Robustness of reliability-growth analysis techniques

The author examines the robustness of techniques commonly applied to failure time data to determine if the failure rate (1/mean-time-between-failures) is changing over time. The models examined are the Duane postulate, Crow-Army material systems analysis activity, and Kalman filtering (also referred to as dynamic linear modeling). Each has as a foundation the underlying premise of changing failure rate over time. The techniques seek to confirm or reject whether failure rate is changing significantly, based on observed data. To compare the ability of each method to accomplish such a rejection or confirmation, a known failure time distribution is simulated, and then each model is applied and results are compared.<
>