The application of neural network for the advancement of the eddy current testing
- Central Research Inst. of Electric Power Industry, Tokyo (Japan)
All the steam generator (SG) tubes of Japanese pressurized water reactors (PWRs) are inspected by the eddy current testing (ECT) method in every annual scheduled inspection. Here, a neural network system to estimate the class and size of defects from signals obtained by the eddy current testing (ECT) method has been developed. A trajectory of ECT signal is characterized by four representative parameters, and totally eight parameters obtained from two trajectories by different AC current frequencies are used as input parameters for neutral networks. A probabilistic descent method is employed to minimize the error at the learning process of neural networks. It is indicated that using multiple neutral networks which are separately responsible to each class of defects is effective to the improvement of their estimation accuracy. And, it is demonstrated that the neural network system which the authors developed can estimate the class and size of defects from unlearned trajectories with high accuracy.
- OSTI ID:
- 418167
- Report Number(s):
- CONF-960706--; ISBN 0-7918-1769-5
- Country of Publication:
- United States
- Language:
- English
Similar Records
Likelihood of mid-cycle shutdown due to steam generator tube leak
Hybrid digital signal processing and neural networks for automated diagnostics using NDE methods