Abstract Generative design can accelerate the early exploration of design solutions, yet generated designs frequently fail to meet the requirements for manufacturing. This shortcoming stems from a disconnect between the generation process of the solutions and the domain-specific design knowledge that experienced engineers apply to ensure manufacturability. To address this gap, this paper presents SheetGen, a rule-based generative design algorithm for sheet metal bending parts that systematically translates domain-specific design knowledge into algorithmic rules enforced directly during part generation rather than through post-hoc filtering. The algorithm operates on edge connections between planar surfaces and embeds manufacturing requirements at distinct stages of the generative design process, distinguishing between pre-processing, in-process, and post-processing integration points. Generated solutions are exported directly into a commercial CAD environment and validated through industrial manufacturing simulation. SheetGen is evaluated across three test cases. Compared to a baseline algorithm without rule implementation, it raises manufacturability from 40% to 64%. For an input where a generative approach from the literature produces no manufacturable parts, SheetGen achieves 86%. In a plausibility comparison with human designers who achieved 71% manufacturability, SheetGen reaches 96%. The results demonstrate that systematic knowledge integration can substantially improve the manufacturability of generatively designed sheet metal parts, and the identified strengths and limitations of the rule-based approach point toward hybrid strategies that combine rule-based reliability with AI-driven adaptability to further advance generative design for manufacturing.
Adão et al. (Mon,) studied this question.