The application of the resilent backpropagation algorithm and power spectrum density for recognizing the acoustic emission signals genereted by basic partial discharge forms using artificial neuron networks

The subject matter of this paper refers to the correct recognition of the acoustic emission
(AE) signals generated by basic partial discharge forms (PDs). The paper presents research
results of the application of unidirectional artificial neural networks (ANN) for recognizing basic
PD forms that can occur in paper-oil insulation impaired by aging processes. The research
work results present recognition effectiveness of basic PD forms depending on the number
of basic forms passed simultaneously onto the network inputs and the size of the teaching
sequence. Power spectrum density was assumed as the parameter of the AE signal generated
by the assumed PD forms. The paper also presents the results of the network effectiveness
analysis depending on the number of the points averaging the power spectrum density, the
number of neurons of the concealed layer and the size of the teaching sequence.