A Study on Shape Design Agent for Fluid Machinery Using Large Language Model and Retrieval-Augmented Generation of Design Knowledge

In the design of fluid machinery, shape is a crucial factor that influences performance. Shape design is a highly subjective task that requires going through a trial-and-error process of iterative shape modification and evaluation, as well as an empirical process based on past design knowledge. To address this challenge, we have developed a system that combines a Large Language Model (LLM) capable of generating and interpreting various types of knowledge with a method called Reasoning and Acting (ReAct). This approach incorporates trial-and-error behavior, rapid performance evaluation using surrogate models, and design knowledge extraction through Retrieval-Augmented Generation (RAG), allowing for interactive shape generation based on user instructions.