Arc fault signal detection - Fourier transformation vs. wavelet decomposition techniques using synthesized data

Arc faults are a significant reliability and safety concern for photovoltaic (PV) systems and can cause intermittent operation, system failure, electrical shock hazard, and even fire. Further, arc faults in deployed systems are seemingly random and challenging to faithfully create experimentally in the laboratory, which makes the study of arc fault signature detection difficult. While it may seem trivial to simply record arcing signatures from real-world system, an obstacle in capturing these arc signals is that arc faults in the PV systems do not happen predictably, and depending on the location of the sensors relative to the arc location, may contribute a negligible portion to the magnitude of the sensed current or voltage waveform. The high-frequency content of the arc requires fast sampling, long memory, and fast processing to acquire, store, and analyze the waveforms; this adds substantial balance-of-system cost when considering widespread deployment of arc fault detectors in PV applications. In this paper, we study the performance of the fast Fourier transform arc detection method compared to the wavelet decomposition method by using synthetic waveforms. These waveforms are created by combining measured waveforms of normal background noise from inverters in DC PV arrays along with waveforms of arcing events. Using this technique allows the ratio of amplitudes are varied. Combining these separate waveforms in various amplitude proportions enables creation of test signals for the study of detection algorithm efficacy. It will be shown that the wavelet transformation technique produce more easily recognized detection results and can perform this detection using a much lower sampling rate than what is required for the fast Fourier transform.