Dynamic Scheduling Techniques in Cloud Manufacturing – An Exploration of Deep Reinforcement Learning as a Critical Opportunity for Future Research

Abstract: Objective: Determine the state-of-the-art in dynamic scheduling techniques for cloud manufacturing.Significance: This paper firmly establishes the underexplored technique of Deep Reinforcement Learning as the state-of-the-art for dynamic scheduling in cloud manufacturing, exposes a significant gap in the literature, and sets out critical future research objectives.Abstract:For many years, metaheuristic algorithms have represented the state of the art in manufacturing scheduling techniques, proving to be exceptionally reliable for optimising schedules. However, metaheuristics suffer from inherent weaknesses that inhibit their ability to be applied to dynamic cloud manufacturing (CMfg) scheduling problems in practice. Thanks to the very recent and rapidly accelerating development in deep reinforcement learning (DRL), a small sample of studies have described how those approaches have thoroughly outperformed metaheuristic algorithms in dynamic manufacturing scheduling problems, establishing a new state of the art. However, a significant lag in maturity exists between the algorithms used in CMfg and state-of-the-art DRL. This paper systematically reviews the CMfg scheduling literature published between 2010 and 2020, summarises the development of deep reinforcement learning in this context and offers valuable directions for future research.

Keywords: deep reinforcement learning, dynamic scheduling, cloud manufacturing, real-time, resource scheduling, optimization

DOI: 10.54941/ahfe1001625

Cite this paper:

​