Life And Reliability Evaluation Of Rolling Bearings For No-failure Data

With the rolling bearing manufacturing level improving, the life of rollingbearings has also been greatly extended. Among the processes of rolling bearingreliability test, there will be a large number of failure datas. Especially in thereliability test for some high-cost, high reliability rolling bearings such as windpower bearing, high-speed train bearing, the small sample size, no-failure censoredtest will be choosed generally. In this case, due to the lack of failure information, thetraditional evaluation methods, such as the best linear unbiased estimation providedby GB24607-2009, cannot provide proper evaluation for the reliability of rollingbearings. To solve these problems, a new evaluation method must be studied. In thisthesis, on the basis of previous studies, after making the thorough theory andsimulation studies for the above problem applying to Bayes method, two kinds ofreliability evaluation method suitable for no-failure datas and its model of the rollingbearing were presented.For the First chapter, the source of the subject was introduced and thesignificance of the rolling bearing life testing and reliability evaluation wasexpounded. Rolling bearing life distribution-Weibull distribution parameters andreliability index were discussed. On this basis of above, the main contents of thisthesis were given.The second chapter, the rolling bearing definition of life, experimental principle,life calculation model were expounds in the second chapter, then the basic theoryand method of the rolling bearing reliability evaluation were discussed. Therelationship among the reliability, failure probability, failure rate, MTTF (MeanTime To Failure) and other indicators of the Weibull distribution model was alsodiscussed. Finally, the geometric meaning and physical meaning of Weibulldistribution shape parameter and Bayes theory and Bayesian statistical model werediscused one by one.The third chapter, the evaluation method and model builded by virtualinformation were put forward the virtual failure information in the former censored time point was introduced to the process of reliability estimate in each censored timepoint, which made the reliability of the censored time point estimation highercredibility and better robustness. Two parameters of Weibull distribution wereobtained by the weighted least squares method after completing reliability estimateof each censored time point. The example shows that the proposed method has betterrobustness than other methods when the hyper-parameters of the prior distribution ofreliability changes in certain intervalThe fourth chapter, the new model and method were put forward in the fourthchapter, in which the distribution is fitted from the historical data of Weibulldistribution shape parameter was used as prior information. According to the historydatas of the shape parameter test, the probability distribution of the shape parameterfitted from historical data was used as a priori information. Converting the Weibulldistribution to exponential distribution, the priori information of the failure rate ofexponential distribution was constructed according to the principle of conjugateprior distribution. Then taking failure rate and shape parameter as the entry point,combined with no-failure test datas, the Bayes estimation of failure rate and theshape parameters was drawn, and then the life estimation of Weibull distributionfeatures was calculate. At last, through a set of example the accuracy of theestimation results was verified, and the estimate of the robustness was discussed.The fifth chapter, based on Matlab GUI, the rolling bearing reliabilityevaluation software were designed and carried out in the fifth chapter. The methodof this thesis to evaluate the reliability of rolling bearings was used in thesoftware.The evaluation system includes the Bayes estimation module, the studymodule on Bayes estimation stability, the fitting module for prior distribution of theshape parameter three parts.The first part includs the Bayes estimation method in thethird chapter, Shi-Song Mao’s estimation method, Lai-Lin Wu’s s estimationmethod and two other methods in the fourth chapter in which prior distribution ofshape parameter is taked as uniform distribution and Weibull distributionrespectively. The second part mainly includes study on stability for five estimationmethods in the first part. The method in the third part can judge which distributionamong Weibull distribution, normal distribution, and exponential distribution is bestdistribution for shape parameter from input data firstly, then it fits the best distribution according to the input dataFor the Chapter6, the full text of the research methods and research results weresummarized and the future research direction and the deficiency of this study wereproposed.