General regression artificial neural networks for two-phase flow regime identification

A crucial aspect of nuclear monitoring is the identification of the two-phase flow regimes that occur in heated pipes. A novel efficient, non-invasive, on-line artificial neural network approach to two-phase flow regime identification is put forward; the general regression architecture has been employed. Through the utilization of a single input expressing the mean intensity of each image, satisfactory identification of the flow regime of sequences of images from neutron radiography coolant flow videos is accomplished. The proposed approach is not only more computationally efficient than existing conventional signal processing techniques and computational intelligence methodologies, but also - at worst - comparable to them in terms of identification accuracy.