基于截尾样本下部件相依竞争失效的k/n系统可靠性分析  

Reliability analysis for k/n system with dependence competing failure units under pprogressive censoring

作　　者：时炎杰 彭秀云[1] SHI Yanjie;PENG Xiuyun(School of Science,Inner Mongolia University of Technology,Hohhot 010051,China)

机构地区：[1]内蒙古工业大学理学院,呼和浩特010051

出　　处：《内蒙古工业大学学报（自然科学版）》2022年第5期400-406,共7页Journal of Inner Mongolia University of Technology：Natural Science Edition

基　　金：国家自然科学基金项目(11861049);内蒙古自然科学基金项目(2021LHMS01001);内蒙古自治区生命数据统计分析理论与神经网络建模重点实验室基金资助项目。

摘　　要：研究k/n系统部件由两个相依竞争失效机理组成的可靠性问题,相依失效机理对应的失效时间由Gumbel-Hougaard Copula函数连接,边缘分布服从形状参数和尺度参数均不相同的Weibull分布。基于截尾样本下建立k/n系统部件失效的模型,证明了边缘分布形状参数的后验密度函数具有对数凹性,基于此性质提出了Metropolitan-Hastings(MH)抽样混合Gibbs抽样的Bayes估计方法。采用Inference for the margins(IFM)方法对参数进行估计。通过Monte Carlo实验比较两种方法的可行性和有效性。结果表明Bayes估计结果优于IFM估计结果。最后,将构建的模型运用到一个真实数据中。The reliability of the k/n system with two dependent competing failure components whose lifetime follows Weibull distributions is studied.The competitive correlation is connected by the Gumbel-Hougaard Copula distribution.The failure model of k/n system components is established based on censored data,and it is proved that the shape parameters of the Weibull marginal distributions are log-concave.On the basis of this,a hybrid Markov Chain Monte Carlo(MCMC)technique combined by Gibbs with Metropolitan-Hastings(MH)sampling algorithm is proposed to obtain Bayes estimation.The estimations of the parameters are also obtained by inference for margins(IFM)estimation method.Both methods are evaluated by Monte Carlo(MC)simulation.The simulation results show that the Bayes estimation is better than IFM method.Finally,the establised model is applied into real data.

关 键 词：k/n系统 Gumbel-Hougaard Copula WEIBULL分布 BAYES估计 逐步截尾 

分 类 号：O212.5[理学—概率论与数理统计]