Arc fault is a significant reliability and safety concern for photovoltaic (PV) direct current (DC) electrical systems. Failure to detect an arc fault can lead to system failure, property loss, or even bodily injury. Of the numerous methods that have been proposed, the most common fast fourier transform (FFT) has significant fundamental limitations that stem from the fact that arcs do not conform to the inherent assumptions upon which the FFT is mathematically based. This can lead to the non-detection of arcing events, or the false-detection which is potentially just as bad as it leads to a reduced confidence in the technology and a better technology solution is needed. The goal of this research is to design and demonstrate a real-time, embedded system implementation of a new technique-wavelet transform (WT) for arc fault detection. In order to analyze arc of various power level under the intrinsic high frequency noise feature of PV systems, arc signals are synthesized with different Arc-Signal-Noise-Ratio (ASNR) and a high fidelity testbed is constructed for the reproduction of arc signals. With threshold setting simulated on MATLAB, WT detection technique using ratio of the power increment is realized on the commercial arc detection product RD-195. Real-time arc fault detection is conducted for both synthesized arc signal and real PV arc signal, and the detection result of the WT algorithm is compared against the FFT algorithm. From the experiment and analysis, the rationality of WT based detection technique for DC arc detection application is validated and the superiority compared to FFT technique is demonstrated.
