| Effective and timely maintenance actions can sustain and improve both system availability and product quality in automated manufacturing systems. However, arbitrarily stopping machines for maintenance will occupy their production time and, in turn, introduce production losses into the system. Real-time maintenance decision-making in manufacturing systems is complex because it requires the integration of multiple sources of information, such as the system configuration, current machine health condition, real-time buffer level, and system throughout target.;The research presented in this thesis aims at developing tools to support real-time maintenance decision-making in complex manufacturing systems. First, the concept of passive maintenance opportunity window (PMOW) is introduced, which is defined as the machine idle-time induced by the propagation of downtime of the other machines in the system. Real-time PMOWs are predicted in manufacturing systems with serial and non-serial structures. Second, the concept of active maintenance opportunity window (AMOW) is proposed so that machines can be strategically shut down for preventive maintenance (PM) while the system throughput requirement can be still satisfied. A system decomposition method is developed to investigate the transient behavior of manufacturing systems with different configurations, based on which real-time AMOWs are estimated. Third, maintenance policies are examined by integrating the real-time buffer levels and machine degradation profiles. The system throughputs under a control-limit policy and a Markov Decision Process approach are evaluated and compared. Last, the propagation of downtime in serial-parallel manufacturing systems is studied, based on which three resilience metrics (i.e., production loss, throughput recovery time, and total underproduction time) are defined and evaluated. An optimization problem is formulated for the design of a resilient manufacturing system.;The model and methodology developed in this dissertation provide managerial insights on conducting maintenance operations in complex manufacturing systems. The effectiveness of the proposed model and algorithms is validated by case studies with simulations, and measurements in an automotive assembly plant. |
