Real time arc fault detection in PV systems using wavelet decomposition

Abstract

Reliable arc fault detection is crucial for the safe operation of photovoltaic (PV) system. Fourier transform methods have been previously used to detect arcing by examining the frequency characteristics of the PV voltage or current but are not well suited because arcs are chaotic, not periodic and not stationary. In contract, wavelet-based transforms are well suited because the technique does not assume periodicity and is adept at detecting discontinuities in the signal. This paper reports on results from the development of a real time arc fault detection technique that was built as a wavelet decomposition based arc detector using a TI C2000 platform DSP. The arc fault detector was tested on a composite arc signal constructed from recordings of real-world inverter noise and real-world arc events replayed through a high-fidelity test bed to compare the ability to differentiate inverter only and inverter plus arcing signals. The results demonstrate that the wavelet decomposition and arc discrimination algorithms can be implemented in real-time on a low-cost DSP.