截尾数据下ZZ分布的参数估计
2020-05-21张国志张宁
张国志 张宁
摘 要:在可靠性统计中,寿命分布通常有4种:指数分布、韦布尔分布、极值分布、对数正态分布。但在有些实际问题中,一些元件的寿命用上述4种分布来刻画往往与实际相差甚远,这说明该元件寿命分布并不属于熟知的这4种分布。由此给出了一种新型ZZ分布,该分布较好地刻画了这类元件的寿命,并在截尾数据下对参数作出最佳线性无偏估计与简单线性无偏估计。为了便于使用附录中给出了估计所需要的数表。
关键词:ZZ分布;截尾数据;参数估计;简单线性无偏估计;最佳线性无偏估计
DOI:10.15938/j.jhust.2020.01.023
中图分类号: O231
文献标志码: A
文章编号: 1007-2683(2020)01-0149-05
Abstract:In reliability statistics, there are usually four types of life distributions: exponential distribution, Weibull distribution, extreme-value distribution and log-normal distribution. But in some real life applications, the results of statistical inference were found to be far from the truth if using the four life distributions to describe the life of some components. It shows that the component life distribution is not anyone of the four distributions. This paper proposes a new type of life distribution called ZZ distribution. This distribution is a better description of the life for such components. And based on censored data, the best linear unbiased estimation and the simple linear unbiased estimation of parameters were given. The required tables were given in the appendix.
Keywords:ZZ distribution; censored data; parameter estimation; simple linear unbiased estimation; best linear unbiased estimation
0 引 言
众所周知常见的寿命分布有4种,指数分布、韋布尔分布、极值分布、对数正态分布。很多元件的寿命分布属于这四种分布,因此在各种样本形式下,关于这四种分布的可靠性统计推断的研究非常之多。比如:Lee J. Bain[1]给出了截尾样本下极值分布尺度参数的简单线性无偏估计,在此基础上Max Engelhardt和Lee J. Bain[2]考虑了极值分布次序统计量的影响,给出了更精确的简单线性无偏估计,并与此样本下的极大似然估计进行了对比。W. J. Szajnowski[3]研究了对数正态分布参数估计量,并用蒙特卡洛法模拟检验其合理性。A.Clifford Cohen和Betty Jones Whitten[4]对三参数对数正态分布的参数极大似然估计和矩估计做了修正。N.R. Farnum和P. Booth[5]给出了两参数韦布尔分布的参数极大似然估计并对其唯一性进行了验证。Zheng R[6]研究了三参数韦布尔分布参数估计方法并探究了其在可靠性分析中的应用。Davies I J[7]用最小二乘法估计了三参数韦布尔分布的尺度参数。J. William Shelnutt,Albert H. Moore和H. Leon Harter[8]研究了韦布尔分布线性估计。对于存储产品,早期失效数据对寿命分布有较大影响,K. Muralidharan和P. Lathika[9]对韦布尔分布的早期失效数据做了分析。Wayne Nelson[10]研究了步进应力加速寿命试验模型。N. Balakrishnan,Qihao Xie和D. Kundu[11]研究了指数分布在定时截尾数据形式下的简单步加模型。C. Xiong[12]研究了简单步加模型在II型数据截尾下参数统计推断方法。在对参数做出估计后,对这些估计的分布进行研究非常必要。Thoman, D. R. ,L. J. Bain和C. E. Antle[13]研究了韦布尔分布参数在截尾数据形式下的简单线性估计及极大似然估计的近似分布推断方法。对参数近似分布及置信限[14]的研究也为参数置信区间的确定[15-17]奠定了基础。
以上的研究工作都是基于常见的4种分布展开的,然而有些存储产品,在厂家给出的设计寿命之前几乎很少失效,过了设计寿命之后失效的比例大幅增加。这时根据样本对寿命分布进行检验,往往做出的推断是:或者拒绝这四种分布;或者接受某些分布,但统计推断的结论与实际相差甚远。因此,需要寻找一种新的寿命分布来刻画这类元件寿命。下面给出的例子就是一个实际遇到的问题。对于这个实际问题,自然希望:①新分布的检验获得通过;②新分布下平均寿命的估计接近实际。
4 结 论
本文针对可靠性统计中常见的四种寿命分布使用时具有局限性的问题,提出了新型ZZ分布刻画一类元件存储寿命的理论和方法,得到了预期的效果,在一定程度上丰富了寿命分布的研究内容。在此基础上给出了ZZ分布参数的简单线性无偏估计与最佳线性无偏估计。
参 考 文 献:
[1] LEE J. BAIN. Inferences Based on Censored Sampling From the Weibull or Extreme-Value distribution[J]. Technometrics, 1972, 14(3):693.
