Online Partial Discharge Detection & Localization of a Transformer Winding Using Machine Learning

Power transformers are critical and costly components in power systems. The primary cause of severe failures in power transformers is insulation breakdown and partial discharge (PD) is a significant contributor to insulation failure due to gradual erosion. Therefore, accurate detection and localization of PD are essential for effective condition monitoring. This research introduces a robust approach for detecting and pinpointing PD using machine learning, eliminating the need for human intervention and dismantling the transformer, thereby enhancing the reliability of power systems. Using a lumped parameter RLC model in PSCAD, PD is simulated by injecting current impulses at various winding locations. Terminal current waveforms are analyzed to extract features for machine learning training. The study also focuses on identifying multiple winding locations of the transformer in cases where partial discharge might be found in multiple locations. The study indicates that statistical features are most effective in accurately representing PD, with Support Vector Machine exhibiting higher accuracy compared to other machine learning methods for partial discharge monitoring.