Multiobjective production and opportunistic maintenance dispatching rule optimization: A graph reinforcement learning approach

For manufacturing systems that execute multiple production mission phases with variable requirements throughout their lifecycle, a multiobjective operation and maintenance (O&M) optimization method is necessary for interphase adaptation. Additionally, an integrated method that simultaneously optimizes two in-process O&M policies, namely the Production and Opportunistic Maintenance (P&OM) schedules, is necessary, given that these two are highly interdependent. Correspondingly, this study introduces a joint multiobjective dispatching rule optimization method for P&OM problems. To handle the increasing system complexity and requirement variability, Graph Reinforcement Learning (GRL) is applied to implement optimization. The main contributions of this paper include: (1) formulating the P&OM optimization framework that in-volves time, quality, and cost as objectives, (2) proposing the universal dispatching rule graph for comprehensive representation of P&OM instances, (3) decomposing the dispatching rule optimization problem into a sequential process, and (4) developing a weighed-adjacency graph multihead attention agent to implement GRL and optimize the dispatching process. The applicability and effectiveness of the proposed method have been verified via a real industrial case study of flexible nuclear fuel rod shielding component production line.