LLM-based design process for manual assembly

In times of increasing demand for flexible manufacturing, efficient and flexible design processes are crucial for companies in high-wage countries. Therefore, this work deals with the potential of Large Language Models (LLMs) for the creation of assembly instructions based on an assembly design. To evaluate this potential, we present an evaluation method that assesses the contextual accuracy of a set of assembly instructions. The method provides a measure consisting of the three components of correct assembly sequence of components, correct assembly operations and correct positions. To validate the method, we create 50 variants of a sample assembly, simulate them in the CATIA Delmia assembly simulation environment and generated assembly instructions using a ChatGPT-based RAG model. With the proposed evaluation method, the generated assembly instructions are evaluated for contextual correctness and unambiguity and compared with each other. The results show that our evaluation method for AI-generated assembly instructions is a good first indicator to evaluate the quality in terms of contextual correctness. It also shows that LLMs have the potential with the necessary input and fine-tuning to automate the assembly instruction process.