Research On GIS Mechanical Fault Recognition Method Based On Auditory Feature

With the rapid development of electric power technology and systems in our China,the electric power system tends to develop toward higher voltage and larger capacity,which raises requirements for the safe and stable operation of electric power equipment.Because the gas-insulated switchgear(GIS)has the advantages of small space occupation,high security and stability,long maintenance cycle,and is not easily disturbed by external environment,it is widely used in substations at all levels.However,with the expansion of the application of GIS,various problems are increasing.Due to the key components of GIS are enclosed in metal shells,it is difficult for maintenance personnel to directly detect faults.Moreover,there is currently limited research experience on GIS mechanical faults,and traditional mechanical fault detection methods are relatively complex when facing complex working environments and strong background noise in the diagnosis of GIS mechanical faults.Inspired by workers using their auditory sense to judge the working conditions of mechanical or electrical equipment during operation,this thesis proposes a GIS mechanical fault diagnosis method based on auditory feature.This method has superior performance in extracting sound signal features under strong background noise,and the deployment of the diagnostic model requires low hardware computing power and is easy to deploy.In addition,it has high accuracy in GIS mechanical fault diagnosis tasks,specifically as follows:(1)The structure of GIS and the cause of mechanical failure are analyzed,and the experiment of sound signal acquisition under various GIS conditions is carried out.The sound signals generated by GIS equipment in normal working state and three mechanical fault states were acquired,and equal length data samples were intercepted from the collected sound signals to form GIS multi-state working sound sample data set,which provided data conditions for GIS mechanical fault diagnosis method based on sound signals.(2)An auditory model is introduced to construct a tool to extract the auditory features of GIS working sound signal.By analyzing the hierarchical structure of the ascending auditory pathway and its sequential nonlinear feature extraction calculation,an auditory model simulating the sound processing characteristics of the auditory center is constructed using a hierarchical spiking neural network(HSNN)based on optimal coding theory.The Gammatone filter bank,which simulates the frequencyselectivity properties of the human cochlea,is employed as the pre-processor of the auditory model.By decomposing the input signal into multiple frequency bands,the time-domain multi-frequency characteristics obtained by frequency division of the input signal and represented as spectral input to HSNN.Finally,the Spike Firing Rate Spectrum(SFRS)of HSNN output is used as an auditory feature for GIS mechanical fault diagnosis.(3)An GIS mechanical fault diagnosis method is been constructed based on SFRS features and lightweight convolutional neural network.Taking into account the high-frequency component in the SFRS characteristic matrix of GIS signal features extracted by HSNN,this thesis optimizes the lightweight convolutional neural network,Ghost Net.By incorporating a Spatial Frequency Block(SFB)that is sensitive to high-frequency features into the network,the detailed components of the features are modeled to improve the network’s modeling performance for highfrequency signals.Then,by comparing the influence of different network structures and hyperparameters on classification accuracy,the superiority of the network classification ability after the insertion of SFB module is verified,and the optimal network model is selected.Finally,the optimal network model was used to compare the classification accuracy of the proposed SFRS feature extraction method with commonly used feature extraction methods under different levels of background noise,demonstrating its noise resistance performance.(4)GIS mechanical fault intelligent diagnosis system based on Lab VIEW is designed,and the verification experiment is carried out on the platform.The integrated virtual instrument includes functions of data acquisition,data processing,data management and fault diagnosis.The system provides intuitive display of timefrequency domain waveform,SFRS features,and fault types of the collected signals.The Lab VIEW and MATLAB mixed programming approach is employed for feature extraction of GIS sound signals,and a classification model is called through Python nodes to predict the mechanical fault types of GIS.Through the verification experiment,it is proved that the intelligent diagnosis platform can meet the requirements of GIS mechanical fault diagnosis in four working states.