No abstract

GAVER AND JACOBSWe would first like to thank Gaver and Jacobs for their thoughtful discussion. With respect to the issues raised, we regard the potential masking problem to be of greatest concern. Viewing a discrete-use mission "trial" as an ordered sequence of stages as they do clearly points to the problem that the occurrence of a failure mode at stage i can preclude a potential failure mode associated with a later stage from occurring on the trial. Such preclusion effects violate the model's failure mode occurrence independence assumptions. However, if the graphical and statistical goodness-of-fit procedures do not provide evidence that the observed failure mode first occurrence trial pattern is inconsistent with the model's expected number of failure modes metric, we would deem such preclusion effects to be insignificant for the observed test data. On the contrary, if there were statistical evidence against the model, alternate projection assessment techniques will have to be employed. Such techniques that provide additional (and sometimes needed) model fidelity not only increase model complexity, but much more importantly, typically increase the number of model parameters that must be assessed. This in turn can potentially degrade the accuracy of the reliability projection. Thus, we would recommend applying a projection model that is as parsimonious as possible with respect to model parameters, subject to the condition that the observed data is consistent with the model. The discussants write: "the first model" where "no fixes occur until T tests/trials elapse, during which no repair/redesign is made. (The second model does permit sequential fixes.)" Our article addresses what the discussants refer to as the "second model." This model accommodates a corrective action strategy that incorporates fixes to failure modes at any time after the failure mode has been observed in test. Under such a scenario, the initial probability of occurrence of an observed failure mode may not be constant over the T trials. Thus it may be difficult to obtain an accurate statistical estimate of the initial mode probability of occurrence from the test data. Such an estimate is desired if one wishes to utilize individual mode FEFs to reduce the initial probabilities of occurrence of observed failure modes to which corrective actions have been applied during the test period. This consideration motivated using the average of the assessed FEFs for the observed failure modes.