On-Line Detection of Dc Arc Faults Using Hurst Exponents for Hybrid-Electric Vehicles

Series arc faults are a severe safety hazard in modern pure electric and hybrid electric vehicles, as well as dc electrical distribution systems. A new method utilizing Hurst Exponents was recently proposed that is strongly resistant to switching harmonics, load profiles, and high frequency noise. In this paper, further analysis and implementation of the Hurst based detection is performed. A method of filtering out signal noise is presented, and an on-line operation prototype is constructed using common hardware. The on-line prototype is demonstrated to successfully detect series arc faults in under 50 ms. A statistical analysis of the method effectiveness is performed to determine the success rate of the on-line prototype, which is expressed through the metrics of Precision and Recall.