Optimum estimators for the Weibull distribution from censored test data. Progressively-censored tests [breakdown statistics]

摘要

The paper considers the estimation of the parameters of the 2-parameter Weibull distribution from tests in which data has been progressively censored in different ways (50% censoring and two 30% censoring schemes). Four techniques are compared using Monte-Carlo simulations: maximum likelihood, least squares using the Bernard and Weibull rank estimators, and the White technique. The latter three have been specially adapted for use with censored data. It is found using several criteria that the White technique never performs badly and usually performs best. The maximum likelihood technique is reasonable under most conditions for estimating the scale but not the shape parameter. The least squares techniques generally introduce severe errors. It is recommended that the White technique is adopted widely.