Arc fault diagnosis and analysis based on wavelet neural network

Abstract

Research on the external characteristic of the switch with contact is an important subject in breaking techniques nowadays. In this paper, based on the proposed experimental platform covering the arc fault diagnosis, load selection and wavelet neural network (WNN) design, a combination method of the experiment and simulation of the actual lines for the arc fault testing system is presented. And the external parameters of the series arc fault has been obtained under different loads, such as the resistance-reluctance loads and a series of electrical loads. By using the wavelet packet transform, frequency bands subdivision of experiment data and energy statistics under different frequency band are accomplished. And the current energy distributions for different loads under fault condition have been obtained and figured out. Additionally, the proposed energy criterion is applied as the input and the trained relax-model WNN can identify the arc fault effectively. By analyzing the arc faults of different loads for the demonstrated cases, the feasibility and the validity of the proposed scheme for arc fault diagnosis is verified.