Language models to assess printability in additive manufacturing: a future perspective

Abstract

Large language models and vision-language models have the potential to change how printability is assessed in additive manufacturing. By learning from the rapidly expanding body of literature, process logs, simulation outputs, and image‑based defect data, these models can move printability evaluation from trial‑and‑error experimentation to scalable, knowledge‑rich reasoning across materials and processes. They can classify and explain defects, link them to process–parameter choices, and act as intelligent supervisors that suggest targeted process-variable adjustments while integrating physics‑based simulations and specialized machine‑learning models. At the same time, realizing this potential requires new multimodal printability datasets, physics-preserving algorithms, rigorous uncertainty quantification, and benchmarking frameworks, as well as exploration of advanced computational infrastructures, including quantum‑inspired approaches for high‑dimensional optimization. This perspective outlines how language models can capture human expertise, unify heterogeneous data, and enable more reliable, data‑driven printability decisions for next‑generation metal additive manufacturing.