Experimental Characterization of Series Arc Faults in DC PV Systems

In photovoltaic (PV) systems, arc faults are dangerous events that may lead to fire ignition if not promptly detected and de-energized. They are due to various reasons, such as damaged wires and connectors, worn electrical insulation, components aging, overheat or stress. The detection of arc faults poses several issues, because of the intrinsic random nature of the arc fault phenomenon; furthermore, several factors (PV system topology, operating conditions, inverter noise and distortion and so on) can modify the arc signal profile and characteristics; thus the arc signal can be filtered, masked or attenuated and the arcing condition may go undetected or a normal operating condition can be mistaken for an arcing one. This is more critical for series arcs, as they occur because of electrical continuity failures (of conductors, connections, modules or other PV system components), thus the current amount is limited by the load of the PV system components themselves.Several methodologies and solutions have been studied and patented for AC arc fault detection, and the research on DC arc faults is ongoing too, concerning both DC arc fault modeling and detection methods; however a unique and complete solution, able to correctly operate in all working condition is not yet available. In very brief, most methods are based on current signal analysis; one of the most addressed signal characteristic is the broadband noise and its analysis in specified frequency bands, typically from tens of kilohertz up to 100 kHz. However, the possibility of exploiting low frequency analysis can allow using lower sampling frequencies, thus enabling the use of commercial acquisition and signal processing systems (as those typically used for smart metering purposes). Furthermore, a suitable choice of sampling frequency and number of acquired samples can allow reaching a good tradeoff between spectral resolution and the observation window (OW); this is important for limiting the need of sophisticated and relatively expensive signal processing systems and high processing speed requirements, especially where multi-criteria arc detection strategies are involved and the analysis of more than one arcing characteristics must be performed in reasonable short time.In this framework, the experimental study herein presented has been aimed at investigating the suitability of using low frequency harmonic current analysis for arc detection purposes, even considering real case noisy situations [1]. An experimental characterization of series arcs has been carried out, with both laboratory and on-field tests, reproducing some arcing conditions, in accordance with the UL 1699B Standard requirements [2]. A preliminary analysis of the arcing current has been made in the low frequency range, in order to highlight some relevant characteristics for the arc detection purpose, which can be easily measured without the need of sophisticated measurement instrumentation.