Proportional intensity models robustness with overhaul intervals

The class of semi-parametric proportional intensity (PI) models applies to recurrent failure event modeling for a repairable system with explanatory variables (covariates). Certain repairable systems (e.g. aircraft and electrical power generating plants) experience a substantial period of downtime due to performing maintenance (i.e. major overhaul) at scheduled intervals or following a major failure. Other systems (e.g. emergency power units) experience extended periods of non-operating dormancy. These discontinuities in observation time have potential effects oil the accuracy of estimation for covariate effects, particularly where calendar time is the life metric. This paper examines the robustness of two PI methods (Prentice-Williams-Peterson gap time (PWP-GT) and Andersen-Gill (AG)) as a function of the overhaul or dormancy duration. The PWP-GT model proves to perform well for sample size of 60 (30 per level of a class covariate), constant or moderately decreasing/increasing rate of occurrence of failures, and relative overhaul (dormancy) durations less than half of the immediately preceding interval between failures. The AG model performs consistently well for a small sample size of 20 (10 per level of a class covariate) for homogeneous Poisson processes, regardless of the relative overhaul (dormancy) duration. Copyright (C) 2005 John Wiley & Sons, Ltd.