Date of Award

Spring 2014

Degree Type


Degree Name

Master of Science (MS)


Aeronautics and Astronautics

First Advisor

Karen Marais

Committee Member 1

William A. Crossley

Committee Member 2

John P. Sullivan


This thesis presents an explicit consideration of the impacts of modeling decisions on the resulting maintenance planning. Incomplete data is common in maintenance planning, but is rarely considered explicitly. Robust optimization aims to minimize the impact of uncertainty--here, in contrast, I show how its impact can be explicitly quantified. Doing so allows decision makers to determine whether it is worthwhile to invest in reducing uncertainty about the system or the effect of maintenance. The thesis consists of two parts. Part I uses a case study to show how incomplete data arises and how the data can be used to derive models of a system. A case study based on the US Navy's DDG-51 class of ships illustrates the approach. Analysis of maintenance effort and cost against time suggests that significant effort is expended on numerous small unscheduled maintenance tasks. Some of these corrective tasks are likely the result of deferring maintenance, and, ultimately decreasing the ship reliability. I use a series of graphical tests to identify the underlying failure characteristics of the ship class. The tests suggest that the class follows a renewal process, and can be modeled as a single unit, at least in terms of predicting system lifetime.

Part II considers the impact of uncertainty and modeling decisions on preventive maintenance planning. I review the literature on multi-unit maintenance and provide a conceptual discussion of the impact of deferred maintenance on single and multi-unit systems. The single-unit assumption can be used without significant loss of accuracy when modeling preventive maintenance decisions, but leads to underestimating reliability and hence ultimately performance impacts in multi-unit systems. Next, I consider the two main approaches to modeling maintenance impact, Type I and Type II Kijima models and investigate the impact of maintenance level, maintenance interval, and system quality on system lifetime. I quantify the net present value obtained of the system under different maintenance strategies and show how modeling decisions and uncertainty affect how closely the actual system and maintenance policy approach the maximum net present value. Incorrect assumptions about the impact of maintenance on system aging have the most cost, while assumptions about design quality and maintenance level have significant but smaller impact. In these cases, it is generally better to underestimate quality, and to overestimate maintenance level.