Control-limit policies for coordinated dispatching and preventive maintenance in multi-product manufacturing systems

This research proposes a real-time dispatching and preventive maintenance (RDPM) framework for multi-product manufacturing systems. The objective is to minimize the weighted long-run average holding cost. Under random deterioration of machine health, a Markov decision process (MDP) model explicitly considers the following common manufacturing system characteristics that are rarely combined in the literature: 1) Processing times that decrease monotonically with machine health deterioration but unevenly between different products; 2) Machine deterioration that depends on dispatching decisions since producing different products may stress the machine unevenly; and 3) Preventive maintenance (PM) times that depend on machine health. Since dispatching decisions are intertwined with the health of the machine, the coordination between dispatching and PM decisions is crucial. Under these settings, our theoretical findings suggest the existence of an optimal control-limit policy for PM decisions. This control-limit policy dictates whether the machine should prioritize PM or the processing of a particular product under each state. Compared to dispatching methods from the literature, our numerical analysis shows that the proposed RDPM framework reduces the weighted average holding costs by more than 25% on average and outperforms condition-based maintenance strategies on average throughput. Furthermore, additional experiments show that these advantages remain valid under Markovian settings with imperfect health information and non-Markovian settings, i.e., the robustness of the RDPM framework in general production environments. The RDPM framework also shows superiority compared with other commonly used policies on a publicly available dataset from a semiconductor manufacturing test platform.