On Problems of Estimation for Two Parameter Decreasing Failure Rate Distributions Applied to Reliability

摘要

In this paper some of the problems of parametric estimation for samples from distributions, which have decreasing failure rate, are discussed. Some of these distributions are obtained by mixing different exponential populations, others represent mechanisms analogous to 'work hardening' for each component where 'old is better than new'. Sufficient conditions are obtained that maximum likelihood estimators of appropriately chosen shape and scale parameters exist in both cases. The sample data which are available, when a decreasing failure rate distribution governs life, are usually censored with only a few failures observed; this limitation is dealt with. Actual data, obtained from the testing of integrated circuit electronic packages, are provided to illustrate these concepts and verify the applicability and usefulness of the techniques described. (Author)