Remaining Life Prediction Of Bearing Based On Degradation Pattern Matching

Rolling bearing is an important component of rotating machinery.Once it breaks down,it may cause serious economic losses,even casualties.Therefore,it is of great significance to accurately predict the remaining useful life of rolling bearing for condition based maintenance of rotating machinery.At present,the key of bearing RUL prediction is to build appropriate degradation index and prediction model.Deep learning is widely used in rolling bearing life prediction based on data-driven method because of its strong ability of feature expression and adaptive extraction,however,most of the degradation index construction methods based on deep learning do not consider the time information of failure degradation process;and ignore the mismatch between the training model and the test bearing caused by the individual differences of the bearing.In life prediction,accurate prediction model and appropriate failure threshold are often difficult to obtain;and the interference of healthy stage data on trend mining in degradation stage is also an important factor restricting the prediction accuracy.Therefore,based on the degradation mode matching,this paper studies the determination of degradation starting point,the construction of degradation index,the setting of failure threshold and the construction of prediction model in the residual life prediction of rolling bearing,the content is as follows:(1)The depth features extraction method of performance degradation based on convolutional Self-organizing Maps and feature reduction method based on feature adaptation analysis are proposed.Firstly,the frequency domain signal of the bearing is mapped into a two-dimensional graph containing complete feature information and time information by using GAF,and then put it into CAE model for adaptive depth feature unsupervised extraction.Aiming at the features of unsupervised extraction is difficult to guarantee its correlation with time,as well as the existence of redundancy,in this paper,the grey contribution degree and its inherent deterioration characteristics of features are comprehensively considered,and the mixed fitness function of features is constructed to select the depth features which are conducive to the description of degradation state.(2)A bearing anomaly detection method based on continuous hidden Markov model optimized by K-means clustering algorithm is proposed.In order to avoid the influence of bearing normal stage characteristics on the trend prediction of bearing degradation stage,K-CHMM is used to detect the abnormal of the optimized degradation characteristics,and find out the fault occurrence time,and the correctness of the FPT is illustrated by envelope spectrum analysis.(3)The method of constructing degradation index based on improved self-organizing map and the method of matching degradation trajectory based on improved dynamic time warping are proposed.Using the feature samples before FPT obtained in(2)as the training sample set of the improved SOM,so as to complete the construction of degradation index.Aiming at the blindness caused by the empirical setting of bearing failure threshold and considering the similarity of degradation process of different bearings.In this paper,DTW is used to realize the similar matching analysis of the degradation track between the test bearing and the whole life bearing,and the degradation index of the whole life bearing in the degradation stage is fitted by the gray forecasting model with full order time power terms,and the failure threshold is obtained,which is used as the failure threshold of all test bearings of the same kind.(4)The RUL prediction based on empirical mode decomposition and bidirectional convolution long-short term memory network model is proposed.In order to alleviate the accuracy limitation of LSTM in the trend prediction of complex time series signals,EMD decomposition is used to extract the components of degradation index in different time scales,and Conv LSTM is used to mine the internal laws of these components,so as to achieve more accurate prediction results.Driven by the historical DI of the test bearing degradation stage,EMD-Bi-Conv LSTM model is used to predict the RUL of the test bearing point by point.The experimental results show that the proposed method not only retains the useful degradation depth trend characteristics of the bearing,but also determines the degradation starting point adaptively,and realizes the independent and reasonable setting of bearing failure threshold on the basis of fully considering the similarity of bearing degradation mode,and the prediction accuracy of bearing RUL is improved by mining multi-scale feature information.