Optimal Threshold Selection and Efficacy Evaluation for a Generic dc Series Arc Detection Algorithm

Dc power systems are becoming prevalent in areas such as aerospace, photovoltaic arrays, data centers, and automotive vehicles. As these systems strive for more compact and efficient operation, their operating voltages increase which in turn increases the hazard of arcing. Being able to quickly detect and handle arcs is important to prevent insulation damage, fire, or other system damage. Arcs fall into one of two categories: low impedance arcs and high impedance arcs. Low impedance arcs occur in parallel with a load and are easier to detect through the large overcurrent produced. High impedance arcs occur in series with a load when an energized conductor breaks or connectors loosen. These are harder to detect as they cause a drop in load current which may appear as normal system operation. Many high impedance arc detection algorithms exist in literature and differentiate between arcing and non-arcing states through a threshold or `tripping ' level. Each algorithm uses a threshold to show functionality, but often fails to describe its selection process. This paper details an optimal threshold selection process for a generic running window algorithm. This threshold is determined by an a-posteriori analysis of algorithm output given a known test input. Empirical cumulative distribution curves are generated for the algorithm output in the arcing and non-arcing states. Using these curves, an optimization method is formed for choosing the threshold with respect to Type I and Type II error rates. Once a threshold is chosen, a predictive measure of performance is known by considering the a-posteriori error rates. As an example, the threshold selection method is compared between two common detection algorithms using experimental data.