Estimation of corona location from audible noise

Abstract

Electrical discharges on electrical insulation, though they do not have an immediate effect, could be very hazardous in power systems over time. For that reason, before taking power system apparatus in use, the possible discharges should be identified. In this study, a neural network is configured to determine the exact location of corona discharges by using the sound recordings of audible corona discharge. For the experiment, corona was intentionally produced by placing an x-shaped string on a relatively smooth high-voltage line model. The x-shaped string generates a sharp point on the wire and that results in high stress concentration on this x-shaped string. Obtaining localized high stress on the x-shaped string ensures that a corona starts on the string before anywhere else on the wire and location of the corona can be controlled for data collection. Corona sound samples and corresponding x-shaped string coordinates are applied to a neural network in which sound information forms the input data and coordinates were the output data. Network was configured so that corona location can be estimated by using only 1 s of a recording per discharge. Results show that it is possible to use this method to estimate the location of corona with around 5% error.