Determination of the level of degradation of generator stator bar insulation using a onedimensional convolutional neural network

Abstract

This paper presents the results of an experiment to classify the levels of insulation degradation of generator stator bars using a one-dimensional convolutional neural network. The stator bars were subjected to increased electrical stress during the time until insulation breakdown. The bars were periodically injected with a specially designed reference signal during the stress application to generate a dataset for training the neural network. The injected signal was acquired and then subjected to pre-processing. The paper evaluates each pre-processing variant in terms of its effect on the performance of the classification algorithm. It provides the neural network structure and its optimal parameters to accomplish the task of determining the insulation degradation state.