Research Of Bearing Vibration Analysis And Life Evaluation Methods

With the rapid development of automation and industrialization, the requirementsof mechanical equipment’s stability and life time have also increasing quickly. For thispurpose, the effective maintenance strategy should be planned by the correct judgmenton equipment’s station. As an important part of mechanical equipment, bearing take theresponse of lubrication and regulating the shaft. But it also is a vulnerable component ofthe equipment. So, to extend the lifetime of the equipment and maintain its reliability,it’s necessary to have a correct diagnosis and state evaluation on bearing fault.In order to diagnose and evaluate the bearing’s fault and state, a series of researchhas been carried out in this paper. Mainly includes the following aspects:(1). After reviewing the knowlage of bearing’s strucutre and failure, introduced thenormal methods on bearing fault diagnosis and condition assessment, which includingtime domain and frequency domain methods. Then, discuss the use in fault diagnosis ofShannon entropy and signal-to-noise ratio.(2). Then introduced the principle of support vector machine (SVM) and theapplication in multi-classification system. Afiter discussing the voting mechanism inSVM, the one-to one vote was improved in this paper. Basis on this, the appropriate tagvalue from the experimental data was discussed and been used in the SVM for gettingthe condition assessment of the test data.(3). For the inaccuracies of machine learning classcation, an optimization methodwere proposed. And this method was used on the experimental data; the results showthat the proposed optimization method can improve the accuracy of the bearingcondition assessment effectively.(4). After the principle of continuous wavelet analysis and its advantages in bearingfault diagnosis were introduced, the marlet wavelet filter family was constructed forbearing fault diagnois. Then, Autocorrelation function indicator was introduced to selectthe appropriate the coefficient of morlet wavelet filter to extract the bearing cyclecomponents of vibration signal. Compared this method with the tradional mathods, likeHilbert transform and discrete wavelet transform, this method can be more effective todetect early fault information from the application in experimental data. (5). Finally, a brief introduction to the origin and theoretical stochastic resonancewas given, to discuss how to apply it into early fault dignosis of rolling bearings. Afterdiscussed the parameters of numerical calculation in stochastic resonance, this methodwas be used on the bearing experimental data. Forther more, either the stochasticresonance or the compine of stochastic resonance and method in (4) can get a goodresult.This thesis is the discussion and exploration of signal pricessing methods for earlybearing fault diagnosis.