Model-Based Temperature Noise Monitoring Methods for LMFBR Core Anomaly Detection

Temperature noise, measured by thermocouples mounted at each core fuel subassembly, is considered to be the most useful signal for detecting and locating local cooling anomalies in an LMFBR core. However, the core outlet temperature noise contains background noise due to fluctuations in the operating parameters including reactor power. It is therefore necessary to reduce this background noise for highly sensitive anomaly detection by subtracting predictable components from the measured signal. In the present study, both a physical model and an autoregressive model were applied to noise data measured in the experimental fast reactor JOYO. The results indicate that the autoregressive model has a higher precision than the physical model in background noise prediction. Based on these results, an "autoregressive model modification method" is proposed, in which a temporary autoregressive model is generated by interpolation or extrapolation of reference models identified under a small number of different operating conditions. The generated autoregressive model has shown sufficient precision over a wide range of reactor power in applications to artificial noise data produced by an LMFBR noise simulator even when the coolant flow rate was changed to keep a constant power-to-flow ratio.