A neural network methodology for check valve diagnostics

The emerging computational tools of artificial neural networks are applied to the area of check valve diagnostics. Recent years have seen increased attention to check valves as a result of several valve failures in safety-related systems in nuclear power plants, demonstrating the need for a non-intrusive method for monitoring check valves. A neural network methodology has been developed for check valve diagnostics using data from a test flow loop. The methodology developed uses an artificial neural network architecture which couples both unsupervised and supervised learning networks in a unified structure. The results of this research demonstrate the ability of the coupled-network methodology to be applied to check valve diagnostics, more specifically the check valve operating condition. This methodology shows improved results over other self-organizing neural networks investigated (competition, self-organizing map, and probabilistic neural networks). The results of this methodology can be improved using various methods to represent the input data.