Partial Discharges: Experimental Investigation, Model Development, and Data Analytics

Insulation system is an inseparable part of electrical equipment. In this study, one of the most important aging factors in insulation systems known as partial discharge (PD) is targeted. PD phenomenon has been studied for more than a century and yet new technologies still demand the investigation of PD impact. Nowadays, electrification is penetrating into various fossil-fuel-based industries such as transportation system that demands the reliability of electrical equipment under various harsh environmental conditions. Due to the lack of knowledge on the behavior of insulation systems, research in this area is intensively needed. The current study probes into the partial discharge phenomenon from two aspects and the groundwork for both aspects are provided by experimentation of multiple PD types. In the first goal, a finite-element analysis (FEA) approach is developed based on measurement data to estimate electric field distribution. The FEA model is coupled with a programming scheme to evaluate PD conditions, calculate PD metrics, and perform statistical analysis of the results. For the second target, it is aimed to use deep neural networks to identify and discriminate different sources of PD. The measurement data are used to generate thousands of phase-resolved PD (PRPD) images that will be used for training deep learning models. To meet the characteristics of the dataset, a deep residual neural network is designed and optimized to discriminate PD sources in an accurate, stable, and time-efficient way. The outcome of this research enhances the reliability of electrical apparatus through a better understanding of the PD behavior and lays a foundation for automatic monitoring of PD sources.