PD detection and recognition based on UHF method for typical models in air

The UHF method can not only be used in gas insulation substation-GIS and transformer, but also can be used to detect PD happening in air or other insulation dielectric. On account of the three kinds of PD happening in dielectric, out of dielectric and surface of dielectric respectively, four typical...

Full description

Saved in:
Bibliographic Details
Main Authors: Duan, Da-Peng, Wang, Peng, Han, Liang, Lu, Yu-Hang, Li, Wei
Format: Conference Proceeding
Language:English
Subjects:
Artificial Neural Network
Artificial neural networks
Atmospheric modeling
Classifiers
Dielectrics
Feature extraction
Feature recognition
Gas insulation
Insulation
Kernel Principal Component Analysis
Mathematical models
Partial Discharge
Partial discharges
Pattern Recognition
PD Models
Recognition
Sampling
UHF
UHF measurement
Wavelet Transforms
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The UHF method can not only be used in gas insulation substation-GIS and transformer, but also can be used to detect PD happening in air or other insulation dielectric. On account of the three kinds of PD happening in dielectric, out of dielectric and surface of dielectric respectively, four typical PD models were designed and manufactured with aluminium, namely, air gap, air void in epoxy, needle-flat and surface of epoxy, by which the PD experiments were carried out in air. The UHF PD signals were detected by the developed twin-spiral UHF coupler and amplifier with band of 100MHz-1500MHz. For every model, 50 UHF PD records were stored by digital storage oscilloscope with sampling rate 20GHz. After denoising with the improved wavelet adaptive threshold method, and envelope extraction, 19 features parameters were extracted in time domain and frequency domain to recognize the PD source type. Finally, a classifier was designed with 3-layers back propagation artificial neural networks, and the test results show that the extracted features and classifier are sufficient to recognize the PD source correctly with high robustness.
DOI:10.1109/ISEIM.2011.6826381