Research On Dynamic Characteristic And Fault Diagnostic Method For Low Voltage AC Arcing

Electricity and electrical devices are important resources for modern daily life.In buildings or residences,when the electric wires working,air gap insulation breakdown or arcing may happen,resulting from wiring aging,loose connection of junctions or insulation damage by external intrusion.The electrochemical reaction and releasing energy of arc would in turn worsen the insulation and be likely to set the surroundings ablaze,leading to electrical fire.Therefore,it is of great necessity to study the dynamic electrical characteristics of arc and work on reliable diagnostic methods to detect the signs of incipient arcing and eliminate the hidden danger,for the sake of device and transmission line safety.This dissertation contributes to the dynamic characteristic study and fault diagnosis of low voltage AC arcing in various load types application,and the alternating arc in air under 220V/50 Hz power source is employed.The whole research is on three aspects,which are dynamic volt-ampere property modelling of arc,diversity in arc current and randomness mechanism of arc,and arc fault diagnostic method.For the dynamic volt-ampere property modelling,an arc length related mathematical model is established for the horizontal AC arc with carbon-copper electrodes.The arc cycle is divided into three stages:starting,burning and nearly quenching.The functional relationship between the field strength near the cathode and the external voltage in the process of breakdown and reburning of arc,as well as the arc energy dissipation function in the arc burning stage are analyzed.Combined with experimental measurement,the basic volt-ampere relation equation of arc process is obtained,and the specific parameters of the model are calibrated by experimental measurement and numerical fitting.Simulation and experimental results show that the computed arc current waveform and arc voltage waveform of the proposed Arc?L model are close to the experimental measurements under the same condition,and the performances are better than those of the classical arc models.The calculated waveforms under different arc lengths of the Arc?L model are in accordance with the measured relationship between the arc length and the change of arc voltage and current.Furthermore,the Arc?L model is verified under resistive load,resistive-inductive load and dimmer load applications.The computed results are close to the measured waveforms.These performances indicate that the Arc?L model describes the volt-ampere characteristic of arc properly and also helps to reveal the basic characteristics of arc fault under various load applications.Modelling and categorizing a load according to its impedance property throws light upon subsequent arc identification method construction under multiple loads.As the diversity of arc current challenges the series current analysis based arc fault detection,this dissertation studies the arc current diversity under various electrical appliances.Vast experiments are conducted to obtain the current data under the condition of various burning intensities of arc for the selected loads,and the corresponding current sample database is established.Together with the model simulation arc current and voltage in Chapter 2,these data are for the arc characteristic study.Feature engineering,such as the frequency domain analysis,time-frequency domain synchronous analysis,numerical statistics analysis etc.are carried out on the samples.The obtained general and individual characters of the raw current and the derived indicators could be guidance for the feature extraction,feature selection and data model construction in arc detection in the following chapters.In order to solve the key problems such as identifying the complex characteristics of different state currents under various loads,balancing the performance and complexity of the detection method,and considering the practicability of the method,this dissertation proposes a hybrid time and frequency analysis and fully-connected neural network(HTFNN)based method.The HTFNN method has a structure of three stages when doing the identification: firstly computing the necessary time and frequency features,then conducting the brief category division based on the fundamental frequency component,finally discriminating between normal and arcing state within each category by the corresponding concise neural network(NN).This strategy solves the problem of feature overlapping of normal and arcing states among different load types.By using lightweight NN and load sensitive features in each category,the accuracy of identification is improved and the complexity of network is reduced.Experimental results show that the HTFNN method achieves an average of 99% accurate state identification for typical load types with low complexity operation.The method is implemented and tested on the development board,and can identify the sample state within 3ms,with99% test accuracy for thousands of samples,which verifies the validity of the method and the feasibility of hardware application.For the randomness of low voltage AC arc,this dissertation observes the macroscopical and microcosmic random phenomena of arc through experiments and numerical simulation.The functional relationships between the mobility of charged particles and the external electric intensity in the process of air gap discharge under the action of external electric field are derived.The varied input energy of arc,and the coupling restriction between the conductivity of the charged particles and the external electric field make the arc working in a dynamic changing process.Together with these factors,the randomness of the collision among charged particles inside the arc,and the randomness of the initial position and process of the current column discharge induced by electron avalanche are the internal causes of arc's randomness.Furthermore,feature engineering indicates that the macroscopic randomness of the arc could be characterized by some indicators of arc current.A method based on covariance and correlation coefficient of the adjacent periods current is proposed,which could reflect the characteristics of the high stability and consistency between the adjacent periods of normal current,and the significant difference and randomness between the adjacent periods of arcing current.Confront with the diversity and randomness of arc current induced by various load types and arc unstable combustion,crucial problems have to be tackled,such as the poor identification performance and generality of the characteristics obtained by using the inherent basis when analyzing the current signal,the difficulty in finding the common characteristics among loads and the complexity of the corresponding classifier would increase along with the increament of load types.This dissertation proposes an accurate load type and arc fault identification method based on sparse representation and fully-connected neural network(SRFCNN).The methodology is elaborately structured by a pretreatment layer,a sparse representation layer and a decision layer.It designs customized dictionaries for the signal to be identified,extracts the characteristics of various signals adaptively,and recognizes the working state and load type of the computed sparse coefficients with the fully-connected neural network.The sparse coefficients serve as the distinction between normal and fault states under each load,as well as among various load applications.Experimental tests on the samples of various load types show that the identification accuracy of the working state and load type of single half-cycle current reaches above 94%,and the state identification accuracy of every half-cycle sample is more than 97%,with a nealy 100%accuracy of arc fault detection.The performance indicates the effectiveness and reliability of the proposed method in the identification of load types and normal and arcing states.