Autoregressive model-based boiling detection in a Liquid Metal Fast Breeder Reactor

Abstract

This paper presents a new approach for acoustic detection of sodium boiling in a Liquid Metal Fast Breeder Reactor (LMFBR) based on Autoregressive (AR) models. The AR models are estimated on a sliding window and classified into boiling or non-boiling models by comparing the on-line estimated values of their components to the predictions of their components from the environment parameters using linear regression. In order to avoid false alarms the proposed approach takes into account operating mode information. Promising results are obtained on the background noise data collected from the French Phenix nuclear power plant provided by the French Commission of Atomic and Alternative Energies (CEA).