Joint Optimization of Spare Part Manufacturing and Maintenance Workforce Scheduling Under Heterogeneous In-Warranty and Out-of-Warranty Demands

The efficient operation of the maintenance service system is key to achieving sustainable operations, with its core lying in the coordinated scheduling of spare parts production and maintenance personnel, as well as the holistic management of in-warranty and out-of-warranty demands. This approach optimizes resource allocation and enhances long-term service value. This paper investigates the integrated scheduling of distributed spare parts production and maintenance personnel with differentiated in-warranty and out-of-warranty demands (ISSPD). To solve the ISSPD, an improved non-dominated sorting genetic algorithm-II that uses Q-learning to adaptively select local search strategies (QLNSGA) is proposed, which incorporates a decoding strategy for differentiated order types, eight knowledge-driven local search strategies, and a Q-learning mechanism for the adaptive selection of key local search operators. Compared to random local search operators, the Q-learning mechanism achieves a 55% decrease in IGD metric and a 65% increase in HV metric. Through comparative experiments with four mainstream algorithms, QLNSGA outperforms RIPG by 58% in terms of the IGD index, and its HV index is generally superior to that of comparative algorithms such as MOEA/D. This indicates that QLNSGA exhibits superior performance in both computational efficiency and solution quality, effectively enhancing service levels and significantly reducing operational costs, thereby providing scientific decision support for service-oriented manufacturing enterprises.