Integrated Scheduling of Distributed Hybrid Flowshop Production, Spare Parts Inventory, and Equipment O&M Activities Using a Learning-Assisted Co-Evolutionary Algorithm
Main article
Abstract
The interdependence between production scheduling, spare parts inventory management, and equipment operation and maintenance (O&M) represents a critical yet underexplored opportunity for cross-domain resource optimization in manufacturing systems. Conventional scheduling approaches treat these three domains independently, resulting in suboptimal resource utilization, excessive preventive maintenance downtime, and inventory holding costs that could be substantially reduced through coordinated planning. This paper formulates the Integrated Production-Inventory-O&M scheduling problem (IPIOM) for distributed hybrid flowshop environments, where manufacturers simultaneously optimize job sequencing across geographically distributed production facilities, on-hand spare parts inventory allocation to support maintenance activities, and preventive maintenance (PM) schedules coordinated with spare parts delivery timelines. The IPIOM model employs an optimal speed adjustment strategy that modulates processing speeds to synchronize production completion with spare parts availability, and defines PM trigger conditions based on equipment degradation state and predicted maintenance windows. The bi-objective formulation minimizes total manufacturer costs (production, inventory holding, maintenance, and energy) and total customer capacity loss from delayed deliveries and equipment downtime. To efficiently solve the NP-hard IPIOM, we develop a Learning-Assisted Co-Evolutionary Algorithm (LACA) that integrates a Proximal Policy Optimization (PPO) reinforcement learning mechanism for adaptive operator selection with problem-specific global search operators and element-specific local search. Computational experiments on benchmark instances with 10-100 machines and 20-200 jobs demonstrate that LACA achieves 15-25% cost reduction compared to independent scheduling approaches and outperforms NSGA-III, MOEA/D, MOPSO, and classic GA baselines on standard multi-objective quality metrics (IGD, HV, SP). Sensitivity analysis confirms LACA maintains superior performance under demand uncertainty, processing time variability, and maintenance duration perturbations.
