Interval estimation from a life-test through a sequential Kolmogorov-Smirnov type procedure

A possibly-censored sequential procedure, based on the 1-sample 2-sided Kolmogorov-Smirnov statistic, for estimating-with a prespecified probability of error-the unknown (vector) parameter of any continuous distribution, is presented. One of the main features of this procedure is the capability of determining, during a life-test, the time when the observed data are sufficient to stop the experiment without losing the accuracy of a complete life-test. An example is given which shows that, by using this procedure, in a test conducted on 100 items having exponentially distributed lifetimes with mean 1000 hours, over 5000 test hours can be saved. In another example, to estimate the 2-parameters of a Weibull life distribution, this procedure immediately applies to the case of 2- (or more) parameter distributions