Study of Partial Discharge Audio Recordings Towards Identifying Features For Classification

Episodes of Partial Discharge (PD) induce changes to the conductors or insulation that may degrade electrical performance and lead to larger faults and ultimate failure of assets or network outages. Several ways have been used in developing different PD-inspection techniques. The focus of this work is on acoustic (audio, ultrasonic) PD detection. In this work, we have used the Discrete Fourier Transform (DFT) to analyze a samples of real-world audio recordings of PD to identify their types as they may be classified by their unique signatures in both the time domain and frequency domain. Through the analysis of the field data, this project explored the signatures of each type of PD (tracking, arcing and corona) and how they can be distinguished from background noise. The result of this work will be used for further algorithm development for detection of episodes of PD. Having a system in place to autonomously detect PD would help to direct maintenance toward prevention of major faults. This would lead to large cost-reduction and improved quality of service.