Research On Rolling Bearing Fault Diagnosis Method Based On Data-driven And Deep Learning

As one of the main parts of industrial machines,A great impact on normal industrial production will emerge if the rolling bearings fail.However,due to the complex working environment,it is difficult to find the fault in time.The most common method now is to use vibration data for fault diagnosis.However,the collected fault data are generally the non-stationary signal with high noise and high frequency variation,which contains a lot of redundant information.So it is difficult to effectively extract and diagnose fault features.Therefore,this thesis conducts research on fault diagnosis methods based on data-driven and deep learning.The main work is as follows:(1)A bearing fault diagnosis method based on Local Fisher Discriminant Analysis(LFDA)and Sparse Representation(SR)is proposed.In view of the large amount of bearing fault data,redundant information and the signal changing drastically,the global LDA method is used for dimensionality reduction and feature extraction,which cannot fully reflect the local characteristics of the signal fisher.Therefore,this thesis first uses LFDA to extract local fisher features of bearing signals;secondly,in order to further filter out the high-frequency disturbance information contained in the signal,the SR method is used to process the dimensionality-reduced data.By constructing an adaptive feature dictionary,the fault signal is sparsely represented using the orthogonal matching pursuit algorithm;Finally,the test samples are classified using the minimum reconstruction error method.The experiments are carried out with bearing datasets from Case Western Reserve University and Xi’an Jiaotong University for analysis and verification.The results show that the proposed method can effectively diagnose faults.(2)A bearing fault diagnosis method based on Soft SSC Empirical Mode Decomposition(Improved EMD,IEMD)and adaptive residual shrinkage network is proposed.Due to the influence of rotation speed,load and environmental noise,the characteristic frequency of the collected vibration signal is large,which has complex characteristics on different time scales.Therefore,starting from the nature of the signal,this thesis first uses the IEMD method to process the fault data.By calculating the IMF components on different scales,the IMF components with greater correlation with the original signal are retained,which can not only better characterize the characteristics of the original data,it can also reduce the interference of harmonic noise.Then,in order to make full use of the features of each scale in the data,an improved residual shrinkage network is established.The adaptive parameter Re LU is added to the residual shrinkage structure to effectively retain the data features of different scales and improve the performance of the network model.In order to verify the effectiveness of the proposed method,the bearing fault data collected from Case Western Reserve University and the actual data collected based on our platforms are used.The effectiveness under the noise conditions was verified by adding noises with different signal-to-noise ratios to signals.The experimental results show that the proposed method can effectively realize the fault diagnosis of rolling bearing with high noise and multiple redundancy.