Research On Multi-feature Aviation Arc Fault Detection Algorithm Based On Pattern Recognition

In today’s society,the number of electrical occasions is increasing,and electrical fires account for more than 30% of the total number of fires,which greatly endangers the safety of people’s lives and property,especially electrical fire accidents caused by arc faults in aircraft.Today,in the aviation electrical system,continuous integration and high portability are being sought.Detection is a key research area that people are paying attention to.In an aircraft,cables are usually bundled and bent inside the fuselage,and they work in extreme environments such as high temperature,low temperature,radiation and vibration for a long time.It is easily caused by aging of insulation,friction between cables or between cables and fixed frames.Damage to the insulation layer results in arc faults.Therefore,in this paper,the theoretical analysis,model simulation,data acquisition,feature extraction,training of arc fault identification models,and other aspects of arc faults in aeronautical AC systems are studied,and the model’s ability to identify arc faults in different environments is tested.,Designed to reduce or eliminate the potential dangers caused by arc faults to the aircraft and protect the personal safety of the crew and passengers.First,referring to the description of the arc fault experiment in the relevant international standards for arc fault circuit breakers(AFCB),the aviation AC static transformer power supply is used to carry out series and parallel arc fault experiments under various load conditions.The characteristics of the current signal extraction circuit under normal and arc fault conditions.Subsequently,the time-frequency domain,frequency domain,and time domain methods were used to extract the maximum overlapping discrete wavelet packet transform(MODWPT)percentage energy spectrum of the loop current signal,and the linear frequency sweep Z modified by the bimodal spectral line interpolation technique Transform(CZT)refines the first 20 odd and even frequency harmonic amplitude sums of the frequency spectrum,the difference between the positive and negative peaks,and the time it takes for the current to cross zero,totaling 28 feature quantities.The analysis shows that the feature quantity can effectively distinguish the two working states of the loop,and can be used as the input feature quantity when training the classifier for identifying arc faults.Considering the effect of the feature dimension on the training process of the classifier,binary particle swarm optimization(BPSO)is used to reduce the feature dimension to find the optimal feature subset,and the feature recursive elimination technology is combined with BPSO to provide excellent optimization algorithms.The initial particle accelerates the convergence speed and accuracy of the optimization,and introduces a K nearest neighbor(k-NN)classifier in the calculation of the BPSO fitness function.Analysis of the results of the optimization iteration shows that the k-NN arc fault recognition model trained on the optimal feature subset has significantly improved the recognition rate of the test set features.Finally,the accuracy,misdetection,and missed detection rates of the arc fault identification model in unknown load environments,variable frequency test environments,and crosstalk interference immune test environments are examined,with a higher accuracy rate and a very low false detection rate.And the missed detection rate meets the requirements for detection capability in the aviation AFCB standard.