Degradation State Identification And Life Prediction Of Rolling Bearing Based On Differential Symbolic Entropy

Rolling bearing,as one of the important components of rotating mechanical equipment,is widely used in large-scale precision mechanical equipment.The quality of its health seriously affects the operation of mechanical equipment.Most mechanical equipment does not allow frequent shutdowns during operation,so the extent of bearing damage cannot be visually observed by maintenance personnel.In actual operation,slight changes in mechanical system caused by early failure of internal bearings are often neglected,which leads to deterioration of bearing health.Once the rolling bearing is damaged,it will accelerate the early scrapping of other adjacent parts and machines and even lead to safety accidents.Therefore,it has become a precondition to ensure the safe and smooth operation of mechanical equipment to track the change trend of rolling bearing operation state in real time and accurately,to correctly judge the current bearing stage,and to predict the degradation trend of early faults and the remaining useful life.With the support of the national key research and development program and the National Natural Science Foundation of China,the application of differential symbolic entropy in the degradation state identification and life prediction of rolling bearing is carried out.The main research contents and innovative achievements are as follows:(1)To solve the problem that it is difficult to characterize the degradation state of rolling bearings,the method of constructing the performance degradation index of rolling bearings based on nonlinear dynamic entropy is studied.Firstly,the basic theory of differential symbolic entropy(DSE)is proposed,and the influence of parameters on the calculation results is studied through simulation signals.Secondly,through the experimental verification,it is compared with approximate entropy,sample entropy,fuzzy entropy,and dispersion entropy,and the results verify the superiority of differential symbolic entropy in characterizing the running state of rolling bearing.Finally,through the power spectrum analysis of the original vibration signal,the accuracy of differential symbol entropy in detecting the bearing operation stage is proved.(2)Firstly,aiming at the problem that single-scale DSE is not enough to extract the state degradation feature of rolling bearing,a multiscale differential symbolic entropy algorithm is proposed.Then,to avoid the disadvantages of high dimension and information redundancy caused by multiscale,the manifold learning method,namely the laplacian eigenmap method,is introduced.Finally,in the aspect of degradation trend prediction,short-term static prediction,and long-term dynamic prediction are proposed based on long and short term memory neural networks.The experimental results show that the multiscale differential symbolic entropy has better consistency and monotonicity in characterizing the running trend of the bearing life cycle.(3)The single-channel analysis method is easy to be affected by the low signal-to-noise ratio(SNR)signals,and it can not extract enough degradation features.Therefore,a multivariate multiscale differential symbolic entropy algorithm is proposed.Compared with the single-channel analysis method,the proposed method comprehensively considers the correlation between different sampling channels.The simulation results show that the proposed method has a better stability.(4)Due to the vulnerability of the prediction model to low-correlation degraded feature components,the prediction accuracy is insufficient.The attention mechanism theory is introduced and embedded in the long and short memory network.The weighted degenerate feature sets are weighted by the attention mechanism,and the weighted degenerate feature sets are taken as the input of the life prediction model so that the long and short term memory network prediction model based on the attention mechanism is finally obtained.Grid search support vector regression(GS-SVR),particle swarm optimization extreme learning machine(PSO-ELM),and original long short term memory network(LSTM)methods are compared.the results show that the influence of low correlation degradation feature components is effectively avoided and the prediction accuracy of LSTM is improved after feature weighting.In summary,the application of differential sign entropy in the construction of rolling bearing health index,degradation trend prediction,and remaining useful life prediction is mainly studied in this paper,which provides a new means for the performance degradation tracking and residual life prediction of key components of the rotating machinery system.