NDTnet 1998 Aug, Vol.3 No.8

Full-Text - Abstract

In this paper we present the results obtained using a neural network simulator based on the back propagation algorithm in the classifying the artificial flaws. We scanned six axial EDM notches with the radii ranged 0.lmm to 1.5mm. The flaws had same depth -1.5mm and same lengths 12mm and were done on a stub of pressure tube (made from Zr-2.5%Nb) for a CANDU nuclear power plant. The input data used in the learning stage were the radio frequency signal and its Fourier transform obtained on the artificial flaws [1]. In fact was eight defect classes:
• classes 1 - 6 artificial flaws with the following tip radius: 0.1mm, 0.25mm, 0.6mm, 0.8mm, lmm and 1.5mm. From learning and testing was used the signal with the maximum amplitude from each flaw.
• class 7 "flawless". We have considered the signal for this class froma free flaw zone
• class 8 "backwall echo" . The signal for this class was a backwall echo
The results obtained at the testing of neural network have shown that the probability of classification for defects with tip radius bigger than ultrasonic wavelength (0.6 - 1.5 mm) was the best.
The same neural network simulator was used in the classification of the defects located by C-scan in the brazed joints of CANDU nuclear fuel. We have used the four classes of defect corresponding the flaws from block calibration:
• C1 :"the axial circular flaw "; In this case the signal used was obtained from a cylindrical flaw with the axis along the brazed joint . The diameter of this flaw is 0.5mm
• C2: "the transversal rectangular flaw"; We was used, for the learning process, the signal obtained from the artificial rectangular flaw of the 0.2mm width
• C3: "the longitudinal hole"; cylindrical flaw with the axisperpendicular on the axis of the brazed joint. The diameter for this flaw was the same ( i.e. 0.5mm)
• C4: "the flawless indication";
The testing was performed using the other signals obtained from each type of defect. In addition we have used a signal obtained from a natural defect located by a C-Scan image of the brazed joint. The presence of defect was confirmed by methalography. The neural network simulator has classified this defect with a 87% probability in class C1 and with a 18% probability in class C3.

Reference:
1. M. Kitahara, J. D. Achenbach, Q. C. Guo, M. L. Peterson, Review of Progress inQuantitative Nondestructive Evaluation, Vol. 11, pag. 701

Abstract Source:
Book of Abstracts, 7th European Conference on Non-Destructive Testing, 26-29 May 1998, ISBN: 87-986898-0-00

Full-Text Source:
Proceedings of the 7th European Conference on Non-Destructive Testing, 26-29 May 1998, ISBN:

Publication Contact:
7th ECNDT, Park Alle 345, DK-2605 Broendby, Denmark, Fax: +45 46 26 70 11, Email: 101373.3414@compuserve.com