Deep Reinforcement Learning for Integrated Production-Maintenance Scheduling in Smart Manufacturing Systems
Main article
Abstract
The escalating complexity of modern smart manufacturing environments demands integrated optimization strategies that simultaneously address production scheduling, equipment maintenance planning, and quality control. Traditional approaches treat these domains in isolation, resulting in suboptimal performance, elevated operational costs, and increased equipment downtime. This paper presents a novel Deep Reinforcement Learning (DRL) framework that achieves holistic integration of production task scheduling, predictive maintenance (PdM), and quality control within Industry 4.0 and Industry 5.0 manufacturing ecosystems. Drawing upon a systematic literature review following a PRISMA-like methodology across six major electronic databases (n = 2,847 initial records; n = 143 final included studies), we identify the state-of-the-art AI algorithms and integration mechanisms employed in this domain. The proposed framework introduces a hierarchical DRL architecture that leverages real-time IoT sensor data, digital twin representations, and multi-objective reward functions to dynamically optimize scheduling decisions under uncertainty. Experimental validation on a simulated flexible manufacturing cell demonstrates that the DRL-based approach achieves a 45.7% reduction in maintenance costs, a 31.2% improvement in on-time delivery performance, and an 18.4% increase in overall equipment effectiveness (OEE) compared to genetic algorithm baselines. Furthermore, the integration of industrial information systems and interoperability protocols enables seamless data exchange across heterogeneous manufacturing modules, aligning with the human-centric, resilient principles of Industry 5.0. The findings establish DRL as a compelling paradigm for next-generation intelligent manufacturing optimization.
