确定截尾试验数据统计分布模型的方法

APPROACH FOR DETERMINING THE STATISTICAL DISTRIBUTION MODEL OF TEST DATA FROM TRUNCATION TAIL TESTS

研究有替换定时与定数、无替换定时与定数共4种截尾可靠性试验数据统计分布模型的确定方法。首先把试验数据区分为失效与非失效数据两部分,然后建立包含这两部分数据可能的假设分布的似然函数;根据极大似然原理与微分极值原理,发展比较似然函数值、预测与生产实际情况的一致性和尾部预测的安全性确定较好统计模型的方法。实例说明方法的可用性。

An approach is presented for determining the statistical distribution model of test data from four kinds of truncation tail tests. The four truncation tail tests is time truncation test with replacement, time truncation test without replacement, failure truncation test with replacement, and failure truncation test with replacement. The data are firstly divided into failure data and uncompleted data. The likelihood functions for seven possible statistical models are constructed with the two types of data. The seven models are exponential, normal, log-normal, 2-parameter Weibull, 3-parameter Weibull, extreme maximum value, and extreme minimum value distributions. Their statistical parameters are solved by maximum likelihood method. Appropriate model is finally determined by comparing the mounts of the their likelihood function values, the consistencies of their predicted and physical failure mechanisms, and the safety predictions in their tail regimes. Usability of the present approach is indicated by analysis to production data.