Classification of Transformer Oil Based on Ageing Severity Using Probabilistic Analysis of Partial Discharge Measurements

Characterization of transformer oil is crucial for ensuring long-term reliability in the operation of power transformers. IEEE C.57.106 (2015) and IEEE C.57.637 (2015) classify fresh and in-service transformer oil based on its ageing severity with five mandatory dielectric properties, including AC breakdown voltage, water content, interfacial tension, dissipation factor, and neutralization number. However, this classification strategy allows quite a significant variation in the dielectric properties of oil samples belonging to the same oil class. It also does not directly suggest the exact reconditioning/reclamation process to pursue. Hence, the present work proposes categorizing in-service and fresh oil samples into six oil propositions: A, B, C, D, E, and F, based on the five above-mentioned tests with justified follow-up actions for each oil category. It also proposes a single partial discharge measurement as an alternative to the five tests in order to save time and resources. The prominent features of the phase-resolved partial discharge patterns are extracted to form Z marginal histograms. The oil samples are then classified using the Multinomial Naive Bayes analysis and Weibull probability density function. Finally, the oil classification, with a least error rate of 1.67%, is reported after post-processing with the Weibull analysis at 1.4
$^{\circ }$
phase windows.