| As an essential part of the transmission system,bearings are characterized with advantages such as low frictional resistance,high mechanical efficiency,high precision and low production costs.Therefore,they are heavily used in automobiles,aircrafts and ships,as well as in micromachinery.However,complex working environment of bearings can impact their expected life due to many external factors,often leading to pitting,flaking and peeling in the inner and outer races in the early stage,after a long period of operation,it will develop into serious failures such as spalling,smearing,scuffing and fracturing,causing bearings to fail,or even major accidents and substantial economic losses.To address the loss of mechanical equipment caused by bearing failures,domestic and foreign researchers used many methods to study bearing failure diagnosis,and contributed more to the research on bearing failure mechanism and dynamics modelling.In many studies,bearing failures are approximated as unitary axial running-through rectangular failure,but for circular failure,existing unitary activation functions have difficulty with expression of actual contact between the rolling element and the failure.It is in desperate need of efforts in mechanism and dynamics modelling of circular bearing failure,to reveal the time-varying activation of the rolling element through the circular failure zone,to identify the dynamics laws of double shock phenomenon of bearing failures,based on which to study the modified multi-network fusion model,so that to realize intelligent recognition of bearing failures,providing solid theoretical foundation and practical support for the recognition of early bearing failures and the extension of diagnosis methods.To study deep groove ball bearings,a time-varying displacement model and timevarying stiffness model for outer circular flaking and peeling failures are constructed.To study the dynamics response of outer circular flaking and peeling failures,a failure dynamics model is established,which is verified with experimental data and theoretical calculation.A neural network is used to classify the intelligent diagnosis methods of bearing failures for healthy monitoring of bearings in the black box with no downtime.Here are the contents of this study:(1)A time-varying displacement model for the rolling element passing through the flaking and peeling in the outer race is constructed to study the varying pattern of the contact gap when the rolling element passes through the circular failure zone in the outer race.According to Hertzian contact theory,a time-varying contact stiffness model for the rolling element passing through the failure zone is established to analyze the loading conditions of the rolling element at the lowest point of the loading area.The two models are combined to reveal the mechanism of the double shock phenomenon of bearing failures,laying the theoretical foundation for the study of dynamics response of bearing failures.(2)With the time-varying displacement and contact stiffness models and the established two-degree-of-freedom(2DOF)bearing failure dynamics model,the characteristics of dynamics response to different sizes of circular flaking and peeling in the outer race and different working conditions such as speed are investigated.The data under different working conditions are processed by the method of wavelet threshold denoising to extract double shock time interval of de-noise signals and failure frequency,which are used to verify the accuracy and validity of the 2DOF baring failure dynamics model through theoretical formulas.(3)The research on the intelligent diagnosis of bearing failures is conducted,the public data from Case Western Reserve University processed through noise reduction are split.The time frequency domain of sample sets is reproduced through generalized S transform as the input of the convolutional neural network.Two bearing failure intelligent diagnosis models with small sample size is constructed based on the Modified Convolutional Neural Network(MCNN)and the Long Short Term Memory Network(LSTM),respectively,to study intelligent diagnosis of bearing failures.By integrating MCNN and LSTM,a MCNN-LSTM model for intelligent diagnosis of bearing failures is established.The superiority of the MCNN-LSTM model is validated by comparing its diagnosis results with those of MCNN,LSTM and other common diagnosis methods. |
