Research On Cumulative Features And Integrated Prediction Model Of Rolling Bearing Performance Degradation

Rolling bearing is widely used in rotating machines,its performance degradation assessment and prediction is beneficial to better grasp the actual operation state and avoid accidents.Most of the current degradation assessment and prediction methods are based on data-driven,and the key is to obtain appropriate degradation index and build accurate prediction model.In the process of obtaining the degradation index,some feature evaluation methods which are suitable for fault diagnosis are often used to evaluate the degradation features,while these methods have less consideration for the degradation process itself,so the applicability is not strong;At the same time,the lack of overall consideration of data leads to weaker trend characteristics of the extracted degradation features.In the prediction of degradation tend,it is often difficult to construct an accurate prediction model.Therefore,in this paper,the feature extraction,feature selection and integrated prediction model of rolling bearing performance degradation assessment and prediction are studied as follows:(1)A noise reduction method based on Variational Model Decomposition(VMD)and kurtosis criterion and a degradation feature selection method based on multiple evaluation index are proposed.The VMD method is used to decompose the signal into components of different frequency and then the decomposed components is reconstructed by kurtosis criterion,thus the noise in the original signal is eliminated.Then,the features are extracted from the vibration signal of the rolling bearing,including time domain features,frequency domain features and wavelet packet energy features.Finally,degradation feature selection is carried out by considering the three evaluation indexes of monotonicity,trend and robustness,and the features that are favorable to the description of the degraded state and trend prediction are selected.(2)The construction method of degradation index system is proposed based on cumulative transformation and feature fusion.The traditional feature extraction method processes only a single signal fragment and therefore ignores the relationship between the whole signal;Thus,this paper proposed a cumulative transformation method based on successive stacking principle,from the perspective of bearing damage accumulation.Through the cumulative processing based on the original characteristics,the overall extraction of data information is realized.Finally,based on the cumulative feature,the construction of fusion degradation index system is realized by Principal Component Analysis(PCA).The bearing test data are used to verify that the proposed index system can characterize the bearing degradation process comprehensively and clearly.(3)An integrated prediction model based on Extreme Learning Machine(ELM)and Adaboost iterative algorithm is proposed.Aiming at the problem that the prediction model is difficult to construct in the prediction of degradation trend,the Adaboost-ELM integrated prediction model is proposed.ELM is used to construct the weak learner,and the Adaboost algorithm is used to construct the strong learning instrument for the integrated learning framework.When the degradation index is input as the prediction model,the accuracy of the prediction of the rolling bearing degradation trend is improved.