DC Series Arc Failure Diagnosis Using Artificial Machine Learning With Switching Frequency Component Elimination Technique

The intricate spectrum of arc faults elicited by diverse load types introduces a complex and formidable challenge in residential series arc fault detection. Series DC arc faults pose a significant concern as they can potentially instigate fire incidents and exert adverse ramifications on power systems if left undetected. Nonetheless, their detection within practical power systems remains challenging, predominantly attributed to the meager arc current magnitude, the absence of a discernible zero-crossing interval, and the manifestation of multifarious aberrant behaviors contingent upon the diverse array of power loads and controllers. Importantly, the conventional safeguards, notably encompassing protection fuses, may exhibit inefficacy in promptly activating during the occurrence of series DC arc faults. The ramifications of undiscerned arc faults are profound, with the potential for engendering erroneous operational modes within power systems, thereby amplifying the risk of material and human casualties. In light of these exigencies, the development of an efficacious detection mechanism targeting series arc faults within DC systems becomes a paramount imperative. This research proposed a preprocessing signal to eliminate the switching noises, which could degrade the performance of artificial machine learning algorithms. The diagnosis results valid the effectiveness of the proposed diagnosis scheme for all ranges of switching frequencies.