Proposing a model based on deep reinforcement learning for real-time scheduling of collaborative customization remanufacturing

Abstract

The mass production of products in recent decades has led to the excessive exploitation of global resources and environmental degradation. Researchers tackle this challenge by proposing methods for reusing end-of-life products, including remanufacturing strategies. On the other hand, today's consumers seek products that completely fulfill their needs. For this reason, leading manufacturers prioritize customization to improve consumer satisfaction. In contrast to previous studies, this research investigates the real-time scheduling problem of intelligent systems in remanufacturing collaboratively customized products. To address this problem, the multi-agent deep Q-network method is proposed and designed. The elements of this method are defined for each remanufacturing department, including disassembly, cleaning-repair, and assembly stations. The experimental data is simulated to evaluate the proposed method based on a realistic smartphone assembly environment that can produce 46,656 unique products. Despite the disruption caused by the arrival of new jobs, the proposed method's results outperform those of the combined genetic algorithm. They can reduce factory costs by >6 %.