A Probabilistic Approach to Series Arc Fault Detection and Identification in DC Microgrids

In this article, series arc fault detection and identification is investigated for dc microgrids using a statistical model based on nodal analysis. The consecutive sample difference of the injection currents are modeled as a random vector whose distribution depends on the network conductance matrix. When a series fault occurs, the conductance matrix changes, which leads to a change in the data generating distribution. The goal is to quickly detect and identify faults on different lines while maintaining low false alarm rates. A quickest change detection (QCD) approach is proposed in this article, utilizing a cumulative sum (CUSUM) algorithm. The proposed method is robust to nominal network operations, such as load and reference changes, and the CUSUM statistic is used for detection increase during faults, ensuring faults are not missed. In addition, a Kron reduction approach is developed to eliminate the internal nodes, and an optimal sensor placement strategy is proposed using vertex cover to ensure fault detection on any line with reduced number of sensors. The proposed framework is tested on dc microgrids typically found in the more electric aircraft, composed of multiple generators, internal nodes, and various load types. Lastly, experimental results are shown on a microgrid testbed to validate the feasibility of the QCD approach for series arc fault detection.