System Design with Physical Law Self-verification Type Language Models

Language models can sometimes produce statements that do not correspond to reality. This phenomenon,known as hallucination, poses a major challenge when applying language models to real-world product design. Inthis work, we propose a system-design framework based on a physical law self-verification language model, in whicha physics-based simulator and a large language model are integrated in a closed loop so that the final product is freefrom the inconsistency. The simulator-in-the-loop concept is validated on a high-complexity aerospace engineeringapplication. Specifically, using the proposed method, we autonomously and self-directedly carry out system design toenhance the thrust of a solid-state electroaerodynamic propulsion system, an application in the aerospace engineeringdomain, and examine both the reasoning process and the resulting final design.