Research On Weak Fault Feature Extraction And Diagnosis Method Of Planetary Gearbox

Planetary gearboxes are extensively used in large-scale complex mechanical equipment such as wind turbines,helicopters,heavy trucks,stacker-reclaimers,ships and warships.However,due to long-term operation under high load,strong impact,high pollution and variable speed working conditions,the key parts of planetary gearboxes,such as gears and bearings,are vulnerable to be damaged and even failed.Once the gearbox breaks down,it may bring out the deterioration and invalidation of the entire power transmission chain,and even cause catastrophic accidents.Therefore,it is of great significance to develop condition monitoring and fault diagnosis methods of planetary gearboxes to ensure the safe and stable operation of mechanical equipment,reduce economic losses and avoid major accidents.This paper takes the key parts of planetary gearbox as the research object.Firstly,the vibration signal response models are established according to the service characteristics of planetary gearbox to explore its failure mechanism,which provide a theoretical basis for the follow-up fault diagnosis method research.Subsequently,this paper focuses on four aspects:weak fault feature extraction under strong background noise,compound fault decoupling and weak feature enhancement extraction,weak fault feature extraction under time-varying non-stationary conditions,and intelligent fault diagnosis under complex variable conditions and data imbalance.The main work of this paper includes:(1)The structural characteristics and operation manner of planetary gearbox are analyzed,and the failure mechanism of the key parts based on the vibration signal response is studied.Firstly,the fault vibration signal models of sun gear,planetary gear and planetary bearing under steady-state condition are constructed.Based on these models,the analytical expressions of fault signal spectrum and envelope spectrum and the calculation formula of fault characteristic frequency are derived.Then,based on the signal models of steady-state condition,the fault vibration models of key parts under time-varying condition are established,and the characteristic law of fault signals under time-varying condition is given.Finally,through the analysis of planetary gearbox fault experiments under different working conditions,the correctness of fault signal models and failure mechanism research is verified.(2)Aiming at the problem that the early fault features of planetary gearbox are weak and difficult to extract under strong background noise,a weak fault feature extraction method based on improved variational mode decomposition and singular kurtosis difference spectrum is proposed.Firstly,the improved variational mode decomposition is applied to decompose the complex multi-component planetary gearbox fault signal into a series of narrow-band modal components,and the sensitive component is screened out according to the cosine similarity index.Secondly,the Hankel matrix is constructed for the sensitive component in the phase space and singular value decomposition is performed.The effective singular values are adaptively selected by singular kurtosis difference spectrum for signal reconstruction and noise reduction,thereby highlighting the weak fault features.Then,envelope demodulation analysis is carried out on the reconstructed signal to effectively extract the early fault features masked by strong noise.Finally,the effectiveness of the proposed method is verified by simulation signal with strong noise and measured signals.(3)Aiming at the problem that when the planetary gearbox occurs compound fault,different faults are coupled and interfered with each other,difficult to identify and prone to missed diagnosis and misdiagnosis,a compound fault feature extraction based on parallel dual-parameter optimized resonance sparse signal decomposition and multipoint optimal minimum entropy deconvolution adjusted(MOMEDA)is proposed.Firstly,the parallel dual-parameter optimized resonance sparse signal decomposition algorithm effectively decompose the compound fault features into high and low resonance components by constructing the wavelet basis functions matching the faults,thus realizing the nonlinear decoupling and separation of compound fault.Secondly,the MOMEDA is used to deconvolute the high and low resonance components respectively to enhance the weak fault impact characteristics in each component,so as to accurately and comprehensively identify the compound fault features.Finally,the effectiveness and practicability of the proposed method are verified through simulation and practical compound fault examples.(4)Aiming at the problem that it is difficult to accurately extract the fault features of planetary gearbox under time-varying conditions,and the speed measuring device cannot be installed under special circumstances,a time-varying nonstationary fault feature extraction method based on improved adaptive chirp modal decomposition(IACMD)and adaptive maximum second order cyclostationary blind deconvolution(CYCBD)is proposed.Firstly,IACMD is used to decompose the multi-component time-varying nonstationary vibration signal to obtain the modal component containing abundant fault shocks.Secondly,the instantaneous dominant meshing multiply trend line is accurately extracted from the original vibration signal by fast path optimization algorithm and equivalent to the rotating frequency,thereby mapping the non-stationary modal component to the angular domain to eliminate the influences of variable speed.Then,the CYCBD is applied to enhance the weak fault features in the obtained angular domain signal.Finally,the accuracy of the proposed method is verified through single and composite fault simulation signals and actual variable speed fault cases.(5)Aiming at the problem of low recognition accuracy and lack of adaptability of traditional diagnostic models under complex variable conditions and data imbalance,a novel adaptive normalized convolutional neural network model is proposed.Firstly,Teager order spectrum is used as preprocessing to enhance the weak fault characteristics of the signal,and eliminate the influences of variable speed and variable load,thereby improving the quality of sample data.Secondly,batch normalization is introduced to solve the distribution differences caused by the changes of operation conditions and data imbalance.Then,the multi-dimensional hyperparameters of the proposed model are optimized by particle swarm algorithm to improve the self-adaptability of the model.Finally,the proposed model is applied to planetary gearbox fault diagnosis under the scenes of complex variable operating conditions and data imbalance,which can accurately identify different fault types.In addition,compared with other mainstream intelligent diagnosis methods,the proposed method has higher diagnosis accuracy and more reliable stability in the case of highly unbalanced data.Eventually,the research content and innovation of this paper are summarized,and the future research direction is given.