关键词: 产品-机器指派, 动态调度, 预防性维护, 马尔科夫决策过程

Abstract: A joint production dispatching and preventive maintenance problem in multi-product, heterogeneous-machine manufacturing is addressed through a Product–Machine Assignment (PMA) framework. The assignment is formulated as a mixed-integer linear program that matches product types to machines and controls per-machine variety through a penalty parameter κ, achieving a balance between throughput and machine specialization. The assignment decomposes the system into independent multi-product single-machine subproblems, each solved as a continuous-time Markov decision process to coordinate dispatching and preventive maintenance. Combining these yields a near-optimal system policy with computational complexity growing approximately linearly in the number of machines. Numerical experiments on 30 instances, compared with Cμ, First-Come-First-Served, and Round-Robin under throughput and cycle-time metrics, show that PMA maintains throughput while notably reducing cycle time. Additional tests under non-Markovian inter-event times and a real semiconductor workstation dataset exhibit consistent performance. The results indicate that PMA provides a scalable and interpretable approach for real-time joint dispatching and maintenance decisions in large heterogeneous manufacturing systems.

Key words: Product-machine assignments, Dynamic scheduling, Preventive maintenance, Markov decision process