Research On Rolling Bearing Fault Diagnosis Based On Data Fusion

Rolling bearing is a key component of rotating machinery equipment,if it fails,it may cause serious accidents,resulting in serious economic losses and even casualties.Therefore,the fault diagnosis of rolling bearings,timely fault detection and maintenance is of great significance to ensure the safe and stable operation of rotating machinery.When the rolling bearing is damaged,its vibration signal will appear periodic impact components,but in practical work,due to the influence of the environment,the collected vibration signal will be disturbed by noise,the fault characteristics are weak,and it is easy to be submerged by noise.It is not easy to detect and identify.And the traditional fault classification method is mainly to analyze the signal and then classify it manually,which not only requires a lot of prior knowledge,but also is not easy to extract deep features,and the stability of the diagnosis result is poor.In view of the above problems,the fault diagnosis of rolling bearing is studied as follows in this paper.In order to solve the problem of noise interference and weak fault characteristics in the early fault signals of rolling bearings,a vibration signal preprocessing method based on Complete Ensemble Empirical Mode Decomposition(CEEMD)-wavelet threshold denoising and Multipoint Optimal Minimum Entropy Deconvolution Adjusted(MOMEDA)is proposed,which is used to preprocess the original vibration signals,reduce noise interference and enhance fault features.Firstly,the signal CEEMD is decomposed to generate a series of components,and the high-frequency components with high noise are denoised by wavelet threshold,and the signal is reconstructed to achieve signal denoising.Secondly,a new composite index,kurtosis-envelope waveform factor,is proposed,and a variable step search method is designed to optimize the filter length of MOMEDA algorithm as fitness function.Finally,the de-noised signal is deconvoluted by MOMEDA based on the optimized filter length to enhance the fault features.Experiments show that this method can effectively extract weak fault features under the background of strong noise.In order to solve the problem that manual classification requires a lot of prior knowledge,strong subjectivity and poor stability of diagnosis results,Convolutional Neural Network(CNN)is used to classify faults.First of all,according to the characteristics of less data and one-dimensional vibration signal,Le Net-5 is optimized and an one-dimensional convolution neural network(1D-CNN)suitable for rolling bearing fault diagnosis is generated.Then,in order to solve the problem that there is a lack of features in the process of feature extraction due to the limited information carried by single-domain features,which can not fully reflect the inherent characteristics of the original fault signal,a fusion network model is proposed.taking the time domain and frequency domain signals as the input of the network,the inherent characteristics of the original fault signals can be fully reflected,and the diagnosis results are more accurate.finally,the effectiveness of the method is verified by experiments.