A Novel LLM-Assisted Approach for Automated Mechanical Design of Regularized Geometries in CAD System

Modern product development spans a long lifecycle from planning to design, simulation, manufacturing, and customer delivery. Manufacturers place great emphasis on reducing development lead time to strengthen competitiveness. In recent manufacturing industries, most designed components feature standardized and structured geometric configurations, which can be efficiently generated through parametric design methodologies. With the rapid advancement of artificial intelligence (AI) and large language models (LLMs), research that integrates these technologies into the design and 3D simulation stages has gained increasing attention. This paper proposes a practical LLM-assisted framework for automated mechanical design of regularized geometries in a CAD system. Unlike data-driven generative CAD approaches that require large-scale CAD training datasets and may not guarantee strict dimensional correctness, the proposed method executes predefined parametric templates implemented via CAD APIs, while using an LLM to translate natural-language design intents into executable command-line instructions and to support interactive parameter refinement. A case study on ball bearing design demonstrates that the proposed framework reduces per-instance design execution time by approximately 94–95% compared to conventional manual CAD modeling, while generating CAD-native outputs that can be directly extended to downstream CAE and manufacturing workflows.