|
Abstract:
|
Failure events associated with systems reliability admit the possibility of repair and continued availability. For analyzing systems reliability data, the principal approach uses mean cumulative functions (MCFs), the repairable-system analog to hazard functions. MCFs implicitly focus on the time-ordered slices at which failure events occur. Using models conditioned on these time slices (these risk sets), MCF-based methods can adapt the Mantel-Haenszel and Cox proportional hazard approaches for quantifying the relative-risk coefficients associated with particular factors. Further, when the number of candidate risk factors is large, a modification to classification trees ("CART") is suggested for model reduction. This modification is conditional on the risk sets, and exploits random permutations within the risk sets to calculate significance. We illustrate with a series of examples.
|
