Studies on the detection of incipient coolant boiling in nuclear reactors using artificial neural networks

Abstract

The sensitivity of coolant boiling monitoring based on the analysis of signals of neutron detectors in a nuclear reactor is studied. Thermal hydraulic processes related to coolant boiling have typical time constant in the order of a few seconds. An efficient coolant-state monitoring system should have a response time comparable with this value of the time constant in order to detect changes at an early stage. The proposed system described in this paper has the required fast response. The proposed monitoring system utilizes advanced signal processing methods based on artificial neural networks in order to achieve early detection of changes in the state of the coolant. The networks have been trained to identify small variations in the power spectral density functions of neutron detector signals. The boiling monitoring method has been tested by using in-core neutron detector signals measured at the NIOBE loop located in the Hoger Onderwijs Reactor (HOR) of Interfaculty Reactor Institute, Delft, The Netherlands. It is shown that boiling detection can be accomplished within about 16 s after the onset of surface boiling in a coolant channel. Results obtained by artificial neural networks have been compared with the efficiency of anomaly detection based on the analysis of band-passed variance of neutronic fluctuations. It is shown that artificial neural nets detect the anomaly faster and more reliably than variance-based statistical methods.