Valve health monitoring with wavelet transformation and neural networks (WT-NN)

Servovalves are one of the most important components of the complex machinery of space exploration. They have to be at the perfect condition for safe and efficient operation of very valuable complex machines. In this paper, use of wavelet transformation (WT) and adaptive resonance theory 2 (ART2) type self learning neural network (NN) combination is proposed for detection of defective valves. The current signature of the energization stage of the valve was encoded by using the WT. ART2 classified the approximation coefficients of the WT. WT-NN classified all the normal valve data in single category and assigned new categories to the data of defective valves as long as the vigilance was selected properly. WT-NN combination was found an effective alternative to customized diagnostic software if the operating conditions change drastically