Research On Data-driven Fault Feature Extraction And Diagnosis Of Rolling Bearings

Rolling bearing is more widely used in industrial production of machinery parts,which will vibrate when the failure occurs.The fault signal from vibration signal analysis of the characteristics of the component is effective for bearing fault diagnosis.Because the rolling bearing is working under environment with noise and the early fault signal is weak,the traditional fault analysis method is more difficult to extract the fault characteristic components effectively.From the perspective of vibration signal analysis and processing,this thesis analyzes the vibration signal constantly and makes improvements to extract more effective fault characteristic values so as to achieve the more accurate fault diagnosis on bearing inner ring and outer ring,rolling body.The main work contents are described as follows.(1)The ensemble empirical mode decomposition(EEMD)method can not completely eliminate the mode aliasing phenomenon due to the noise of the working environment of rolling bearings and weak early fault signals.When extreme points are selected for upper and lower envelope fitting,some extreme points will affect the fitting effect and mean curve.The intrinsic mode function(IMF)component decomposed by EEMD has the same high dimension as the original fault signal,which will lead to the reduction of the fault diagnosis accuracy of the classifier.Therefore,a radial basis function(RBF)neural network fault diagnosis method for rolling bearings based on improved EEMD and singular value decomposition(SVD)was proposed.When EEMD algorithm is used,the extreme points affecting the fitting envelope are removed,and then the fault signals are decomposed to obtain several groups of intrinsic mode function components(IMFs).Then these IMF components as well as the largest energy ratio of several IMF component are constructed into a new matrix,and then based on the matrix SVD decomposition,a set of low dimensional singular values are obtained to be fend into the RBF neural network instead of the original fault signals for rolling bearing fault diagnosis.Finally,the simulation experiments based on the rolling bearing fault datasets of Case Western Reserve University verifies the effectiveness of the proposed method,and the experimental results show that the improved EEMD algorithm has a great improvement in the fault feature extraction ability compared with the traditional algorithms.(2)The features are easily lost by using a single feature extraction method.A novel rolling bearing fault diagnosis method was proposed based on the variety of feature extraction and information fusion strategy,which can effectively avoid the phenomena that single feature extraction method is vulnerable to the external environment and the signal itself characteristics caused by the fault characteristics of leakage problems.Firstly,the wavelet accumulation energy parameters are calculated by wavelet decomposition of the signals,and the empirical mode decomposition(EMD)and the improved EEMD algorithm are used to decompose the signals so that the dimension reduction is realized to obtain the characteristic parameters of the signals.The timedomain performance index of the signals is calculated to obtain the time-domain characteristic parameters.Then all the obtained feature parameters are filtered with low variance and Pearson coefficient to reduce the dimension of feature vectors so as to facilitate the setting of neural network classifier.Finally,Bidirectional-Long Short Term Memory(BI-LSTM)neural network was used as a rolling bearing fault diagnosis classifier to test the effectiveness of the proposed method.Experimental results show that the fault diagnosis accuracy of the proposed method is higher than that of single algorithm.(3)Rolling bearing fault diagnosis is influenced by many factors,such as industrial environmental noise,which results in the decomposed signal components with some redundant components.At the same time,empirical mode decomposition(EMD)and its related improved algorithms have the modal aliasing phenomenon,which also makes the component have more invalid characteristics.These phenomena caused great influence on bearing fault diagnosis.On the basis of the above improvements,a rolling bearing BI-LSTM fault diagnosis method based on segmental interception auto regressive(SIAR)spectrum analysis and information fusion was further proposed.After the improved EEMD,complementary ensemble empirical mode decomposition(CEEMD)and robust empirical mode decomposition(REMD)algorithms are respectively used to decompose the rolling bearing fault signals,the obtained components are analyzed by AR spectra.By comparing the AR spectra of corresponding components at different fault locations,the effective AR spectral values are intercepted as the eigenvalues of the data,and finally all the eigenvalues are fused to achieve the purpose of more efficient screening of effective features.It can reduce the impact of feature redundancy caused by modal aliasing on neural network training.The BI-LSTM neural network is used as the rolling bearing fault diagnosis classifier,and then the simulation experiments based on the rolling bearing fault signals from Case Western Reserve University was carried out to verify the effectiveness of the proposed feature extraction and fault diagnosis method.The experimental results show that the performance of fault diagnosis is significantly improved by adding segmented AR spectrum analysis.