Abstract

The application of neural networks as a tool for reactor diagnostics is examined here. Reactor pump signals utilized in a wear-out monitoring system developed for early detection of the degradation of a pump shaft [17] are analyzed as a semi-benchmark test to study the feasibility of neural networks for monitoring and surveillance in nuclear reactors. The Adaptive Resonance Theory (ART 2 and ART 2-A) paradigm of neural networks is applied in this study. The signals are collected signals as well as generated signals simulating the wear progress. The wear-out monitoring system applies noise analysis techniques and is capable of distinguishing these signals apart and providing a measure of the progress of the degradation. This paper presents the results of the analysis of these data and provides an evaluation on the performance of ART 2-A and ART 2 for reactor signal analysis. The selection of ART 2 is due to its desired design principles such as unsupervised learning, stability-plasticity, search-direct access, and the match-reset tradeoffs. ART 2-A is selected for its speed. Two simulators are built. One is ART 2, and the other ART 2-A. The result is a success for both paradigms, and the study shows that ART 2-A is not only able to learn and distinguish the patterns from each other, its learning speed is also extremely fast despite the high-dimensional input spaces. © 1992 IEEE

Department(s)

Nuclear Engineering and Radiation Science

International Standard Serial Number (ISSN)

1558-1578; 0018-9499

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2023 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 1992

Download

Full Text Link

Share

 
COinS