Cost optimal sudden-death life testing

Abstract

In sudden-death life testing, a group of m specimens is put on a test machine together and run until the d-th failure. The total cost of such a test is a function of m, of the expected test length, and the number G of such groups tested. For a Weibull model, new results here optimize m and G to minimize the variance of the estimate of a specified low fractile subject to a specified total test cost. Sudden-death testing reduces test time, and the resulting data from the lower tail of the distribution yield an estimate with less model error. The results are illustrated with a client application.