Finite-horizon maintenance under geometric-process models: Quantifying model and decision uncertainty

Abstract

Geometric process (GP) models are widely used for repairable systems whose inter-failure times evolve over successive failure-repair cycles. Maintenance decisions under GP models are often made with limited failure data and are subject to uncertainty in baseline selection and the resulting maintenance decision. Existing approaches typically rely on model ranking and plug-in estimates, which can understate uncertainty surrounding next-cycle replacement decisions. This paper develops a unified generalized-inference framework for finite-horizon, event-driven maintenance decision making under GP models. The proposed framework quantifies uncertainty in GP model parameters, cycle-specific reliability, hazard rate, one-cycle cost rate, and the associated optimal replacement time. For exponential and gamma baselines, the framework yields generalized confidence intervals through generalized pivotal quantities. For the Weibull baseline, it is implemented through a generalized fiducial approach. It also extends to a finite-sample discrimination procedure for quantifying baseline-selection uncertainty. Theoretical results establish the asymptotic validity of the proposed procedures and their interval estimates. Simulation study shows reliable finite-sample coverage for model parameters and nonlinear maintenance-relevant quantities, and satisfactory discrimination performance. A case study on aircraft air-conditioning failures shows that, in early cycles, replacement schedules with similar expected cost can imply materially different risk profiles, so a single plug-in recommendation may be misleading.