Series DC Arc Fault Detection Using a Wavelet-Based Filter Bank with Statistical Analysis

A real-time algorithm is developed for the detection of series dc arc faults in solar photovoltaic installations. A wavelet-based filter bank is used to condition the sensed dc input current, which is sampled using an isolated sigma-delta ADC. The filter bank improves the robustness of the algorithm to false trips by attenuating the 120 Hz ripple and switching harmonics. In the algorithm, the variance of the filter bank output is calculated and compared to an upper threshold value to determine if a fault has occurred. The algorithm senses a fault when the upper threshold is exceeded, and the dc bus voltage falls below a predefined lower limit. The algorithm can detect a series arc fault in under two seconds and is found to be robust to false trips induced by power converter switching harmonics. The filter bank is implemented using 32-bit floating point values, and the detection algorithm is programmed on a microcontroller.