【作者】 林静； 韩玉启； 朱慧明；

【Author】 LIN Jing1, HAN Yu-qi1, ZHU Hui-ming2 (1.School of Economics & Management, Nanjing University of Science & Technology, Nanjing 210094, China; 2.Department of Statistics, Hunan University, Changsha 410079, China)

【机构】 南京理工大学经济管理学院； 湖南大学统计学系；

【摘要】 基于传统竞争失效分析中对截尾数据(包括寿命数据截尾及失效机理信息的删失)、以及环境因子伴随变量的研究不足,提出贝叶斯生存分析理论中一种多重Weibull回归模型;在该模型中假设各失效机理条件下的产品寿命均服从某种Weibull分布,利用该模型能够较为灵活地拟合竞争失效分析中存在数据截尾的情形;针对高维数值积分的不便,讨论了基于Gibbs抽样的MCMC方法,提出通过运用该方法动态模拟出多重Weibull回归模型中相关参数后验分布的马尔可夫链;并通过对模型结构的输入输出分析,给出随机截尾条件下模型参数的贝叶斯估计;利用BUGS软件包进行实例分析的结果,证明了该模型在竞争失效分析中的直观性与有效性。更多还原

【Abstract】 Aiming at the fault of analysis for truncated data (including the truncated data of life time and censored data of causes of failure), as well as the different covariates in traditional competing causes of failure test, a poly-weibull regression model using Bayesian survival analysis theory was introduced, where weibull distributions for each competing causes of failure were supposed . Lifetime data with truncated data can be fitted flexible by using this poly-weibull regression model. Markov chain Monte Carlo method based on Gibbs sampling was discussed, which were used to simulate dynamically the Markov Chain of the parameters’ posterior distribution. Also, the parameters’ Bayesian estimations were given out with model structural analysis for its input and output. What’s more, the results of the data’s simulation were utilized to show the process of setting the model by using the BUGS package. It proves the objectivity and validity of the model.更多还原