Research On Fault Diagnosis Of Motor Car Bearing Based On Sparse Representation

In recent years,rapid development of high-speed railways and safety issues in train operations have become the focus of attention.The high-speed running bearing is an important part of the locomotive which is the most easily damaged part.Once the fault occurs,it will rapidly increase,resulting in bearing failure and train accidents.Therefore,the detection of the state of the bearing is of great significance to driving safety.The state of the equipment can be judged in time through the analysis of the vibration signal,which is the most intuitive reflection of mechanical equipment.Due to the initial bearing fault impact signal often submerged in strong noise,it is difficult to extract the fault feature of the signal.In order to improve the effectiveness of fault diagnosis for low speed and heavy haul train bearings,we extract the impact of bearing vibration signals based on sparse representation and EMD.In this paper,the signal characteristics of bearing faults and the frequency of bearing fault feature extraction are studied.According to the sparsity of bearing fault signals,the method of extracting bearing fault features based on sparse representation is proposed.The main work is as follows:Firstly,the bearing fault signal characteristics,the bearing structure and the calculation method of the frequency of the fault feature are analyzed.The research progress of the bearing fault detection technology is introduced.The sparse representation is selected to extract the fault feature according to the signal characteristic of the bearing fault.The KSVD algorithm is used to train the adaptive bearing fault dictionary to solve the difficulty of dictionary constructionSecondly,the characteristics of signal de-noising by Minimum Entropy Deconvolution(MED)are studied and introduced in sparse representation dictionary training.Because the bearing fault dictionary training sample itself is noisy,it will directly affect the sparse reconstruction result.Thus the MED is used in dictionary training to preprocess samples to reduce dictionary atomic noise and improve the purity of the atom.Finally sparse representation combined with EMD decomposition is used to reduce noise.In order to eliminate the noise further,the sparse results are further decomposed by EMD.According to the kurtosis criterion,the higher the Kurtosis value,is the more the impact component the signal contains.Thus,we use the intrinsic modal function(IMF)decomposed by EMD for Hilbert envelope analysis and extracts feature frequencies of fault signals.It is proved by simulation and bearing data test that this method can effectively highlight the impact components of the signal,increase the signal kurtosis value,improve the envelope spectrum signal-to-noise ratio,detect more effectively the bearing fault frequency.