Arc Source Recognition and Locating Based on Electromagnetic field analysis

Abstract

An electric arc is the breakdown of gas discharge between electrodes. In daily life and industrial production, electric arcs can be divided into two categories: normal arcs and arc faults. An arc refers to the spark discharge during the normal operation of electrical equipment, such as a contact arc generated by a switching-in in power supply, a spark used in electric welding, or an arc produced in the electric insertion process. Arc faults are the discharging caused by equipment failure or working under abnormal conditions, such as circuit failures including aging, overload, short circuit, current instability, fusing, loosening of a conductor, or ground fault. It is a dangerous sign in electric system operation when an unusual arc occurs, since an arc fault can not only affect the equipment’s normal operation, but it can also cause an electric breakdown that triggers a power outage. In certain working environments, arc faults can even cause a fire hazard or explosion. According to fire statistics, about 30% of industrial accidents are fire disasters, and most fires are caused by arc faults. Therefore, arc fault monitoring is necessary to guarantee the safety and security of power system operations. If arcs can be found in advance, accidents can be effectively avoided, and electric faults can be discovered accordingly to prevent systems from being affected. In order to effectively protect a system, an analysis and recognition of the arc signal is important. This research not only includes type identification of electric arcs, but also identifies the locations of arc sources for timely detection of electrical faults and the prevention of potential accidents. In the process of arc discharging, there are a series of physical phenomena such as optical radiation, light, high temperature, and detonation. An initial current pulse with a fast rise time and short duration is generated in the arc production process, along with an electromagnetic field and wave generated by strong electromagnetic radiation. This provides a new way for analysing an arc signal. The main objective of this thesis is to recognise, classify, and locate arc faults by analysing electromagnetic signals and provide a basis for fault diagnosis and monitoring. The recognition and classification involved in the thesis includes clustering AC and DC arc discharge, arc signal recognition from an electromagnetic environment with strong noise, recognition of quantified features extracted from arc signals, and location recognition of an arc source. The key procedures of the framework include five steps, A) Arc-generation mechanism analysis and electromagnetic field (EMF) model establishment: The numerical analysis of the electromagnetic signal of an arc source explores electromagnetic distribution in order to provide a basis for the location selection of the positioning of sensors and arc source locating; B) Electromagnetic signal identification from electromagnetic noise environment: specifically, research on noise suppression and signal separation, laying the foundation for the analysis of electrical arcs and breakdowns; C) Rough clustering of electrical arcs: analysis of arc signals in a time-frequency domain, which has important significance for the practical diagnosis of an arc fault; D) Extraction features of an arc signal: Takes the brush work state of a DC motor as an experiment target, and extracts and quantifies the features from the signal of the arc generated in abnormal brush conditions, based on the different working conditions identified; E) Identification of arc source location: includes establish an EMF map for visual observation of arc source in a small range of two-dimensional space and calculation of arc coordinates in a wide range of three-dimensional space. The method used for electromagnetic signals generated by electric arcs includes finite difference time domain (FDTD), the wavelet transform algorithm, the blind separation algorithm, the local mean decomposition (LMD) algorithm, particle swarm optimisation algorithm (PSO) combining extreme learning machine (ELM), or PSO-ELM, the band entropy algorithm, a bi-spectrum analysis, a trend surface polynomial model, linear interpolation, ternary symmetric matrix method, and the weighted centroid algorithm. The FDTD method is applied to the analysis of the electromagnetic model of the electric arc. The algorithm of wavelet transform and blind source separation algorithm are for arc signal separation and identification from noise. The LMD algorithm and PSO-ELM are for AC and DC arc distinguishing; the band entropy and bi-cepstrum analysis are for quantification and extraction of arc features to achieve arc fault recognition; and the trend surface polynomial model and linear interpolation are for two-dimensional spatial electrical arc source location identification. The ternary symmetric matrix method and weighted centroid algorithm are combined for three-dimensional space electric arc source location identification.