Fig. 6-3 Monitoring display Measured (on-line) signals (white) coming from the reactor and estimated values by NN (green) are displayed. The color of the estimated value changes to red and the time-series graph appears when the monitoring system detects an anomaly.

Fig. 6-4 Monitoring result (Atmospheric dump valve leak at 9 s) Monitoring system with neural network detects the anomaly 3 seconds after the valve leak occurs. The conventional alarm system needs 30 seconds to announce "Rod Stop" alarm.

Monitoring the condition of a nuclear reactor is a major concern during operation because of operational safety. An on-line nuclear power plant monitoring system has been developed using the neural network (NN) concept. The dynamics model of a reactor plant is constructed on a three-layered autoassociative neural network by using a back propagation learning algorithm. The basic idea of this anomaly detection method is to monitor the deviation between process signals measured from the actual plant and corresponding output signals from the NN plant model. The monitoring system has successfully detected various anomalies in the early stage of malfunctions in an actual plant. The integration to an expert system accumulating knowledge of anomaly conditions will be a promising way to realize the desirable coordination of the human-machine interface. The final goal is to detect all anomalies in an early stage and to support the operators.

Reference K. Nabeshima et al., On-line Nuclear Power Plant Monitoring with Neural Network, Proc. 3rd. Int. Conf. on Nuclear Engineering, 3, 1551 (1995).

Select a topic in left column