Faults in electric machinery such as generators, although rare, still costs hundreds of thousands of dollars for diagnostics, and repairs along with extra costs incurred by the downtime of the machine. This thesis proposes a new way, which is cheaper in terms of hardware cost, and advantageous in ways like live and visual monitoring, live data collection and feature extraction for fault detection, and portability of apparatus. The proposed methodology uses an edge compute device along with software-defined radio to achieve the goal. The thesis includes a discussion on feature extraction from the collected data, which will be used for fault or anomaly detection using machine learning solutions. Signals collected via the neutral ground are used for fault detection, in this approach. The concepts of "partial discharge analysis" and time-series data extraction from the collected data, are focused on primarily.