Abstract

Implementing design for manufacturing in the early stages of product development can have significant effects on timeline and budget, but can be a difficult undertaking. Current approaches to automating rule-based feedback using large language models (LLMs) and machine learning do not adequately capture the complex manufacturing context around a product. A large part of the knowledge and experience used in giving manufacturability feedback is either tacit or lacking in structured documentation. This makes it difficult for designers to access this knowledge when seeking to consider manufacturability. This also presents a challenge for AI-in-design interventions, as there lacks an accurate way for AI models to access and reason with this tacit knowledge. Thus, we present a knowledge extraction method that enables AI to better augment early-stage concept design. Using post data from machining forums, we capture the tacit knowledge and experiences communicated by machinists and engineers, structuring it into a scalable knowledge base. The base consists of a multi-vector embedded space and a 3879-frame knowledge graph. Frame quality is validated using an automated LLM judge calibrated against a stratified 27-frame human-expert panel. This contributes a unique dataset on machining discourse and tacit knowledge, while demonstrating a method for transforming unstructured natural language data into useful design for manufacturing insights. These contributions enable AI to function as an extradisciplinary sensemaker that expands the designer’s capacity to account for external constraints.