Deep reinforcement learning-based dynamic scheduling for resilient and sustainable manufacturing: A systematic review

Abstract

Dynamic scheduling plays a pivotal role in smart manufacturing by enabling real-time adjustments to production schedules, thereby enhancing system resilience and promoting sustainability. By efficiently responding to disruptions, dynamic scheduling maintains productivity and stability, while also reducing resource consumption and environmental impact through optimized operations and the potential integration of renewable energy. Deep Reinforcement Learning (DRL), a cutting-edge artificial intelligence technique, shows promise in tackling the complexities of production scheduling, particularly in solving NP-hard combinatorial optimization problems. Despite its potential, a comprehensive study of DRL's impact on dynamic scheduling, especially regarding system resilience and sustainability, has been lacking. This paper addresses this gap by presenting a systematic review of DRL-based dynamic scheduling focusing on resilience and sustainability. Through an analysis of two decades of literature, key application scenarios of DRL in dynamic scheduling are examined, and specific indicators are defined to assess the resilience and sustainability of these systems. The findings demonstrate DRL's effectiveness across various production domains, surpassing traditional rule-based and metaheuristic algorithms, particularly in enhancing resilience. However, a significant gap remains in addressing sustainability aspects such as energy flexibility, resource utilization, and human-centric social impacts. This paper also explores current technical challenges, including multi-objective and multi-agent optimization, and proposes future research directions to better integrate resilience and sustainability in DRL-based dynamic scheduling, with an emphasis on real-world application.