Analysis of PD-generated UHF Signals and their Classification using PRPD and Machine Learning
Fragment książki (Rozdział monografii pokonferencyjnej)
MNiSW
20
Poziom I
| Status: | |
| Autorzy: | Zmarzły Oskar, Kozioł Michał, Wotzka Daria, Boczar Tomasz, Kołtunowicz Tomasz, Kunicki Michał |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Strony: | 1 - 5 |
| Web of Science® Times Cited: | 0 |
| Bazy: | Web of Science | IEEE Xplore |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | TAK |
| Nazwa konferencji: | 26th International Scientific Conference on Electric Power Engineering - EPE 2026 |
| Skrócona nazwa konferencji: | 26th EPE 2026 |
| URL serii konferencji: | LINK |
| Termin konferencji: | 18 maja 2026 do 20 maja 2026 |
| Miasto konferencji: | Opole |
| Państwo konferencji: | POLSKA |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| Partial discharges (PD) are localized electrical breakdown events in dielectric media, typically initiated by elevated local electric fields, voids or surface irregularities, and can be replicated in laboratory conditions using controlled electrode gaps. In this work, oil-immersed PD activity was generated for five representative electrode geometries (needle–sphere, sphere–plate, sphere–sphere, plate–sphere, plate–plate), serving as analogues of defect scenarios relevant to power transformer insulation systems. UHF-based measurements were conducted using a MPD600 system and phase-resolved partial discharge (PRPD) patterns were extracted from the recorded signals. PD fingerprints were observed, motivating the formulation of a data-driven classification concept. A set of statistical and phase-resolved descriptors is proposed to parametrize the PRPD data and to enable subsequent machine learning–based recognition of PD mechanisms using only recorded PD signals. The study demonstrates that controlled PD measurements can be mapped to transformer-relevant defect mechanisms and provide a viable basis for automatic PD classification frameworks, with potential applications in transformer condition assessment and HV diagnostics. |
