Fault Detection for Naval Pulsed-Energy Mission Loads Using a Novel Machine Learning Approach

摘要

Next generation warships require energy dense distributions to power advanced weapon and sensor loads on the horizon. Medium-voltage dc distribution is well-suited to provide this requirement if certain reservations are addressed. A particularly concerning issue is that these advanced loads draw large currents in short periods of time similar to fault behavior; and may be indiscernible from a fault. This paper presents a novel machine learning based algorithm to detect faults applied to a notional pulsed-energy mission load exhibiting characteristics of these advanced loads. Specifically, time-scale features of the load current are extracted under a wavelet transform before being processed in a recurrent neural network which produces both a classification of the pulsed-power supply firing phase and a signal reconstruction based on a known operating profile. Simulation of the proposed algorithms showed 99.8% accuracy for waveform classification and 100% accuracy for fault detection.