Research On Rolling Bearing Fault Diagnosis Method Based On Shallow Machine Learning And Deep Transfer Learning

Rolling bearings are one of the most commonly used but also the most prone to failure of key components in rotating machinery.The running performance of rotating machinery is closely related to the health status of rolling bearings.Therefore,the research on fault diagnosis of rolling bearings has always been an important research content in the field of mechanical fault diagnosis.Timely and accurate fault diagnosis of rolling bearings plays an important role in ensuring the safe and reliable operation of mechanical equipment,avoiding major accidents and reducing production costs.The ways of rolling bearing fault diagnosis are various,which can be carried out through the traditional signal processing method,or by using data-driven intelligent fault diagnosis model to expand rolling bearing fault diagnosis.This thesis carries out multi-dimensional health assessment of rolling bearings from three ways: traditional signal processing,shallow machine learning and deep transfer learning.The main research contents are as follows:(1)Aiming at the problem that the bearing fault impact will arouse the natural vibration frequency of the entire bearing system and modulate it with the characteristic frequency of the bearing fault,a signal processing method based on adaptive resonance demodulation is used to carry out the fault diagnosis of rolling bearings.In order to amplify and separate the fault characteristic signal from the original signal,a Butterworth filter is first used to filter the original signal,thereby extracting the high-frequency resonance signal.For the selection of the filter frequency band,fast spectral kurtosis is used to adaptively determine the optimal filter frequency band.Then use the Hilbert transform to obtain the envelope of the high-frequency resonance signal to finally realize the resonance demodulation.Finally,the use of faulty bearing data in two datasets verifies that the adaptive resonance demodulation algorithm shows good diagnostic results for different types of bearing faults at different speeds.(2)Aiming at the problem of SVM parameter sensitivity and feature redundancy in the fault diagnosis of rolling bearing based on Support Vector Machine(SVM),a SVM parameter optimization method based on improved Whale Optimization Algorithm(WOA)and a Recursive Feature Elimination algorithm based on Support Vector Machine(SVM-RFE)are proposed.In order to make SVM perform better,WOA was improved and used for SVM parameter optimization,and compared with the optimization effect of SVM based on standard WOA,genetic algorithm and particle swarm optimization algorithm.In order to obtain sufficient features to characterize the health status of rolling bearings,this thesis first extracts the time domain,frequency domain and wavelet packet features from the original signal,and then normalizes the features and uses the SVM-RFE algorithm to discard redundant features and form new feature vectors.Finally,two bearing data sets are used to verify the classification effect of the model proposed in this thesis.The results show that the model can achieve excellent classification results when the training set and test set data come from the same working condition or are natural damaged bearing data,but the training set and test set data come from different working conditions or are artificially damaged bearings data and natural damage bearing data,the classification effect of the model is not ideal.(3)In view of the fact that the rolling bearing fault diagnosis method based on shallow machine learning is difficult to solve the problem of cross-domain fault diagnosis in practical applications,the deep transfer learning method combining Multi-Kernel Maximum Mean Discrepancy(MK-MMD)and Domain-Adversarial Neural Networks(DANN)is used to carry out rolling bearing fault diagnosis.Firstly,the principles of MKMMD and Generative Adversarial Networks(GAN)are introduced.MK-MMD can be used to measure the distribution distance between source domain and target domain,and DANN developed from GAN can realize automatic alignment between source domain and target domain.This thesis combines the two to propose a deep transfer learning algorithm based on MK-MMD and DANN.Finally,the network model is verified by using all kinds of health state bearing data in two datasets,and the results show that the network model can effectively transfer the knowledge of labeled source domain data to unlabeled target domain data.