AI-Powered Scheduling in High-Mix, Low-Volume (HMLV) Manufacturing: An Appraisal of the Challenges, Approaches, and Future Directions

Abstract

High-Mix Low-Volume manufacturing faces persistent scheduling pressure from high product variety, frequent changeovers, alternative routings, volatile demand, and disruption events such as rush orders and equipment failures. These conditions expose limits in fixed dispatching rules, static expert policies, and large-scale mathematical programming, especially under uncertainty and tight decision windows. This review synthesizes AI-powered scheduling approaches developed for HMLV settings, with emphasis on predictive, adaptive, and decentralized control. Machine learning models support processing-time estimation, bottleneck prediction, and schedule recommendations. Reinforcement learning and deep reinforcement learning frame scheduling as sequential decision-making and learn policies that adapt under nonstationary shop-floor states, often trained and validated through simulation or digital twin environments. Metaheuristics and learning-guided hybrid methods address combinatorial scale by steering search toward feasible, high-quality schedules within practical runtimes. Multi-agent systems distribute scheduling decisions across job and resource agents through negotiation and coordination, supporting resilience and scalability in modular production lines.The review also consolidates integration requirements across IoT sensing, MES and ERP connectivity, edge deployment for low-latency execution, and digital-twin-driven verification. Comparative evaluation metrics include makespan, tardiness, utilization, and disruption responsiveness. Open challenges include sparse and inconsistent data, limited interpretability of high-capacity models, compute burden, trust and governance in human-in-the-loop operation, and sustainability and fairness constraints. Future directions prioritize explainable scheduling, continuous learning, federated training across sites, LLM-assisted scheduling knowledge, and alignment with autonomous manufacturing architectures.