What happened
Researchers developed a framework that makes it safer and easier for AI language models to automatically build digital simulations of manufacturing systems by separating the structural design (what components exist and how they connect) from the performance tuning (adjusting numbers based on real sensor data). In practice, this means factories can now get working simulation models faster without sacrificing human ability to understand and verify what the AI actually built.
Why it matters
Digital twins — computer models that mirror physical systems — are expensive and slow to build by hand, but letting AI do it alone creates black boxes that engineers can't trust. This work shows how to keep the speed advantage while making the AI's decisions visible and verifiable to human experts, which matters because manufacturing decisions based on opaque models can lead to real production failures.