Abstract

Partial discharges (PD) are generated by anomalies in the electrical insulating systems and are used to obtain information to help establish its health status. PDs generate electromagnetic emissions that can be detected by antennas in the Ultra High Frequency (UHF) range, but are affected by different types of noise, such as Gaussian or non-Gaussian. While techniques like Wavelet Transform (WT), MultiScalePCA (MSPCA), and Variational Mode Decomposition (VMD) have shown effectiveness in noise reduction, choosing the most suitable one for specific real-world conditions remains challenging. Traditional indicators such as the Noise Reduction Ratio (NRR) rely on linear mixing assumptions with Gaussian noise, which may not always apply. This paper introduces Sample Entropy (SE) as a more versatile and comprehensive performance metric, offering enhanced analytical capability across diverse conditions. This work validates SE utility through experiments with real UHF PD signals from two distinct laboratories in varied geographical locations, demonstrating its effectiveness as a superior indicator for technique selection in noise-afflicted environments.