Neural Networks in Process Diagnostics

Diagnostics and Control are critical to any continuous process set up. It is not only necessary that proper diagnostic conclusion be drawn but also with accuracy and speed. Artificial Neural Networks, which are biologically inspired are accurate, elegant and fault tolerant. In the contexts of known scenarios, they are faster in comparison with the case-based diagnostics systems. In this paper the authors propose a two layer, feed-forward Neural Network with back propagation as a modeling paradigm for continuous process diagnostics. This research is pursued in the domain of nuclear reactor control. Both the training and operational phases are described.