, British Journal of NDT, vol..35, No 8, August 1993.Google Scholar
- Fu. LiMin, “Neural Networks in Computer Inteligence”, McGraw-Hill Companies, March 1994.Google Scholar
- Ch.T. Yahn, R.Z. Roskies, “Fourier Descriptors for Plane Closed Curves”, IEEE Transactions on computers, March 1972Google Scholar
- R. Palanisamy, W. Loyd, “Finite Element Simulation of Support Plate and Tube Defect Eddy Curre nt Signals in Steam Generator NDT”, Materials Evaluation, Vol 39, June 1981.Google Scholar
- J. Grman, “Neural Network Application in the Defectoscopy”, IMEKO 2000 — XVI IMEKO World Congress, Wienna, Austria, 25.–28.September 2000.Google Scholar