The invention relates to a
DC fault arc detection method. The method is characterized by comprising the following steps of 1, sampling, filtering and conducting the fast Fourier transformation on an input current of a high-frequency power electronic converter to obtain an amplitude-versus-
frequency curve for the high-frequency components of the current; 2, selecting at least one
frequency band containing a
switching frequency or a
multiple frequency from the amplitude-versus-
frequency curve, and calculating the
peak value D1 of the magnitude of the curve and the mean value D2 of the magnitude of the curve within the selected
frequency band. Based on the change of the distance between the
peak value D1 and the mean value D2, whether an arc is generated or not can be judged. If the mean value D2 gets closer to the
peak value D1, the arc is judged to be generated. Otherwise, no arc is generated. According to the novel
DC fault arc detection method, based on the own characteristics of the converter,
noise signals are generated by the high-frequency components of the
direct current at the frequency multiplication points of the
switching frequency of the converter. By utilizing the
noise signals, the arc spectrum of an arc fault can be analyzed and calculated. In this way, whether the arc fault occurs or not can be detected.