| Reliability assessment is an important part of CNC machine tools reliability engineering technology, that also is the indispensable key to develop machine tool manufacturing industry. With machine tools’ increasingly complication, informatization and integration, high-grade CNC machine tools have the characteristics of small sample size and less failure data, which lead to the result that classical statistical theory based on the law of large numbers can’t effectively assess its reliability. Bayes method comprehensivly taking advantage of prior information and field test data provides a very important theory support for small sample machine tool reliability assessment. How to establish appropriate accurate prior distribution is the key to the Bayes method reasonable application.In this thesis, with small sample CXK5463 lathe-mill machining center as the research object, to reduce the great sampling error and deal with too much subjectivity problem of assuming distribution types during the process using traditional Bootstrap to confirm prior distribution upon similar machine tools data for Bayes reliability assessment, an establishing prior distribution method based on Least squares support vector machine(Lssvm), parametric Bootstrap and kernel density estimation is proposed. Utilize weibull distribution outliers test method to calculate the credibility value of the established prior distribution. According to the credibility value, the established prior distribution is revised and obtain the final trusted prior distribution.On this basis, implement Bayes reliability assessment.Firstly, with the failure data of CXK5463 lathe-mill machining center as field test information and the similar CNC machine tool CK5250 failure data as priori information, the compatibility test between the two kinds of information was performed. Analyze the error existing in the process using traditional Bootstrap to confirm prior distribution. Use Lssvm method to compute weibull parameters of the prior information, and then parametric Bootstrap is implemented, which has reduced the sampling error. According to the sampling results, the probability density of the scale parameter α and shape parameter m is estimated based on kernel density estimation method. In this way, fitting prior distribution directly from the characteristics of the sample data makes the fitting result and the actual distribution of more in line with, which has solved the too much subjectivity problem of assuming distribution types.Secondly, analyze the defects of the existing prior information credibility value calculation methods. Aiming at the reliability estimate situation, in which, prior information is prior distribution and field information is machine tools failure data obeyed weibull distribution, calculate the credibility value of prior information based on weibull distribution outliers test method and realize the algorithm calculation by Matlab. For CXK5463 prior distribution as an example, the simulation is carried on and get its credibility value. The simulation result compares with the credibility value of prior information fitting by traditional Bootstrap method, which shows that the prior distribution obtained by the method proposed in this paper has higher credibility. The CXK5463 prior distribution is revised according to its credibility value and then obtain the final trusted prior distribution.Finally, the posterior distribution of machining center is simulated using the Open BUGS software with Markov Chain Monte Carlo method, and the weibull distribution parameters estimation value of CXK5463 Mean Time Between Failures is obtained. On this basis, the estimation value of CXK5463 MTBF is also obtained. The superiority and accuracy of method in this paper is verified by comparing the result obtained through this method with the Bayes assessment result computed upon traditional Bootstrap establishing prior distribution. |
PDF Full Text Request