Study On Automobile Arc Fault Detection Methods

The number of vehicle holdings in Chinais increasing, and automobileelectricalfiresbreak out more frequently, whichcauses severe losses of social property and threatens personal safety. Arcfault is the main composition of automobile electrical fires and the existing types of protective devices could not provide automobilesreliable protectionsagainst arcfault.Therefore, it’s necessary for studies on the detection of the automotive arc-fault.This paper puts detection methods of automotive arc-fault as the research objects, and seriously analyzes the arc fault characteristic of automotive typical load operation. Then, the paper puts forwardan automobile arcfault detection method on the basis of BP neural network algorithm.According to the electronic structure of medium or large buses we establisharcfault experimental platform and design experiment scheme. Taking three kinds of electrical appliances including the lights, motors of wipers, horns as single loadsand composite loads, we do experiments of the automobile fault of arcs that aregenerated by an arc generating device in a series circuit of automobiles.Contrast analysis of signals of single loads and composite loads in the state of arc fault or normal are made in time domain and in power spectrum.The contrastive results showed thatwhen an arc-fault occurs,the current waveform in time domaingoes with a significantly higher frequency of signal. At the same timethe amplitude ofpower spectrum infrequency band from 60 KHz to 110 KHz risesand the overall power in this frequency band has very different from normal operation.On the basis of analysis of the fault characteristics of the signals this paper designs the band-pass filter with the pass band from 60 KHz to 110 KHz to extract the signal inthat frequency band, which band will be taken as feature band of automobile arc fault. This paper divides the characteristic frequency band of the arc fault into five power bands and calculates the power value of these bands, of which five characteristic vectors of an arc-fault are made up, when automotive single loads and composite loads are in the state of arc fault or normal. With the network inputs of these five characteristic vectors and outputs in present state of arc-fault or non-fault, BP neural network algorithm is applied to train and test samples ofautomobile arc-fault. Testing results show that this method of detecting automobile arc-faults is in effect and useful.Finally, According to the distribution of five power bands of automobile arc fault this paper designs hardware and software of the detection prototype that is based on the DSP digital signal processor and completes the prototypedebugging to realize online detection of automobile arc fault.