Study On Weak Fault Feature Extraction Of Rolling Element Bearing Based On Signal Sparse Representation

As one of the important components in the mechanical transmission system,the rolling bearing plays a critical role to ensure safety of mechanical equipment.The characteristics of rolling bearing vibration signal are non-stationary and non-linear.At the same time,due to the strong background noise in the working environment and the signal attenuation caused by the long transmission path,the fault feature extraction of the rolling bearing is difficult.In addition,there are often many kinds of faults coexisting in the actual operation.The mutual coupling between multiple faults poses a challenge to the accurate diagnosis of rolling bearing faults.Therefore,the study of weak and compound fault diagnosis of rolling bearings is of great significance for adopting active maintenance strategies,low-cost maintenance and avoidance of major accidents.The application of vibration signal processing technology to achieve fault diagnosis of mechanical equipment and crucial components is an important research direction in the field of mechanical fault diagnosis.In the actual engineering application,the fault characteristics of the rolling bearing are often very weak,such as the following three situations: the early fault stage in the whole-life process,signal attenuation caused by long acquisition paths,and compound fault.Traditional signal processing methods are often ineffective in dealing with the above-mentioned problems of complex fault diagnosis.The sparse representation method is a processing method that can effectively match the impact characteristics of rolling bearing faults.It can achieve the capture of the essence of information and the most efficient expression.Based on the classical sparse representation theory,this paper proposes an improvement on the basis of the in-depth study of weak faults and compound fault diagnosis of rolling bearings.The purpose is to diagnose the weak faults and compound faults of rolling bearings and to minimize the occurrence of faults.The main research contents of this article are as follows:1.For existing sparse representation fault diagnosis methods,most of them are based on the prior condition that the coefficient is sparse under the given dictionary transformation of the fault signal.The accuracy of the diagnosis result largely depends on whether the dictionary selection is appropriate or not.An inappropriate dictionary will be directly causes inaccurate diagnosis.A method based on weighted sparse decomposition algorithm is proposed to extract the weak fault feature of rolling bearing.The sparse fault diagnosis model is constructed by using the prior knowledge of the sparse bearing fault signal itself and the periodic impact component in the analysis signal is obtained by solving the model.The optimization objective function contains the fidelity term of the recovery impact feature,in which the high-pass filter H is a zero-phase non-causal second-order Butter worth filter,and the strip matrix form of A and B in the filter can effectively improve the computational efficiency of the algorithm.In order to improve the sparse component recovery ability,the l₁ norm term in the model is weighted to make it closer to the l₀ norm term.The method is verified by simulation and experimental signal of the early failure stage of the rolling bearing.The results show that the proposed method has a good effect in extracting weak fault features.2.When a local failure occurs in the rolling bearing,a periodic impact will occur during the contact between the surface of the fault and the surface of other components,which will excite the high-frequency resonance of the entire system,and a high-frequency resonant band will appear in the frequency spectrum of the signal.A method for early fault diagnosis of rolling bearings based on basic pursuit denosing and adaptive Morlet wavelet filtering was proposed.Firstly,the basic pursuit denosing can be used to remove the invalid frequency components in the frequency domain,which can highlight the resonant frequency band where the fault is located.The adaptive Morlet wavelet filter algorithm is then used to construct a band-pass filter to extract the high-frequency resonant frequency band.The combination of the advantages of both is used to extract weak fault features of rolling bearings.Combining the advantages of the two algorithms can achieve weak fault feature extraction.The shuffled frog leaping algorithm can adaptively acquire the band-pass filtering parameters of the Morlet wavelet.The envelope entropy is used as the fitness function can be a good representation of the periodic impact component in the mixed signal.By analyzing the simulation and experimental signals,the results show that the proposed method can effectively extract the periodic impact feature and achieve the early fault diagnosis of rolling bearings.3.For the existing sparse representation based on binary wavelet transform,the time-frequency resolution of the dictionary is not adjustable,and as the frequency increases and the frequency resolution decreases.A denoising algorithm based on kurtosis-wavelet total variation is proposed to extract the weak fault features of rolling bearings.First,the modified variational term is added to the optimization objective function to effectively remove the spike noise and pseudo-Gibbs generated by the threshold noise reduction.In addition,an overcomplete rational discrete wavelet transform dictionary is introduced,and the time-frequency resolution is adjustable by adjusting the control parameters,so that an appropriate time-frequency division can be selected according to the characteristics of the signal.The kurtosis index is used to select the control parameters and wavelet subbands in the wavelet dictionary to achieve the purpose ofextracting the frequency bands related to the fault features.Then,the algorithm's time-frequency resolution,total variational characteristics,and algorithm complexity are analyzed.Finally,the weak faults in the inner and outer ring of rolling bearing under strong background noise are taken as examples to verify the effectiveness and superiority of the proposed method.4.Aiming at the problem that multiple faults coexist in the actual operation of rolling bearings,and the long signal transmission path causes signal attenuation,a novel fault feature extraction method for rolling bearings based on Morlet wavelet-overlapping group sparsity is proposed.Due to the overlapping of different fault features in the time domain,the wavelet coefficients must also overlap.The overlapping group sparsity model is constructed by using this overlapping property,and the model is solved using a majorization–minimization algorithm.By analyzing the significant features of the signal to adaptively construct the weight coefficients in the model,it is possible to enhance the impact feature in signal.The simulation model and experimental signals show that this algorithm can effectively extract weak compound faults of rolling bearings.5.Because artificially setting the bearing damage can not reflect the early failure state of rolling bearing in the process of natural deterioration,a whole life test rig of rolling bearing is built.It means that the rolling bearing will run from the normal state until its fatigue failure.Without changing the contact fatigue failure mechanism of the rolling bearing,the accelerated full life experiment method is adopted to shorten the experiment time.The experimental data of early failures in the whole life cycle of rolling bearings is used to verify the effectiveness of the research algorithms.The proposed algorithm is compared and discussed in terms of weak fault feature extraction ability and algorithm efficiency.