Research On Low-frequency Fault Feature Extraction For Gearbox Based On Blind Source Separation

Gearbox,as a key part of mechanical equipment to transmit motion and power,is widely used in various huge and heavy mechanical equipments of modern industry,but its working and running condition is usually very poor.Once the unexpectedly failure occurred,the whole machine would not done well,increases equipment maintenance cost,and causes significant economic losses,or even catastrophic accidents.Therefore,the research of gearbox fault diagnosis technology and methods has important academic significance and application value.Fault feature extraction is a key problem for the gearbox condition monitoring and fault diagnosis,especially in the low-speed region,the effective fault information is very weak for the sake of interference from the high frequency gear meshing vibration,transmission path,strong noise,and so on,which brings many problems to the low-frequency fault feature extraction of gearbox.Therefore,It is necessary to find a suitable vibration signal processing technology and method to effectively separate the fault signal and extract the weak fault feature from the complex vibration signal.Considering gearbox steady-rotating working condition,new low-frequency fault feature extraction methods based on blind source separation(BSS)are researched in detail,and its applications for experimental gearbox and mine geabox are investigated.The main contents are as follows:(1)The background and significance of the selected topic are discussed,and the development of gearbox fault diagnosis methods based on time-frequency analysis and blind signal processing is introduced.According to the existed problems analysis of low-frequency fault feature extraction for gearbox based on BSS,the specific reasearch routes and contents of this paper are decided.(2)Taking into account source signal may be directly mixed with source noise,the statistical independence character and nongaussianity will be very weak,it leads to the difficult problem of fault information extraction with BSS.The source noise is added to the linear instantaneous mixture and convoluted mixture model,which is much more suitable for the real gearbox vibration system,and it is the base of the proposed new methods of low-frequency fault feature blind extraction for gearbox with source noise.(3)Constrained independent component analysis(cICA)algorithm has strong denosing ability for measured noise mixed in multi-channel measured signals,but very poor for source signal with source noise.Aiming at this problem,a method of gearbox fault feature extraction based on WT feature-enhanced and cICA is proposed.It canreduce the interference of strong noise and other vibration sources,improve signal-to-noise ratio(SNR),and enhance analysis effect of cICA algorithm.The proposed method is used to the fault diagnosis of test gearbox and mine belt conveyor gearbox,and low-frequency vibration feature is extracted from its vibration signal,respectively.(4)The number of observed signals is no less than that of source signals for cICA algorithm,and it cannot directly extract fault information from the single-channel measured signal.Ensemble empirical mode decomposition(EEMD)can effectively restrains mode aliasing,but with false components.The appropriate intrinsic mode functions(IMFs)are selected by computing the kurtousis and correlation coefficients,then constitute a new observed vector combined with the original signal.Merged the advantages of the two algorithms,a method of gearbox fault feature extraction based on EEMD feature-enhanced and cICA is proposed.By analyzing simulation,experiments and engineering application,the results show that the proposed method is effective for low-frequency fault feature extraction of single-channel measured signal of gearbox.(5)Minimun entropy deconvolution(MED)algorithm is unsuitable for strong noise and outliers,considering this problem,a method of fault feature extraction from gearbox based on maximum correlated kurtosis deconvolution(MCKD)is introduced,which can overcome the shortcomeing of MED algorithm.However,the effect of MCKD algorithm is probably poor according to priori information to select fault period.Therefore,an idea of fault period searching is discussed,the fault period can be limited to a certain range of computation period,and the maximum correlated kurtosis(KC)converges to the global maximum about the optimum fault period with large M and suitable L,and ensures the effective results for MCKD algorithm with different M.The results of weak and low-frequency fault feature extraction throught experiments and engineering applications for the mine gearboxes indicate that the feasibility of optimum fault period searching and the effectiveness and superiority of MCKD method is testified.