Current issues concerning reliability estimation in operational test and evaluation

Abstract

There are several perennial issues concerning rehability estimation in Operational Testing and Evaluation (OT&E). Two of these include, 'how should one model the underlying failure distribution of a continuous-time system,' and, 'how can a testing agency use information from DT in order to reduce OT resource requirements.' In the former issue, some OT&E analysts have questioned whether or not the exponential failure distribution should be used in all cases for continuous-fim systems, and have suggested the Weibull distribution as an alternative in some instances. In the latter, the notion of combining DT and OT data has been an anathema to those involved in OT&E, however, with ever-tightening military budgets, it may be time to revisit the issue. First, this thesis compares the exponential and Weibull failure distributions in terms of the amount of test time needed to demonstrate, to a given level of confidence, that the true MTTF of a system is at least as large as the minimum acceptable value, and also in terms of the actual confidence level associated with the lower confidence level procedure when the system has an increasing (or decreasing) failure rate function. Second, the thesis examines the behavior of an estimator for the relationship between DT and OT failure data using a Monte Carlo simulation. Finally, the thesis introduces a hierarchical Bayes approach for the estimation of the relationship between DT and OT failure data when a gamma prior distribution is assumed.