| Rolling bearings,as key elements of rotating machinery,have been widely applied in various industries.Once a failure occurs,it will have a direct impact on the performance of the entire machine system,and even cause huge losses in terms of human life or property safety.Given the long service lifetime and poor working environment,the degradation process of bearings is inevitable.Thus,bearing health information is of great importance for subsequent intelligent maintenance and loss reduction.In this thesis,the core of rolling-bearing health management,i.e.,condition monitoring,and remaining useful life(RUL)prediction,is studied,and the main researches are as follows:(1)An initial analysis was made of the types of bearing faults and the characteristics of the vibration signals.Then,the general evolution law of failure was elaborated,and the main purpose of rolling-bearing health management was described.In the following,the basic theory of the prognostics and health management technology was briefly outlined and its technical framework was introduced.Next,the challenges of health management were identified,which are the construction of a strong robust health index and the highly sensitive detection of the first point of change.All of these works provide the theoretical foundation for the following research.(2)In this thesis,the Reinforced Noise Resistant Correlation(RE-NRC)method is proposed for bearing condition monitoring.This method constructs a bearing condition index with high robustness to strong noise.Then,it is theoretically analyzed that the mathematical expectation of the index is equal to the mathematical expectation of background noise when the bearing is in the normal condition,and there are local extreme points in the abnormal state.Afterwards,the mathematical relationship between the local extreme point and the fault characteristic period was derived,and the Cumulative Sum control strategy was used bearing condition monitoring.Finally,verification was carried out using the datasets from the simulation and the University of Cincinnati.Comparison results with classical health index construction methods and fault diagnosis methods were used to further reveal the superiority of the proposed method.(3)For RUL prediction,an adaptive RUL prediction method based on Gaussian Process Regression(GPR)was proposed in this thesis.This method first uses the shift of the health index to characterize the degradation process.The Exponentially Weighted Moving Average control strategy was combined to improve the sensitivity to incipient faults and complete the dynamic first change point detection,thus ensuring the initial fault information acquisition.Then,to screen the optimal features,a hierarchical clustering feature optimisation strategy based on trendability measurement was developed.A robust virtual health index construction method was proposed next to ensure the trendability and robustness of bearing individual differences.Finally,the GPR model was used to predict the RUL.The rolling bearing accelerated degradation datasets from Shengyang Technology and Xi’an Jiaotong University was used to verify the feasibility of the proposed method.Meanwhile,the proposed first change point detection strategy and feature fusion strategy were compared with classical methods to further verify their superiority.In summary,for two cores(condition monitoring and RUL prediction)in rolling bearing health management,this thesis designed a condition monitoring method based on the RE-NRC and an RUL prediction method based on the GPR.It is confirmed that the proposed method is able to detect the fault period of the bearing,to monitor the running condition and to predict the RUL,which is the basis for the health management of rolling bearings. |
