The proposal introduces a modular, closed-loop approach designed to make imperfect AI models highly functional within advanced manufacturing settings. In workflows like Fused Deposition Modeling (FDM) 3D printing, generic AI or default configurations often fail to reliably meet highly specific objectives. To overcome this, the framework treats an LLM not as an all-knowing oracle, but as a source of expert tuning guidance embedded directly inside a Bayesian optimization loop. By systematically evaluating outcomes and executing reliability-aware workflows, this method effectively handles uncertainty, reduces trial-and-error waste, and establishes an automated pipeline for continuous process improvement.
Application/demonstration
The planned demonstration will showcase manufacturing robots utilizing an LLM-guided Bayesian optimization loop to iteratively tune and automatically optimize FDM print configuration parameters for high-quality additive manufacturing.
