Proceedings of the 2019 Winter Simulation Conference N. Mustafee, K.-H.G. Bae, S. Lazarova-Molnar, M. Rabe, C. Szabo, P. Haas, and Y.-J. Son, eds. A NUMERICAL STUDY ON THE STRUCTURE OF OPTIMAL PREVENTIVE MAINTENANCE POLICIES IN PROTOTYPE TANDEM QUEUES Taehyung Kim James R. Morrison Department of Industrial and System Engineering Korea Advanced Institute of Science and Technology 291 Daehak-ro, Yuseong-gu Daejeon, 34141, SOUTH KOREA ABSTRACT While high levels of automation in modern manufacturing systems increase the reliability of production, tool failure and preventive maintenance (PM) events remain a significant source of production variability. It is well known for production systems, such as the M/G/1 queue, that optimal PM policies possess a threshold structure. Much less is known for networks of queues. Here we consider the prototypical tandem queue consisting of two exponential servers in series subject to health deterioration leading to failure and repair. We model the PM decision problem as a Markov decision process (MDP) with a discounted infinite- horizon cost. We conduct numerical studies to assess the structure of optimal policies. Simulation is used to assess the value of the optimal PM policy relative to the use of a PM policy derived by considering each queue in isolation. Our simulation studies demonstrate that the mean cycle time and discounted operating costs are 10% superior. 1 INTRODUCTION 1.1 Overview While high levels of automation in modern manufacturing systems increase the reliability of production, tool failure and preventive maintenance (PM) events remain a significant source of production variability. It is well known for production systems, such as the M/G/1 queue, under certain assumptions and subject to tool failures, repairs, and PMs, that optimal PM policies possess a threshold structure. Much less is known for queues in tandem, or more generally, networks of queues. Here we consider the prototypical tandem queue consisting of two exponential servers in series subject to deterioration in health, failure, and repair and numerically explore the structure of optimal policies in this system. 1.2 Literature Review Much of the research on PM policies has concentrated on non-production systems or has taken an isolated view of the production equipment. When production systems are considered, there are two general categories of work: production systems that build into an inventory and queueing systems. We will briefly discuss equipment, inventory, and queueing systems next, with an eye toward motivating our work. PMs are essential for efficient machine operation and much work has been devoted to the study of PM policies, c.f. (Sim and Endrenyi 1988; Nicolai and Dekker 2008; Moghaddan and Usher 2011). Applications include manufacturing equipment (cf. Laggoune et al. (2009)), windmills (cf. Krishna (2012)), and trucks (cf. Barde et al. (2016)). Often, a PM policy to maximize the mean availability is sought. Alternately, the cost of maintaining the machine may be pursued by considering failure costs, repair costs, and maintenance costs. Recent examples in this vein of work include Barde et al. (2016) and Barde et al. (2019) which focus 2281 978-1-7281-3283-9/19/$31.00 ©2019 IEEE