Fast Reactors and Accelerator Driven Systems Knowledge Base

Conference Article: Detection of argon, hydrogen and water injections in a FBR steam generator

Abstract

The analysis of the IAEA 1995 benchmark data on acoustic leak detection by IRI, The Netherlands, focused on the timely detection of the onset of leak. The aim was to design a leak detection algorithm that can, in principle, be applied for on-line application and will give a fast and reliable detection of any anomalies in the signal with a pre-specified false alarm rate. To meet this goal, long and complicated computations which take long data series have to be avoided. Therefore, anomaly detection in the time domain seems most appropriate. To eliminate all correlation from a measured signal, an autoregressive (AR) model was established based on the first 2048 signal values of the signal, which part is supposed not to contain any anomalous data. Applying the AR-model as a one-step prediction filter, a residual noise signal is obtained which should be uncorrelated (white noise) if the proper AR-model is applied. A fixed model order of 30 was chosen on basis of the reduction of the variance of the residual noise as a function of model order. The Burg algorithm was used to determine the AR-coefficient, while the Yule-Walker methods was not found suitable for this case.

view the full text of this article (12 pages, format: PDF, size= 577kB)

Reference:

Acoustic signal processing for the detection of sodium boiling or sodium-water reaction in LMFRs. Final report of a co-ordinated research programme 1990-1995

International Atomic Energy Agency, Vienna (Austria)

IAEA-TECDOC--946, pp:227-238