Research On Working Condition And Life Time Prediction Of Rolling Bearing

Rolling bearing is one of the most used elements of mechanical equipment,which is easily damaged. Failure caused by rolling bearing is a very important reasonwhen mechanical equipments are broken down. Research on the forecasting of t heworking condition and life time of rolling bearing is the key of evaluating the quality,so studies on rolling bearing are very significant.Forecasting of the working condition and life time of rolling bearing is studied inthe paper. LCD (local characteristic scale decomposition) algorithm is adopted forfeature extraction of vibration signal. Gaussian mixture clustering method iscombined with VPMCD (variable predictive model based class discriminate) toforecast the working condition and life time of rolling bearing.The main contents of the paper are as follows:1. The main fault patterns and characteristics of rolling bearing is analyzed, andthe evolution laws of rolling bearing is discussed. Then some methods used forpredicting the working condition and life time of rolling bearing is introduced, then afoundation is laid for research on predicting of the working condition and life time ofrolling bearing.2. Basic theories of LCD algorithm is studied, LCD (local characteristic scaledecomposition) is a new adaptive decomposing method used for non-stationarysignals. LCD can decompose the complex non-stationary signal to several singlecomponent signals whose instantaneous frequency has physical meaning adaptively.Then a comparison is made between LCD, ITD (intrinsic time-scale decomposition)and EMD (empirical mode decomposition) using simulation signals, the effectivenessof LCD when applied to predicting of the working condition and life time of rollingbearing is testified using experiments.3. Basic theories of Gaussian mixture clustering algorithm is studied. The full-lifedata used for recognizing degenerating condition and forecasting life time isnon-stationary and its’ categories can not be ascertained beforehand, therefore theoriginal data should be clustered into N degenerating conditions at first, then usepattern reorganization methods to train and test the full-life data based on thedegenerating conditions established before. Gaussian mixture clustering algorithm iscombined with LCD method, and introduce time factor into Gaussian mixture clustering algorithm as feature vector. Therefore, the classification of the full timedata of rolling bearing is accomplished. Meanwhile, abnormal points of originalsignals can be recognized.4. Basic theories of VPMCD is studied. VPMCD is a new pattern recognitionmethod, VPMCD can first set up variable prediction models based on the innerrelationship between all feature parameters, and then to recognize and classify thetesting data through variable prediction models. VPMCD is combined with LCD to beapplied to research on predicting of the working condition and life time of rollingbearing. Then a comparison is made between VPMCD and neural network which iswidely used to classify the pattern of rolling bearing, results show that VPMCD hasadvantage in classification accuracy and training speed.