Research On Data-Driven Degradation State Assessment And Remaining Useful Life Prediction For Rolling Bearings

As the core component of rotating machinery,rolling bearings are widely used in important fields such as aerospace,aviation,national defense,infrastructure,and transportation.In these fields,the running state of rolling bearings is directly related to whether complex equipment can work normally.Once the bearing fails,it may cause serious economic losses and casualties,so it is very important to carry out health management research for rolling bearings.As an important part of health management,remaining useful life prediction(RUL)can provide the remaining useful time until the bearing completely fails,which provides an important reference standard for task decision-making and equipment maintenance of the factory.In recent years,with the rapid development of sensor technology and the industrial Internet,the monitoring data of rolling bearings has become easier to obtain.Due to its excellent nonlinear mapping ability,the ability to independently mine data features,and its simple modeling process,artificial intelligence technology has attracted extensive attention from scholars.Therefore,this paper mainly researches the RUL method based on artificial intelligence.However,the current RUL method based on artificial intelligence has some shortcomings,such as the lack of a good early fault identification method,the lack of a unified network modeling theory,and the lack of full-life sample data,which leads to its poor prediction effect and cannot be widely used in engineering.Therefore,this paper researched the early fault feature extraction method of rolling bearings,the deep neural network modeling research,and the whole life data regeneration research,which aims to improve the RUL research system based on artificial intelligence and promote the application of bearing RUL prediction methods in engineering.Among the different studies,the main work of this paper is as follows:(1)In order to extract early fault features for rolling bearings,this paper proposes an adaptive spectrum mode extraction method(ASME),which combines the dispersion entropy index and the envelope spectrum analysis method to extract the early faults of vibration signals.In the ASME method,the minimum bandwidth limit of six times fault characteristic frequency is first fused with the scale space,and the fault feature scale space method is proposed to adaptively determine the mode decomposition layer of the variational mode decomposition method(VMD).Then,according to the different frequency bands divided by the fault feature scale space,a spectral aggregation factor is proposed to adaptively determine the penalty factor of each Intrinsic modal function(IMF)to improve the convergence speed of VMD.Finally,in order to ensure the stability of VMD results,the transboundary criterion is proposed to modify the decomposition parameters of VMD.(2)In order to solve the problem that the RUL method based on artificial intelligence lacks a discrimination mechanism for prediction starting point,first combined with the ASME method,a fault feature gain index is proposed to quantitatively evaluate the damage degree of the bearing.Then,combined with the 3Sigma criterion,the early fault discrimination mechanism is established to identify the initial point of early faults for rolling bearings.Finally,a framework for feature fusion about the time domain,frequency domain,and time-frequency domain of vibration signals is formulated.A Temporal convolutional network(TCN)was constructed to explore the modeling theory of TCN.Compared with the classic long short-term memory network and gated recurrent unit,the advantages of the TCN network in RUL prediction are analyzed.(3)In order to solve the dilemma of the scarcity of full-life samples,a data regeneration framework is proposed to automatically generate full-life samples of rolling bearings with similar full-life trajectories and different local fluctuations.In this framework,a global gain index and a local gain index are first proposed to identify the normal run stages,slow degradation stages,self-healing degradation stages,and accelerated degradation stages for rolling bearings.Then,the full-life sample regeneration rule is defined,including three parts:identity transformation method,the probability distribution of degradation state,and full-life sample regeneration criterion,which serve the process of full-life sample regeneration.Finally,a sample database in different states is established for rolling bearings.In order to solve the problem of unbalanced state samples in the database,a database enhancement framework is proposed to enrich and balance the state sample data,which provides a data source for the regeneration of rolling bearing from run to fail.