A pattern recognition-artificial neural networks based model for signal validation in nuclear power plants

The OECD Halden Reactor Project has been working for several years with computer based systems for plant state identification and fault detection. This work describes a model for Signal Validation which utilizes Pattern Recognition and Artificial Neural Networks techniques. The approach is based on the use of a pattern recognition algorithm (ISODATA) which drives nine specialized supervised networks, each network being trained to provide accurate outputs only in a limited and restricted plant operational region. Results are shown for the application of this technique to a Boiling Water Reactor during some abnormal conditions, simulating multiple sensor failures in critical process signals.