Automatic Creation of Assembly Instructions by Using Retrieval Augmented Generation

The application of Large Language Models (LLMs) for the automated generation of assembly instructions shows significant potential for improving work preparation in production processes. However, challenges remain regarding the overall information quality and precision of the generated instructions. In light of these challenges, this study explores how the information quality of automatically generated assembly instructions can be enhanced through the targeted provision of structured input data, such as Assembly and Quantity BOMs (Bills of Materials), as well as the use of optimized prompt chaining techniques. The methodology employs ChatGPT-4o in combination with Retrieval Augmented Generation (RAG) within the Microsoft Azure environment. The results demonstrate that structured data inputs, particularly the use of Assembly BOMs with defined Tool-to-Component relations, significantly improve the precision and relevance of the generated instructions. Despite these advancements, achieving consistent information quality remains a barrier to broader practical implementation. Therefore, feedback loops should be integrated into the assembly instruction generation process to ensure continuous refinement and reliability. Future research should investigate the use of RAG or similar frameworks, focusing on optimizing data structures and implementing feedback mechanisms to enhance the automated generation of assembly instructions.