Research On Remaining Useful Life Prediction Method Of A Rolling Bearing Based On AdaBoost_RVM

As the key mechanical foundation,the performance of a rolling bearing is directly related to the performance,function and efficiency of the machine.The remaining useful life(RUL)of a rolling bearing is a comprehensive reflection of bearing damage and failure degree during its operation.The accurate prediction of the RUL can provide a basis for preventive maintenance decision,extend the life cycle of the equipment,improve the reliability and utilization rate of the whole machine,and avoid accidents.Data driven method is one of the main methods for predicting the RUL of rolling bearings.Its process mainly includes three parts: data acquisition,feature processing and RUL prediction.Feature processing and RUL prediction are studied in this paper.Feature processing includes three parts: feature extraction,preprocessing and feature selection.In order to extract more comprehensive and sensitive features,the vibration signal is decomposed into a set of mode function components using the variational mode decomposition(VMD).The energy ratio of each mode function is extracted and regarded as the time-frequency domain feature,and together with the time domain features and frequency domain features of the vibration signal,the high dimension original feature set can be constructed.Moreover,in order to achieve the consistency of different bearings features amplitude ranges and the training and testing features normalization,the similarity measure is applied to normalize the original feature set.The experimental results show that the VMD energy ratio of the time-frequency domain features is effective,and the normalization of the features of different bearings can be achieved using the similarity measure.The prediction of RUL includes two parts: the construction of performance degradation trend health index and the prediction of RUL.Aiming at the problems during RUL prediction of rolling bearings,that is,it is not easy to determine the failure threshold and the prediction error is large for the single prediction model,a method based on Adaboost integrated relevance vector machine(RVM)model is proposed.Three assessment indexes of the correlation,monotonicity and robustness are used for selecting the normalized features.The selected features are input into multiple RVMs for building strong prediction model Ada Boost_RVM1 by Ada Boost integration algorithm.And the model is used to construct health indexes and determine the failure threshold.Then,through further training,the model Ada Boost_RVM2 can be obtained,and the future failure time can be predicted and the RUL can be calculated.The experimental results show that,by constructing health indexes,the failure thresholds of different bearings all can be determined to be 1.And compared with the single RVM model,the health indexes and bearing RUL prediction have smaller error and more closer to the true values using the proposed method.