A Frequency Domain analysis of Partial Discharge defects Patterns using Neural Network Model under Cryogenic Temperature

Abstract

The superconducting equipment’s are used for enhancing the transmission capacity of the traditional electric power equipment’s. This is due to many technical advantages which are mostly related to its reduced size, weight, high efficiency and many more. Keeping in view of this fact, many researchers are carrying out experimental investigations to improve its performance which are mostly related to their reliability, when used under cryogenic conditions. One of the most important parameter which is used to describe the insulation status of the apparatus is Partial discharge (PD) detection. However, not many methods had been proposed for the results observed in cryogenic conditions. In current work, three different artificial defects, including protrusion, floating electrode, and turn to turn defects are put into Liquid nitrogen to create cryogenic conditions.  To produce PD, AC voltage is applied to the apparatus. Pattern recognition of the PD signal which are detected based on CAPD method (Chaos analysis of PD). In order to improve the rate of PD pattern recognition which is able to classify different type of insulation defects introducible into the superconducting power apparatus, training data is added with the frequency spectrum data. Afterwards, the changes in Mean Square Error (MSE) and PD pattern recognition rate are compared by using the only CAPD data and CAPD data combined with frequency spectrum data.