Partial discharge localization in power transformers using neuro-fuzzy technique

Abstract

Partial discharge (PD) is the most common sources of insulation failure in power transformers. The most important tools for quality assessment of power transformers are PD detection, measurement, and classification. As for the maintenance and repair of transformers, the major importance is the techniques for locating a PD source. The transfer function-based (TF) method for power transformers' winding in the high-frequency range is commonly used in power engineering applications, such as transient analysis, insulation coordination, and in transformer design. Although it is possible to localize PD in transformer winding using the transfer function (TF) method, this method cannot be used for transformers with no design data. Previous attempts toward finding a feature that localizes PD in transformers in general that lineate with PD location were found to be less successful. Therefore, in this paper, a neuro-fuzzy technique that uses unsupervised pattern recognition was proposed to localize PD source in power transformers. The proposed method was tested on a medium-voltage transformer winding in the laboratory. The results showed a significant improvement in localizing PD for major types of PD compared to currently available techniques, such as orthogonal transforms and the calibration line method.