The Partial discharge (PD) phenomene occur in segmented manner in electrical insulation.
Over time, these occurrences can evolve to a critical state, resulting in short circuits and
significant damage to electrical equipment. PD detection and analysis are essential in the
context of preventive maintenance, contributing to high availability rates of Electrical Power
Systems (EPS). However, interference from various noise sources, such as other electrical
equipment and electromagnetic phenomena, makes PD signal detection challenging. This
work presents a comparative study of filtering and signal analysis methods in noninvasive
measurements of partial discharges in real high voltage systems, using different classical
filters and the Wavelet Transform (WT). The research evaluated the effectiveness of several
filtering methods in noise reduction, providing better identification and characterization of PD
signals. Classic filters such as Moving Average (MA), Butterworth (BW), Chebsyshev (Type
I), SavitzkyGolay (SG) and adaptive filtering such as Least Mean Squares (LSM), in addition
to techniques based on multiresolution wavelet decomposition, were implemented and
compared to verify metrics such as Signal-to-Noise Ratio (SNR), Cross-Correlation (CC),
Root Mean Square Error (RMSE), Kurtosis (K), and others, considering the preservation of
the essential characteristics of PD signals. The results obtained demonstrated that filtering
techniques are crucial for reducing noise effects, although more classical methods present
limited efficiency when compared to those with adaptive capacity. The comparative analysis
revealed critical points that, although they demonstrated restricted efficiency, contribute
significantly to the improvement of electrical equipment monitoring methods.
|
