Study On Degradation Monitoring And Remaining Useful Life Prediction For Mechanical Rotary Components

With the rapid development of materials science and ultra-micro manufacturing technology,Complexity,efficiency and intelligence have gradually become the development trend of mechanical equipment.Rotating parts(such as bearings,gears,rotating shafts,etc.)as a key component in mechanical equipment,its performance status directly determines whether the mechanical equipment can operate safely and reliably for a long time.When a mechanical rotating component fails,it will not only cause a lot of economic losses,but also may cause casualties and serious social impact.Carrying out performance degradation tracking and remaining life prediction of mechanical rotating parts is of great significance for ensuring the safety and reliability of the operation of mechanical equipment.In opinion of the shortcomings of the existing methods in terms of nonlinearity,multi-feature degradation tracking ability and fault prediction accuracy,this paper takes mechanical rotating parts as the research object,and completes adaptive noise set empirical mode decomposition,deep learning and traceless Kalman filtering.Based on other theories,research on performance degradation tracking and remaining life prediction methods is carried out.The main research content is divided into the following three points:1.Aiming at the problem that the existing single-feature-based degraded state tracking method is difficult to reveal the nonlinear characteristics of mechanical rotating parts and has poor robustness under noise interference conditions,an empirical mode decomposition based on complete adaptive noise set—approximate entropy(CEEMDAN-Ap En)performance degradation tracking method.The complete adaptive noise set empirical mode decomposition method is used to decompose the original vibration signals of mechanical rotating parts,and the IMF components obtained by the decomposition are screened by the maximum mutual information coefficient to eliminate noise interference under nonlinear conditions;based on the filtered IMF components Calculate the reconstructed signal,calculate its approximate entropy value as the health state factor of mechanical rotating parts,improve the robustness of performance degradation tracking;select the health threshold through Chebyshev inequality,and determine the initial fault time according to the obtained CEEMADN-Ap En degradation curve Points to lay the foundation for accurate prediction of the remaining life of mechanical rotating parts.Experimental results show that the CEEMDAN-Ap En method proposed in this paper can overcome the noise interference and nonlinear problems of vibration signals of mechanical rotating parts,and its performance degradation tracking accuracy is higher than the characteristic quantities such as kurtosis,root mean square,energy moment entropy,and recursive entropy..2.Aiming at the problem that the existing performance degradation tracking method based on multi-features lacks a reasonable index system and cannot accurately characterize the degradation process of mechanical rotating parts,a performance degradation tracking based on convolutional neural network-gated recursive unit(CNN-GRU)is proposed method.Calculate the similarity features by extracting the time and frequency domain parameters of the vibration signals of the mechanical rotating parts,and combine the time and frequency domain parameters to construct a multi-feature set;construct the comprehensive evaluation index of the feature by time correlation,monotonicity,and robustness,and filter It can accurately describe the excellent feature quantity of the performance degradation process;input the filtered feature quantity to the CNN-GRU network,and use the excellent feature extraction ability of the convolutional neural network and the time-correlation portrayal ability of the gated logic unit to achieve supervised learning.The health state factor prediction based on multi-feature fusion improves the accuracy of tracking the performance degradation of mechanical rotating parts.Experimental results show that the CNN-GRU method proposed in this paper can make full use of the advantages of various feature quantities,and its performance degradation tracking accuracy is higher than various time domain,frequency domain and timefrequency domain feature quantities.3.Aiming at the demand for the prediction of the failure of mechanical rotating parts,based on the health state factors based on single feature and multi-feature fusion,a residual life prediction method based on double exponential model-traceless Kalman filter(DEM-UKF)is proposed.The double exponential model is used to describe the performance degradation process of mechanical rotating parts,and the initial value of each parameter of the model is calculated by Dempster-Shafer theory;the advantage of the unscented Kalman filter in solving the problem of particle degradation and particle depletion is to realize the parameters of the double exponential model The updated calculation of the complete accurate prediction of the remaining life of mechanical rotating parts.Experimental results show that the DEMUKF algorithm proposed in this paper can effectively suppress particle degradation and particle depletion,and its remaining life prediction accuracy is higher than that of traditional particle filtering algorithm.