Aiming at reducing the error recognition rate of low-voltage arcing fault, based on signal conversion, this paper develops a novel detecting algorithm for series arcing fault diagnosis. Via hysteresis comparator, the load current is converted to pulse signal, creating a time series of pulse width. The unpredictable fault current makes the time series fluctuate aperiodicity, thus, after a proper difference processing, the arc-fault signal is extracted from the time series, which results in a fabulous clustering distribution character in the statistical point value figure. The feature vectors are constructed. Based on the different Camberra distance of steady current and fault current, an arcing fault diagnosis algorithm with pattern recognition of pulse-width fluctuation is confirmed. Solutions of false recognition problem of AFCI are also discussed. The prototype of AFDD has been developed after results of algorithm simulation became desirable. Testing results show that this algorithm can be applied to online arcing fault detection with high feasibility and reliability.