Research On Characteristics Extraction Of Vibration Signal And Fault Diagnosis For Low-noise Deep Groove Ball Bearing

As a kind of widely used joint in mechanical equipments,the vibration characteristics of rolling bearings will have a great influence on the mainframe equipment.China is a great power in bearing production.More than 70% of the total output of bearings is deep groove ball bearing,in which the low noise deep groove ball bearings are increasingly used in many fields because of its good performance,and the vibration performance requirements of low noise deep groove ball bearing are higher.In this thesis,the vibration measurement and the defect diagnosis in the production process of low noise deep groove ball bearing are taken as the application background.The key issues such as vibration signal acquisition,signal denosing,characteristics extraction,defect diagnosis and classification are studied.The researches can also be used for reference in the defect diagnosis of other types of bearings.The main research contents of the thesis are as follows:The surface defects of deep groove ball bearings in the production process are often tiny and the corresponding vibration characteristics are easy to be seriously interfered by the bearing intrinsic vibration components.For this problem,an optimized ensemble empirical mode decomposition(EEMD)denoising algorithm based on the suppression of the intrinsic vibration components(OEDSIV)was proposed.The vibration signal was decomposed twice by empirical mode decomposition(EMD).After the first EMD,the intrinsic mode function(IMF)component with the maximal kurtosis was selected.The zero mean normalization was carried out for the IMF,and then the vibration components whose amplitude lain within plus or minus twice the effective value were removed.The white noise with limited amplitude was added into the IMF.Afterwards,the noise-added IMF was secondly decomposed by EMD,and the IMF with the highest frequency band was directly selected from the decomposition results.The experimental results showed that the OEDSIV algorithm could effectively suppress the interference of the intrinsic vibration components.After the proposed noise reduction,the kurtosis values of the actual inner raceway defect signal,outer raceway defect signal and rolling element defect signal increased respectively from 4.49,6.82 and 6.44 to 46.31,55.27 and 135.79.The proposed OEDSIV algorithm highlighted the weak defect characteristics in the original vibration signal.Compared with the EEMD algorithm,the noise reduction algorithm proposed in this chapter only performed two EMD operations,and the time consumption was reduced from 119.23 s of EEMD to 1.05 s.To further highlight the vibration characteristics of minor defects,a defect characteristics extraction algorithm based on parameter-adaptive variational mode decomposition(VMD)algorithm(PAVMD)was proposed.Two parameters of VMD algorithm,i.e.the band-limited intrinsic mode function(BIMF)number M and the quadratic penalty term ?,will have an important impact on the decomposition results of VMD.Meanwhile,the two parameters need to be determined in advance.In order to determine the reasonable M and ?,the defect energy ratio R was proposed,which could reflect the energy magnitude and frequency information of the impact impulses.Then,the multi-beetle antennae search(MBAS)algorithm was developed.According to the prior knowledge,M was assigned as an integer between 3 and 8,and min{1/ R} was set as the optimization goal.Several beetles were employed to independently optimize ? with a fixed M.After the optimization iterations were finished,the global optimal combination of M and ? could be selected from the local optimization results of all the beetles.The global optimal combination was adopted to decompose the vibration signal via VMD.And the BIMF component with the largest defect energy ratio R was extracted as the sensitive one.The simulation signal experiments showed that the PAVMD algorithm proposed in this chapter outperformed the complete EEMD with adaptive noise(CEEMDAN)and fixed-parameter VMD.After processed by the proposed PAVMD algorithm,the defect energy ratio values of the actual inner raceway defect signal,outer raceway defect signal and rolling element defect signal were respectively 0.690,1.822 and 1.284;after processed by CEEMDAN,the corresponding defect energy ratio values were respectively 0.550,1.723 and 1.212;after processed by fixed-parameter VMD,the corresponding defect energy ratio values were respectively 0.227,0.528 and 0.571.The contrast experimental results demonstrated that the proposed PAVMD algorithm was superior to CEEMDAN and fixed-parameter VMD.Meanwhile,the proposed PAVMD algorithm displayed the performance of much less mode mixing than that of CEEMDAN.After extracting the characteristics in the bearing vibration signals,for the subsequent defect diagnosis and classification problems,a defect diagnosis and classification algorithm based on mixed domain characteristics was proposed on the basis of integrating the OEDSIV algorithm and the PAVMD algorithm.Firstly,the PAVMD algorithm was used to extract the sensitive BIMF component from the vibration signal,and then the OEDSIV algorithm was referred to suppress the intrinsic vibration components as much as possible.Afterwards,the PAVMD algorithm was repeatedly processed to obtain the final sensitive BIMF which could demonstrate the prominent defect characteristics.As a result,the difficulty of defect identification and classification could be reduced.Several time-frequency characteristic parameters of the BIMF component were calculated,and a high-dimensional defect characteristic vector could be formed.The principal components analysis(PCA)algorithm is adopted to reduce the dimension of the vector.When the expected value of cumulative percent variance(CPV)was set to 90%,the characteristic parameters after dimensionality reduction covered defect energy ratio,sample entropy,kurtosis and effective value.The defect energy ratio included three parameters: inner raceway defect energy ratio,outer raceway defect energy ratio and rolling element defect energy ratio.After training the extreme learning machine(ELM)algorithm with the prior information,the defect diagnosis and classification of bearing could be realized by a trained ELM classifier.Experiments were separately carried out on the vibration signals from Case Western Reserve University and the field measurements.The results showed that the diagnostic accuracy of the former was 99.75%,and that of the latter was 94.17%.In order to verify the above theoretical research algorithms,following the requirements of vibration measurement and defect diagnosis in the production process of low-noise deep groove ball bearings,an Automatic Bearing Vibration Tester(ABVT)was developed.The ABVT was debugged on the production line of low-noise deep groove ball bearing and the defect diagnosis algorithms were verified.