Winding deformation classification in a power transformer based on the time-frequency image of frequency response analysis using Hilbert-Huang transform and evidence theory

This study presents a novel fully-automated technique to interpret a frequency response analysis (FRA) of a power transformer, using a combined method based on digital image processing and evidence theory. Power transformers are widely used in power systems, and their continuous operation depends on proper monitoring and maintenance. While FRA is considered an efficient method for detecting minor damage in the windings of a power transformer in industrial applications, there is no universally accepted method for interpreting FRA results, and this crucial task relies on the opinion of an error-prone human expert. Our method first obtains a timefrequency image using the Hilbert-Huang transform. Next, the image?s histogram is imported into an evidence theory-based classifier, which ultimately detects radial and axial deformation faults, and their severity. The proposed method is shown to report faults with remarkable accuracy by a finite element model of a three-phase 125 MVA, 230/132/20 kV autotransformer, in which various faults with different severities are simulated and tested.

Automated summary

This study presents a novel fully-automated technique to interpret a frequency response analysis (FRA) of a power transformer, using a combined method based on digital image processing and evidence theory. The proposed method is shown to report faults with remarkable accuracy by a finite element model of a three-phase 125 MVA, 230/132/20 kV autotransformer, in which various faults with different severities are simulated and tested. The method first obtains a time-frequency image using the Hilbert-Huang transform and then imports the image's histogram into an evidence theory-based classifier to detect radial and axial deformation faults.