[论文解读] SwarmFoam: An OpenFOAM Multi-Agent System Based on Multiple Types of Large Language Models

Chunwei Yang, Yankai Wang+7 more|arXiv (Cornell University)|Jan 12, 2026

Lattice Boltzmann Simulation Studies被引用 0

一句话总结

tldr: SwarmFoam presents a multi-agent CFD framework built on OpenFOAM that uses multiple types of large language models to enable multi-modal perception, error correction, and retrieval-augmented generation for complex simulations. It achieves notable pass rates across varied input modalities.

ABSTRACT

Numerical simulation is one of the mainstream methods in scientific research, typically performed by professional engineers. With the advancement of multi-agent technology, using collaborating agents to replicate human behavior shows immense potential for intelligent Computational Fluid Dynamics (CFD) simulations. Some muti-agent systems based on Large Language Models have been proposed. However, they exhibit significant limitations when dealing with complex geometries. This paper introduces a new multi-agent simulation framework, SwarmFoam. SwarmFoam integrates functionalities such as Multi-modal perception, Intelligent error correction, and Retrieval-Augmented Generation, aiming to achieve more complex simulations through dual parsing of images and high-level instructions. Experimental results demonstrate that SwarmFoam has good adaptability to simulation inputs from different modalities. The overall pass rate for 25 test cases was 84%, with natural language and multi-modal input cases achieving pass rates of 80% and 86.7%, respectively. The work presented by SwarmFoam will further promote the development of intelligent agent methods for CFD.

研究动机与目标

• Motivate intelligent CFD simulations by leveraging multi-agent systems and large language models.
• Address limitations of prior LLM-based CFD agents in handling complex geometries.
• Propose a framework (SwarmFoam) enabling multi-modal perception and robust guidance for CFD tasks.

提出的方法

• Integrates OpenFOAM with a multi-agent system to coordinate LLM-based agents.
• Incorporates multi-modal perception to accept images and high-level instructions.
• Employs intelligent error correction to adapt to simulation inputs.
• Utilizes Retrieval-Augmented Generation to supplement reasoning with external information.
• Supports dual parsing of images and textual commands for complex geometries.

实验结果

研究问题

• RQ1 Can SwarmFoam reliably handle CFD simulations with inputs from multiple modalities (text and images)?
• RQ2 How effective are multi-modal and natural-language inputs for achieving correct CFD simulations across varied test cases?
• RQ3 What is the overall pass rate and modality-specific performance of SwarmFoam on a set of 25 test cases?

主要发现

• Overall pass rate across 25 test cases is 84%.
• Pass rate for natural language input cases is 80%.
• Pass rate for multi-modal input cases is 86.7%.