Towards Large Language Model-Driven Personalization of Assembly Work Instructions

Industry 5.0 (I5.0) introduces human-centric automation. An important role within this context and the assembly domain is the creation of assembly work instructions, which often remain generic and inflexible. The manual creation of individual, user-centric instructions is especially time-consuming and requires expert input. Current systems lack dynamic adaptations to user profiles and learning needs. Large Language Models (LLMs) offer new potential for generating personalized work instructions. Therefore, this paper explored challenges, requirements, and evaluation metrics for integrating LLMs in creating adaptable assembly instructions. With special focus on cognitive design, human factors, semantic validation, and performance metrics, the research field of I5.0 is addressed. A literature-based analysis on these topics results in a proposal of a system concept combining user profiles, LLM processing, semantic checks, and feedback loops. The findings show that LLMs can generate high-quality text and adapt to user-specific needs. However, current limitations include a lack of domain understanding and reliability in sequence generation. Semantic validation and adaptive learning loops are essential for robust deployment. The contribution of this paper is a conceptual framework for LLM-based, human-centered assembly instruction systems, which provides the foundation for future research and practical implementation in manufacturing environments.