Strategies for Testing and Assessing Population Reliability for Nonhomogeneous Populations of One Shot Devices

This case study discusses a situation where a subcontractor was producing munition subcomponents with a flawed subpopulation. These particular subcomponents behave in such a way that their functioning could be cycled. The failures were random and independent of the number of cycles. Two parameters were estimated: the proportion of the population that were flawed and the probability of failure of the each of the subpopulations as well as the overall probability of failure if any subcomponent was selected at random from the general population. This measure was then reported as the population reliability.It was found that traditional reliability approaches (homogenous population assumptions and time-to-failure type analysis) resulted in overly optimistic assessments of the true reliability of the population. The results were highly dependent on both the proportion of the population that was estimated to be flawed as well as the estimated probability of a flawed subcomponent actually failing. The rate of increase in overall failure probability as a function of number of found failures increases as the proportion of the flawed population increase. This resulted in a more realistic estimate of the population reliability as well as a better informed path forward for future reliability assessments.