The status of application of neural networks in the nuclear industry

Artificial neural networks (ANN) are now accepted as a very practical Al technique, used for many applications in various fields. The nuclear industry, although aware of the potential benefits of the application of ANN, is slow in accepting this technique in nuclear power plants (NPP) control rooms, even for operator advisory roles. However, the technique is now being considered for predictive maintenance systems of NPP components, and for support roles in non-safety related systems, such as load forecasting, water chemistry, fuel management and safeguards systems. Active research on the properties and possible applications of ANN in NPP's is carried out in universities and national laboratories in several countries. Training data is supplied by the nuclear industry, either as real NPP operating data or from full scale simulators. The ability to train large scale ANN is crucial to their successful implementation in real-life situations, enabling also the development of auto-associative ANN for real-time fault diagnostics. In a demonstration example of a simulated material transfer system with 50 components, an auto-associative ANN was able to sense immediately the presence of a fault during an operation, and diagnose correctly the faulty component. (author). 101 refs