Fault supervision of nuclear research reactor systems using artificial neural networks: A review with results

On-line condition supervision of nuclear reactor (NR) is of major concern and high-priority task during operation to ensure safe operation of systems. Usually, faults can occur in different locations at any time. The task of supervision is to monitor the normal operation, and in the case of faults, it has to detect, diagnose them earlier and take appropriate decision, as correction, before they provoke damage to the plant. The actual paper presents the current state of research in the monitoring and decision-making field. It gives a review of fault supervision applications in dynamic, non-linear, complex, and sometimes not well known systems, such as those of NRs; and the important used methods, particularly neural networks (NNs). The aim of this paper is the proposal of a fault supervision application based on NNs of some nuclear and thermo-hydraulic critical variables in Applied Nuclear Energy Laboratory (LENA) NR systems such as the core and hydraulic circuits. The test results show that NN approach estimates successfully the selected variables with early detection, identification and accommodation of faults in real-time, over a wide power range of operation including start-up, shutdown and steady state. Finally, the application of the developed model may be extended to other variables and different NR systems.