ChatCFD: An End-to-End CFD Agent with Domain-Specific Structured Thinking

Abstract

ChatCFD is an automated agent system that simplifies complex Computational Fluid Dynamics (CFD) simulations for OpenFOAM. It tackles the field's high expertise requirements by using large language models (LLMs) and a multi-agent architecture to process multimodal inputs like research papers and meshes. The system uses a four-stage pipeline—Knowledge Base Construction, User Input Processing, Case File Generation, and Execution/Error Reflection—to automate setups. In validation tests, ChatCFD achieved an 82.1% success rate on 205 benchmark cases, significantly outperforming competitors like MetaOpenFOAM (6.2%) and Foam-Agent (42.3%). It also showed 60-80% success on complex literature-derived cases. Despite its success, the system shows limitations, including LLM bias toward simpler setups and lower success rates for rare turbulence models. By automating hypothesis testing, ChatCFD accelerates scientific discovery in fluid mechanics and engineering, demonstrating strong potential for future AI-driven CFD innovation.

The manuscript of this research is available at https://arxiv.org/abs/2506.02019
 

Speaker

Dr Fan is a Research Associate in the High-speed Thermo-fluid and MAV/UAV Laboratory at The Hong Kong Polytechnic University. His research focuses on the intersection of artificial intelligence and fluid dynamics, specifically LLM-driven CFD Agents, CFD modeling, detonation, and shock-flame interaction.

He earned his PhD in Mechanical Engineering from The Hong Kong Polytechnic University in 2021. Before his current role, Dr Fan served as Technical Director at Shenzhen SimArk Co. Ltd. (2022-2023) and was a Postdoctoral Fellow at the Southern University of Science and Technology (2023-2025).