Traditional signal detection methods require an explicit prescription of partial discharge features that characterize the state of the cable system. When the sample size is sufficiently large, deep learning models allow complex interrelations of auto-generated features. A distinct challenge is the characterization of the waveform signal, which depends on cable length. In this webinar, an overview of the developed deep learning models for an extensive partial discharge dataset to automate the detection of underground cable defects will be provided. The developed deep learning models outperform predictions with traditional methods. In addition to classifying partial discharge signals, the models identify source locations of the defects within a cable system through recurrent Neural Networks. Additional assessments include advanced data augmentation strategies and interpretability to verify the potential use of the model for predictive maintenance. Co-sponsored by: Power and Energy Systems Research Laboratory Speaker(s): Steffen Ziegler Virtual: https://events.vtools.ieee.org/m/387548