Compound Fault Diagnosis Of Rolling Bearing Based On Sparse Representation Theory

Rolling bearing as a key part of mechanical equipment is important to ensure the normal operation of mechanical equipment.However,due to the complex working environment of bearings and coupled with the influence of installation accuracy and other factors,rolling bearing faults frequently happen.Besides,in the actual engineering,a variety of bearing faults often occur at the same time.Different faults interfere and couple with each other,which brings great challenge to the compound fault diagnosis.Therefore,the study of rolling bearing compound fault diagnosis method is significant to reduce or avoid the economic loss and safety accident caused by rolling bearing faults.Bearing compound fault signals have high complexity,but they can be sparse in some transform domains.Based on this feature,sparse representation theory can reduce the redundant information of fault signals and excavate the inherent structural features of signals more deeply.Therefore,based on the sparse representation as the theory foundation with rolling bearing as the research object,this paper solves the problem of the bearing compound fault diagnosis.The research work mainly contains three aspects,which are the impact atom as the overcomplete atom dictionary,wavelet transform as the dictionary and sparse model in time domain,and is verified by simulation experiment and different experimental platforms.The main research content of this paper includes the following aspects:（1）In view of the difficulty and mixing problem of compound fault separation with the impulse atom as the dictionary in the sparse model,sparse compound fault diagnosis based on periodicity weighted kurtosis and periodically local enhancement is proposed.With impulse atom as the dictionary to match the impulse component,periodicity weighted kurtosis considering repeatability and impact feature contributes to determining parameters and obtaining effective fault period.In sparse fault extraction method based on periodically local enhancement,the feature of the sparse coefficient corresponding to the fault period is used to remove noise component interference,enhances the sparsity of the solution,and separates bearing compound fault,which solves fault mixing problem and realizes bearing compound fault diagnosis.At the same time,the proposed method is verified in simulation experiment with different noise intensity,and the experimental bench.（2）For the problem of limited denoising ability and separation ability with wavelet transform as the dictionary,and l₁ norm ignoring the relation between the sparse coefficients of wavelet domain and the bearing fault,weighted total variational sparse compound fault diagnosis method based on clustering similarity is proposed.With tunable Q wavelet transform as the dictionary,weighted l₁ norm fused with total variational norm improves denoising ability and separation ability of the sparse model effectively.Similarity clustering is further proposed to measure fault information of sparse coefficients in wavelet domain.Through similarity clustering reflecting fault information of sparse coefficients in wavelet domain,the fault information is reflected in weighted l₁ norm.Finally,the bearing compound fault diagnosis is realized by sparse model fusing total variational term and weighted l₁ norm based on similarity clustering.The proposed method is verified in the simulation experiment under different noise intensity and the experiment platform of bearing compound fault.（3）Aiming at the problem of strong dependence of the dictionary in the existing sparse representation model and how to select parameters of the model,compound fault diagnosis method based on adaptive sparse denoising and time-frequency spectrum weighting is proposed.Firstly,sparse fault value of frequency domain is proposed as the fitness function of the sparrow search algorithm.The sparrow search algorithm determines parameters of the temporal sparse model and denoises the signal adaptively,which identifies the compound fault types effectively.In view of the problem that denoised signal is still disturbed by noise and bearing faults are still mixed together,time-frequency spectrum weighting is further proposed to deal with the denoised signal.Based on the fault characteristics and the distribution nature of the time-frequency spectrum,the noise in the signal is effectively removed and the compound faults of bearing are successfully separated.The effectiveness of the proposed method is verified on bearing compound fault simulation signals with different noise intensity and bearing compound fault signal of experimental platform.（4）The validity of the three algorithms in this paper is verified by the compound fault data of rolling bearing on the low-speed experimental platform.The results of the proposed algorithms are measured by the index in frequency domain and running time of the algorithms,and the performance of the proposed algorithms is discussed and analyzed.