Vibration-based Fault Feature Extraction And Analysis For Key Components Of Gearbox

As a key part of mechanical equipment to transmit power and motion,the gearbox has been widely used in aerospace,wind turbine,rail transit,automobile,ship,engineering machinery,and many other modern industrial fields.Gearbox fault diagnosis is greatly significant to ensure the safe operation of mechanical equipment,improve industrial production efficiency,and avoid economic losses and catastrophic production accidents.Vibration signals collected from gearboxes are excellent carriers to reflect the gearbox running state and fault information.The vibration-based fault feature extraction is one of the most critical and difficult problems in the field of gearbox fault diagnosis,which is directly related to the accuracy of final fault diagnosis results.However,the gearbox will inevitably be affected by the external random interferences and the internal interferences caused by simultaneous or cascaded multiple faults in the operation process.When the vibration energies of multiple faults are unbalanced,the weak faults are easily submerged by the interference noise and strong faults,resulting in unnecessary missed-diagnosis or misdiagnosis.Therefore,the fault feature extraction of gearboxes from strong noise interferences is a difficult problem in the field of gearbox fault diagnosis,which is also the core problem to be solved in this thesis.Taking into account the above issue,the vibration-based fault feature extraction methods for gearbox key components,i.e.,gears and rolling element bearings,have been proposed based on signal decomposition,signal demodulation and Adaptive Noise Cancellation(ANC)techniques.The main research contents can be summarized as follows:(1)The vibration generation mechanisms and typical fault types of gears and rolling element bearings are analyzed according to their structural characteristics.By establishing mathematical and dynamic models,the vibration characteristics of different fault types are summarized,and the vibration responses of gears and rolling element bearings with local faults are emphatically studied.(2)A gearbox fault test platform is designed and built to simulate the gear root crack and tooth surface spalling.Without dismounting the whole system,single and multiple gear fault vibration tests can be realized,which provide data support for verification of the subsequently proposed fault feature extraction methods.Some existing gear and rolling element bearing fault test rigs are introduced as well.(3)To solve the problem that Variational Mode Decomposition(VMD)based signal decomposition method is prone to mode redundancy,fault feature frequency mixing and missed-diagnosis,an Adaptive VMD(AVMD)method is proposed to extract the fault features of rolling element bearings.Based on the correlation coefficient and the kurtosis of the signal's envelope power spectrum,the Syncretic Impact Index(SII)is constructed to measure the impulsive fault information.After that,by constructing an optimization objective function based on SII,the Artificial Bee Colony(ABC)optimization algorithm is introduced to realize the adaptive extraction on the fault characteristics of rolling element bearings.Compared with existing methods,AVMD has a clear parameter selection basis,which can effectively separate and extract the vibration characteristics of the outer ring and inner ring faults of rolling element bearings under noise interferences,obtaining better results with lower computational cost.(4)To effectively extract the gear fault characteristics from complex gearbox signals containing random shocks and multiple faults,the LEASgram method is proposed.Taking the merits of the logarithm envelope,autocorrelation function and moving average process,a new demodulation spectrum for gear fault signals named logarithm envelope auto-spectrum(LEAS)is proposed,and a measurement index,termed as cyclic frequency index(CFI),is also constructed for calculating impulsive fault information of multi-scale frequency bands.Then,the LEASgram method is proposed for multi-scale demodulation frequency band selection.It can solve misdiagnosis and missed-diagnosis problems when extracting multiple gear fault features by traditional blind gram-based demodulation frequency band selection methods,weaken the interferences of random shocks and strong fault cyclostationary components,and realize the targeted gear multi-fault feature extraction.(5)To accurately extract the gear fault features from strong cyclostationary components without prior fault feature information,a gear fault feature extraction method based on the modified Self-Adaptive Noise Cancellation(MSANC)is proposed.By introducing the adaptive algorithm based on variable convergence factor and the ABC optimization algorithm along with the optimization objective function based on the orthogonality of signal spectrum,the MSANC method is proposed for separating the gear impulsive fault vibration and meshing vibration components.It can overcome the blindness and uncertainty caused by the traditional methods which need to select parameters according to the trial and error method,and can also greatly improve the applicability and convenience of the ANC technology.Furthermore,a gear fault feature enhancement and extraction method is proposed,which combines MSANC with the Fast Spectral Correlation and Multipoint Optimal Minimum Entropy Deconvolution Adjusted algorithms.It can achieve global feature extraction of gear faults under the interferences of strong cyclostationary components and random shocks,without prior fault feature frequency information.