Research On Performance Degradation Assessment And Remaining Useful Life Prediction Methods Of Rolling Bearings

Rolling bearing is a vital rotating component which is widely used in mechanical equipment.It is the main source of mechanical failure since it can be easily damaged.Therefore,mastering the performance degradation assessment and remaining useful life prediction technology of rolling bearings becomes an imperative requirement for the industry.This paper takes rolling bearings as research object and carries out thorough researches of four aspects,including bearing degradation mechanism analysis,vibration feature extraction,bearing performance degradation assessment and bearing remaining useful life prediction.The main contents of this paper are as follows:Aiming at the problem that the monitoring signals of rolling bearing are difficult to express the performance degradation characteristics of bearings intuitively,we analyze the bearing degradation mechanism,and describe the condition monitoring / acquisition method of bearings briefly,which is a preparatory preparation for the research of bearing vibration feature extraction.Aiming at the problem of extracting effective features from the original vibration signals in the task of bearing performance degradation assessment and prediction,we study the vibration feature extraction method of bearing with convolutional sparse coding.We propose the feature extraction method which bases on "using the ADMM to optimizing the sparse convolutional encoding with learning combination".According to the above method,we can extract a feature set composed of RMS,kurtosis and weighted sparse coefficients of the reconstructed error signal.Finally,the validity of the method is verified by the experimental vibration signals of IMS bearings.Aiming at the problem of bearing performance degradation assessment methods,there are two problems,about the construction of degradation indicator of bearings and the setting of the indicator threshold,which are difficult.We study the method of constructing the indicator based on weighted fusing the information of the feature set and the method of setting the indicator threshold based on the confidence interval,respectively.Through the analysis of traditional characteristic indicator's curve and the intuitive analysis of time domain vibration signal,it is pointed out that the traditional characteristic indicator and the intuitive time domain analysis are both difficult to use for evaluating the degradation degree of bearings.This paper propose an method of bearing performance degradation assessment which bases on "combination with three parties including the construction of the degradation indicator through PCA fusing features' information,the threshold setting of the Chebyshev inequality and indicator smoothing of the Exponentially Weighted Moving-Average method ".The validity of the degradation assessment method is verified by a standard IMS bearing full life data set.Aiming at the case of medium-or-small sample,both the traditional life prediction method and Artificial Neural Network based method cannot effectively predict the remaining life of the bearing.We investigate the prediction method of remaining useful life of rolling bearing with Support Vector Regression(SVR)as the main body.This paper implements bearing remaining useful life prediction method which bases on "the combination of Phase Space Reconstruction,Fruit-Fly optimization algorithm and SVR".The validity of the bearing life prediction method is verified by the full life data set from IMS.All of the above methods have been validated on a standard IMS bearing full life data set.