In this study, we developed an automatic program generation system for Design Insight CAD, which checks a variety of design rules on 3D-CAD models. The system applies a Retrieval-Augmented Generation (RAG) approach: given natural language design rules, it extracts relevant feature shapes and retrieves the associated shape recognition functions from a vector database. Information of these retrieved functions are then combined with prompt to generate Design Insght CAD program and it generates the program automatically. We validated the system using a design rule about the relationship between distance between a hole and a plate edge, and thickness of the plate. The generated program for the rule correctly identified all locations in a complex assembly model where the rule was violated. This result demonstrates that the proposed framework can produce valid and reliable rule-checking program based on design rule entered in natual language. It is expected that this system will enable CAD to flexibly respond to various design rules at the design site, and will contribute to the utilization of CAD in the manufacturing industry in the future.
YOSHIDA et al. (Wed,) studied this question.