[2] MAX ENGELHARDT, LEE J. BAIN. Some Complete and Censored Sampling Results for the Weibull or Extreme-Value Distribution[J]. Technometrics, 1973, 15(3):541.
[3] SZAJNOWSKI W J. Estimators of Log-Normal Distribution Parameters[J]. IEEE Transactions on Aerospace and Electronic Systems, 1977, AES-13(5):533.
[4] CLIFFORD COHEN A, BETTY JONES WHITTEN. Estimation in the Three-Parameter Log-normal Distribution[J]. Journal of the American Statistical Association, 1980, 75(370):399.
[5] FARNUM N R, BOOTH P. Uniqueness of Maximum Likelihood Estimators for the 2-parameter Weibull Distribution[J]. IEEE Transactions on Reliability, 1997, 46(4):523.
[6] ZHENG R. Parameter Estimation of Three-parameter Weibull Distribution and Its Application in Reliability Analysis[J]. Zhendong Yu Chongji/Journal of Vibration & Shock, 2015, 34(5):78.
[7] DAVIES I J. Unbiased Estimation of the Weibull Scale Parameter Using Linear Least Squares Analysis[J]. Journal of the European Ceramic Society, 2017, 37(8):2973.
[8] WILLIAM SHELNUTT J, ALBERT H. MOORE, LEON HARTER H. Linear Estimation of the Scale Parameter of the First Asymptotic Distribution of Extreme Values[J]. IEEE Transactions on Reliability, 1973, 22(5):259.
[9] K. MURALIDHARAN, P. LATHIKA. Analysis of Instantaneous and Early Failures in Weibull Distribution[J]. Metrika, 2006, 64(3):305.
[10]WAYNE NELSON. Accelerated Life Testing-Step-Stress Models and Data Analyses[J]. IEEE Transactions on Reliability, 1980, 29(2):103.
[11]BALAKRISHNAN N, XIE QIHAO, KUNDU D. Exact Inference for a Simple Step-stress-model from the Exponential Distribution Under Time Constraint[J]. Annals of the Institute of Statistical Mathematics, 2009, 61(1):251.
[12]XIONG C. Inferences on a Simple Step-stress Model with Type-II Censored Exponential Data[J]. IEEE Transactions on Reliability, 1998, 47(2):142.
[13]THOMAN D R, BAIN L J, ANTLE C E. Inferences on the Parameters of the Weibull Distribution[J]. Technometrics, 1969, 11:445.
[14]MANN N R, FERTIG K W. Tables for Obtaining Conference Bounds and Tolerance Bounds Based on Best Linear Invariant Estimates of Parameters of the Extreme-value Distribution[J]. Technometrics, 1973, 15:87.
[15]BAIN L J, MAX ENGELHARDT. Simple Approximate Distribution Results for Confidence and Tolerance Limits for the Weibull Distribution Based on Maximum Likelihood Estimators[J]. Technometrics, 1981, 23(1):15.
[16]LAWLESS J F. Conditional Versus Unconditional Conference Intervals for the Parameters of the Weibull Distribution[J]. Technometrics, 1975, 68:665.
[17]MANN N R, FERTIG K W. Simplified Efficient Point and Interval Estimators for Weibull Parameters[J]. Technometrics, 1975, 17:361.
[18]張尧庭, 刘鸿翔. 可靠性统计分析中“倒挂”数据的处理方法[J]. 应用数学, 1991, 4(4):23.ZHANG Yaoting, LIU Hongxiang. Disordered Data Analysis in the Reliability Statistics[J]. Mathematica Applicata, 1991, 4(4):23.
[19]杨振海, 安保社. 基于成败型不完全数据的参数估计[J]. 应用概率统计, 1994, 10(2):142.YANG Zhenhai, AN Baoshe. Parameter Estimation Based On Binary Data[J]. Chinese Journal of Applied Probability and Statistics, 1994, 10(2):142.
[20]茆诗松, 王玲玲. 可靠性统计[M]. 上海:华东师范大学出版社, 1989:141.
(编辑:王 萍)