Impact of the Sampling Frequency on the Detection of Series DC Arc Faults in an Aeronautical Environment Using Machine Learning Techniques

The aeronautical industry imposes high security and redundancy in all its embedded systems (electrical, hydraulic and mechanical). Therefore, the increase in voltage levels expected in More Electric Aircraft or propulsion applications pushes studies to consider more carefully the phenomenon of arc faults. This failure can be more frequent and dangerous at high voltage and high current levels. Currently, there is no robust and reliable embedded system capable of detecting, with high accuracy, arcs on aircraft. To recognize the series DC arc faults from the rest of the nominal behavior of the system, this article implements a linear classifier founded on the extraction of relevant features from the conventional current signals. The feature ranking is performed through the orthogonalization Gram-Schmidt procedure and the feature selection through a wrapper approach. The used database contains current signal measurements of series DC arc faults and nominal behaviors (simple and complex). Due to the high requirements in terms of performance and real time implementation, the impact of the sampling frequency (SF) of the current sensor signals is studied.