| Rolling bearings as an important part of rotating machinery and equipment,its operating condition is closely related to the overall performance of the equipment,the performance of rolling bearings decline prediction,not only can be repaired to avoid accidents,but also can reduce maintenance costs,to improve the health management level of rotating machinery is of great significance.This paper takes the whole life cycle vibration signal of rolling bearing as the research object,and conducts in-depth research and analysis on three key technical issues,namely,extraction and evaluation of rolling bearing characteristic indexes,classification of degradation process stages and prediction of performance decline,so as to provide technical support for early warning of mechanical equipment failure and formulation of maintenance strategy.The main research contents are as follows.Firstly,through the analysis of rolling bearing failure mechanism,the vibration signal characteristic index is extracted,and the method of dividing rolling bearing degradation process stage is explored.The strategy of whether the characteristic indexes exceed the constant alarm threshold is used to divide the degradation process of rolling bearings into two stages: health and failure,and the critical point of the two stages is identified by the kurtosis characteristic;for the problem that the two-stage division cannot describe the time-varying degradation trend of rolling bearings due to the superposition of different failure forms,the multi-stage division of the degradation process is carried out.By extracting several time-domain characteristic indexes and frequency-domain characteristic indexes from the accelerated life test data of IMS bearings,a multi-stage division of the degradation process is carried out for the indexes that can better reflect the degradation trend of their performance,revealing the entire time-varying degradation trend of rolling bearings,and analyzing the causes of the“running-in” and “self-healing” phenomena of rolling bearings.Secondly,to address the problem that the traditional time domain and frequency domain signal analysis methods based on linear systems are usually difficult to accurately evaluate the rolling bearing operating condition,the multiscale dispersion entropy(MDE)based rolling bearing feature extraction and fixed window mean value based stage division methods are proposed.By extracting the rolling bearing MDE feature indicators,the time domain features,frequency domain features and MDE feature indicators are combined and weighted for evaluation using multiple evaluation criteria.In order to make the characteristic curves have better time correlation and monotonicity,the characteristic index curve with the best comprehensive evaluation result is decomposed into trend and residual curves through fixed window averaging process to realize the classification of rolling bearing performance degradation stages.The validation by public data set shows that the method can identify the degradation points of rolling bearings in advance,which makes the stage of degradation process of rolling bearings more clearly delineated.Again,for the problem that the prediction model based on recurrent neural network(RNN)and its variants can only process the temporal features of data and ignore the spatial correlation between features,a rolling bearing performance degradation prediction method based on temporal graph convolutional neural network(T-GCN)is proposed.A temporal correlation model is constructed by gate recurrent unit(GRU),and a spatial correlation model is constructed by graph convolutional neural network(GCN)and path graph topology,based on which a T-GCN-based rolling bearing performance degradation prediction model is built.Through experimental validation and comparison with the prediction models of GRU and GCN models,the T-GCN prediction curve fit is the best and the evaluation index results are better.Finally,the rolling bearing failure prediction experimental bench is built to obtain and analyze the whole life cycle vibration data of rolling bearings to verify the effectiveness and generalization of the rolling bearing performance decline prediction method proposed in this paper. |
