Research On Vibration Signal Feature Extraction Method Based On Time-frequency Analysis And Convolutional Neural Network

With the development of science and technology and the advancement of the times,mechanical equipment has become more and more high-performance,high-load,and complicated.It has expanded from mechanical equipment such as spacecraft,vehicles,aviation generators,to precision instruments and meters,in the national economy.Has a pivotal role.Rolling bearings are the most commonly used components in mechanical equipment.The performance of rolling bearings directly affects the operating reliability of the entire system.The failure of rolling bearings may cause complete failure of mechanical equipment,resulting in huge economic losses and even casualties.Therefore,in order to ensure the stable operation of mechanical equipment,it is necessary to propose a reliable and effective rolling bearing fault diagnosis method,timely detection of early faults of rolling bearings and correct diagnosis to prevent their further deterioration.In this paper,the vibration signal of the rolling bearing is taken as the research object,combined with time-frequency analysis to extract the feature of the signal,use the convolutional neural network for fault diagnosis,and for the multi-working conditions and high noise environment,an adaptive intelligent fault diagnosis optimization method is proposed..First,based on the fault mechanism of the rolling bearing,different time-frequency analysis methods are used to extract the characteristics of the vibration signals of each fault state.It is verified that the time-frequency domain characteristics can better characterize the fault characteristics.On this basis,a method of generating timefrequency images using time-frequency transform is proposed to extract features of the signal,and continuous wavelet transform is used to generate time-frequency feature images.The effectiveness of the method is verified through experiments.Using data enhancement technology to generate time-frequency feature image Data Set class data set.Secondly,by studying the basic algorithm of convolutional neural network,and comparing the impact of the model structure and algorithm on the recognition results,build an intelligent fault diagnosis model for fault recognition of time-frequency images,and address the problems of poor diagnosis results on the Paderborn University data set To improve the model in terms of model structure and regularization algorithm,the final experimental result can reach more than 96%.Finally,this paper addresses the problem of unstable model performance under variable operating conditions and high noise environments.An adaptive filtering algorithm for vibration signals and a convolutional autoencoder model for image reconstruction for time-frequency images are proposed for optimization,which greatly improves the model's adaptive ability under noisy and variable operating conditions.The superiority of this method is verified by comparing with other intelligent diagnosis algorithms.