From geometry to function: towards context-aware generative AI for engineering design

Current generative artificial intelligence for Computer-Aided Design (CAD) optimizes for geometric similarity, neglecting engineering criteria like functionality, manufacturability, and sustainability. This paper addresses this gap and proposes a conceptual framework to reorient generative CAD from replicating shapes to achieving function. We introduce two hybrid training strategies: a pre-learning approach using synthetically labeled datasets (evaluated via FEA, CAM, LCA) and a self-learning approach where GenAI uses these knowledge-based tools as a reinforcement feedback loop.