A new method for nuclear plant diagnostics using neural networks

This paper addresses nuclear power plant fault diagnosis using an artificial neural newtwork. In a previous work, single time snapshots of a number of plant variables were used for training the diagnostic network. In this work, however, a moving average of a single plant variable is used in an attempt to reduce classification time. A moving average of the average power range monitor (APRM) flux level is used to build the learning set on which a backpropagation neural network is trained. Preliminary efforts to classify three transients by monitoring this variable are presented.