Series DC arc faults are hard to detect in low-voltage direct current grids because the change in line current is usually too small for classical protection devices. Undetected arcs can overheat conductors and start fires, so dependable detection is essential for future bipolar microgrids. In this thesis, an embedded method that combines the detailed energy from a short Discrete Wavelet Transform with the wide-band energy of a Fast Fourier Transform, then classifies each observation window with a linear support vector machine that runs on a single microcontroller, is developed. Laboratory and field tests confirm that the algorithm detects low-energy series arcs without nuisance trips and operates within the response time required by UL 1699B. The novelty of the work is the mixed wavelet–FFT feature, which captures both local transients and wide-band noise in a compact indicator, making accurate detection possible with modest processing resources.