Research On Asynchronous Motor Fault Diagnosis Based On EWT And Hybrid Kernel Extreme Learning Machine

With the development of industry and the continuous innovation of technology,AC asynchronous motor has become an essential part of People’s Daily production and life.Especially in wind power generation,new energy vehicles and other emerging fields as well as coal mine mechanical and electrical equipment,especially factory machine tools and other traditional industries are still widely used.Ac asynchronous motor can operate normally and stably,it directly affects the stability of the whole equipment system,and even affects the economic benefits of industrial production and even personal safety.The continuous development of machine learning provides new ideas for asynchronous motor fault diagnosis.When a fault occurs in an AC asynchronous motor,different degrees of vibration signals will be generated.This kind of vibration signal often contains large number of time-varying and emergent characteristic components,so how to extract these components is very important.Based on asynchronous motor of publicly available data sets,put forward a kind of experience based on the Empirical Wavelet Transform(EWT)feature extraction method,with the traditional empirical mode decomposition and its improved algorithm(EMD,EEMD,CEEMD)algorithm for performance comparison,proved EWT algorithm can extract the main components in complex information,can provide good training model for fault diagnosis algorithm.EWT for each type of fault data respectively,extract the Empirical Mode Function(EMF),and then according to the Principal Component Analysis(PCA)on the EMF data set dimensionality reduction processing,with t-SNE visual features to compare and analyze the dimensionality reduction results.By combining the above methods,the final output feature dataset sample is used as the input dataset of the fault diagnosis algorithm.Fault diagnosis algorithm is a kind of polynuclear extreme learning machine optimization method is based on the sparrow.At first,this paper introduces a kind of the Extreme Learning Machine(ELM)algorithm is introduced into multiple kernel function combination of hybrid extreme learning machine(HKELM)algorithm,this algorithm nuclear parameters and weight coefficient by means of the Sparrow Search Algorithm(SSA)in accuracy for the fitness function parameters optimization.Secondly,the population updating formula of sparrow search algorithm was improved,and a sparrow search algorithm improved by random walk strategy was proposed.The improved Sparrow optimization multi-core extreme learning machine is used to optimize the hyperparameters,which not only makes up for the influence of the random input parameters of the HKELM model on the fault diagnosis accuracy performance,but also improves the optimization efficiency to a certain extent.Through the experiment of example data set,the influence of manual selection of some parameters on the accuracy of HKELM and the effect of Sparrow optimization and improved optimization on HKELM model are proved.Figure 35 table 15 reference 85...