THE DIAGNOSTIC POTENTIAL OF PATTERN RECOGNITION METHODS APPLIED TO TEMPERATURE NOISE DATA FROM SIMULATED FAST REACTOR FUEL SUB-ASSEMBLIES
The increased computer power of microelectronics makes the on-line use of noise-based diagnostic techniques feasible for monitoring LMFBR core conditions. The main problem with such early diagnostic methods is that parameter changes generated by incipient fault conditions can often be masked by normal plant variations. Because of this it is likely that decision-making will be based on some form of pattern recognition method, requiring the implementation of sophisticated algorithms. In the absence of extensive operational information under normal and fault conditions, provision of appropriate tests could form a viable means of selecting the best pattern recognition method for a particular application. This study generates a number of different test data sets of fluctuating temperatures which are representative of those which can be measured at the outlets of individual fuel subassemblies, with and without cooling blockages. These tests are then used to compare two different decision- making techniques. The majority of tests are based on the Monte Carlo computer code STATEN which allows the prediction of time- varying temperature signals of the correct statistical form produced by the turbulent mixing of hot and cold sodium flows. A significant part of the study provides a further validation of the code against jet block data from sodium and water experiments to justify its application to modelling turbulent flow and temperature fields in a fast reactor subassembly. The validated code is used to model a CDFR subassembly and to produce signals for a developing cooling blockage under a variety of background conditions. The resulting simulated dataset is augmented by experimental data from a water rig. Two different pattern recognition techniques: Cluster Analysis and the Adaptive Learning Network are evaluated with the common test data. The comparison exercise was carried out in conjunction with the University of Compiegne who have extensive experience in the area of pattern recognition. A special method, comparing the disparate outputs of the two algorithms, is formulated. This enables certain differences in performance to be identified. The study indicates further areas of work which would ultimately facilitate the application of fluctuating temperature monitoring techniques to the identification of cooling anomalies in the subassemblies of large LMFBRs.
Bibliographic Reference: EUROPEAN APPLIED RESEARCH REPORTS - NUCLEAR SCIENCE AND TECHNOLOGY, VOL. 7 (1987), PP. 929-994
Availability: Can be ordered online
Record Number: 1989126037401 / Last updated on: 1989-01-01
Available languages: en