Remaining Useful Life Prediction Of Rolling Bearings Under Two Stages Based On Hybrid Filtering

As the key supporting component,rolling bearings have been widely applied in rotating machin ery,and their operation state directly determines the performance of machinery.Due to the harsh environment of high speed and high load for a long time,bearings are prone to damage due to fatigue wear,which further leads to the failure of mechanical equipment.Therefore,if the remaining useful life(RUL)of bearings can be predicted before the failure,better maintenance and replacement strategies can be formulated to provide strong support for the safe and reliable operation of the equipment.Therefore,rolling bearings are taken as the research object in this thesis,and the following researches are carried out around the health index construction,operation stage division and remaining useful life prediction of bearings:(1)In view of the problem of different degradation rates of rolling bearings in normal operation stage and degradation stage,a RUL prediction method is proposed based on stage division and hybrid filtering.Firstly,the kurtosis is determined as the stage division index,and the Kalman filter(KF)model based on linear function is constructed.The sliding window relative error method is combined to detect the segmentation point between the normal operation stage and the degradation stage,that is,the degradation start time(DST)point,so as to realize the operation stage division of the bearing.In the degradation stage,firstly,the complete relative root mean square(CRRMS)index is constructed by complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)to describe the degradation state of bearings.Secondly,a double exponential degradation model is established,and a hybrid filtering(HF)algorithm based on particle filter(PF)is proposed to update model parameters to obtain the CRRMS estimates,and then the RUL prediction of bearings is realized by combining the failure threshold.Finally,simulation experiments are carried out on the PRONOSTIA rolling bearing dataset.The results show that compared with the traditional PF method,the root mean square error of the HF method is reduced by 25.06%on average,and the mean absolute error is reduced by 32.01%on average,which effectively improves the prediction accuracy.(2)Aiming at the limitation of a single feature CRRMS in the characterization of bearing degradation state and the problem that the degradation threshold needs to be set artificially in stage division,a stage division and RUL prediction method based on multi-domain fusion health index is proposed.Firstly,based on the original horizontal vibration signals of bearings,multiple time domain,frequency domain and time-frequency domain features are extracted to construct the candidate feature set.Secondly,a comprehensive index is constructed by monotonicity,robustness and correlation to eliminate redundant features in the candidate feature set,and sensitive features with strong correlation with bearing degradation are screened out.And the kernel principal component analysis(KPCA)is used to integrate the sensitive features to obtain the constructed multi-domain fusion health index(MFHI)to describe the degradation state of bearings more comprehensively.Then,on the basis of MFHI,the box-plot trigger mechanism is proposed to divide the operation stages of bearings.Finally,the RUL prediction results of bearings are given by the above HF algorithm.The accuracy and effectiveness of the proposed method are verified on the rolling bearing data set of PRONOSTIA and Intelligent Manufacturing Systems(IMS)Center of University of Cincinnati.(3)Aiming at the uncertainty of failure threshold setting by artificial experience in the rolling bearing RUL prediction,a RUL prediction method is proposed based on the failure threshold determination under similarity.Firstly,MFHI is constructed based on the above method and the bearing degradation trend is extracted by smoothing method.Secondly,the improved dynamic time warping(IDTW)technique is applied to measure the similarity of the degradation trend between the full-life bearing and the stage-life bearing,and the bearing most similar to the degradation process of the stage-life bearing is selected.Then the GM(1,1)model is utilized to determine the failure threshold of the full-life bearing,which provides a reference for the failure threshold of the stage-life bearing.Finally,the box-plot trigger mechanism and HF algorithm are utilized to predict the final RUL.The simulation results on the PRONOSTIA dataset show that the proposed method can realize the reasonable setting of the failure threshold and obtain more accurate prediction results.