不等定时截尾试验指数分布情形下的可靠性评定  

Reliability evaluation for exponential distribution under non-equal type-I censoring

作　　者：贾祥[1] 王小林 郭波[1] 

机构地区：[1]国防科技大学信息系统与管理学院 [2]78020部队

出　　处：《系统工程与电子技术》2016年第6期1470-1475,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(71371182)资助课题

摘　　要：针对产品的可靠性评估问题,一般需要统计分析寿命试验中收集到的试验数据。假如所有试验样品都在不同时刻终止试验,这种方式就是不等定时截尾寿命试验。针对不等定时截尾数据,现有方法大多只能得到参数的点估计,缺乏针对参数置信区间的估计方法。针对这一问题,在指数分布下,根据样本空间排序法这一理论,推导得到一个比现有方法更简单的计算方法。作为对比,基于极大似然估计的性质和bootstrap方法,又提出两个置信下限的近似求解方法。然后,通过蒙特卡罗仿真实验和一个算例,比较了这3个方法。实验表明基于样本空间排序法的计算结果更优。The problem of reliability evaluation is usually handled based on the sample from the life testing experiment. If all the units in the life test are terminated at different times, it is called the non-equal type-I censoring. According to the non-equal type-I censoring： sample, the existing methods usually estimate the parameter without the confidence interval. To fill this gap, a simpler method is proposed to construct the confidence limit on the basis of the order relations in the sample spaces under the exponential distribution. As comparisons, two approximate methods are also proposed based on the bootstrap method and the property of maximum likelihood estimate. Then, a Monte Carlo simulation and an illustrative example are conducted to compare the results of the three methods, which demonstrate that the new method outperforms the others.

关 键 词：样本空间排序法 指数分布 不等定时截尾 置信限 

分 类 号：TB114.3[理学—概率论与数理统计]