Human-in-the-loop and large language models in smart manufacturing: Current applications, challenges, and perspectives

Abstract

The advent of Large Language Models (LLMs) has led to a transformative shift in modern industrial production, enabling enhanced automation, decision-making, and adaptability. LLMs have demonstrated their potential in diverse manufacturing applications, including design for manufacturability, process planning, and anomaly detection. This paper presents a systematic scoping review of large language model applications in smart manufacturing with a particular emphasis on human-in-the-loop concepts. By analyzing recent industrial and academic studies, the review examines how LLMs are integrated across manufacturing functions, how human roles are defined and operationalized, and what limitations remain in terms of trust, validation, and human centricity. Based on the findings of the review, an exemplar conceptual structure is synthesized to organize existing approaches and highlight research gaps from an Industry 5.0 perspective. This four-modular structure integrates human-in-the-loop (HITL), cyber-physical systems (CPS), LLM, and verification, validation, and uncertainty management (VV&UM). The HITL module addresses real-time human oversight, refining AI-generated insights and reducing decision-making errors; the CPS module aims to bridge the gap between digital twins, real-world sensor data, and AI-driven predictions, enabling real-time validation of AI recommendations; the LLM module is responsible to enhances manufacturability awareness; and the VV&UM module establishes structured verification approach, uncertainty quantification, risk assessment mechanisms, and authentication and authorization mechanisms to ensure the AI-generated outputs are reliable and compliant with industry standards. By integrating these four modules, the LLM with human-in-the-loop-based smart manufacturing (LLM-HSM) exemplar conceptual structure creates a hybrid intelligence model where AI enhances automation, while human expertise ensures contextual accuracy, performance assurance, and quality control. This paper explores the potentials, challenges, and future perspectives of LLMs and HITL in smart manufacturing, outlining a forward-looking exemplar conceptual structure for their responsible and practical implementation in next-generation industrial systems.