AutoFOAM's AI agent automates OpenFOAM CFD simulations
New AI agent runs, refines and evolves CFD simulations from natural language...
Researchers from [institution redacted] have developed AutoFOAM, a self-refining autonomous agent that automates Computational Fluid Dynamics (CFD) workflows using OpenFOAM. The system leverages the Qwen-coder 2.5-14B model, fine-tuned on 252 text prompts spanning 7 OpenFOAM solvers and 13 parametrized mesh templates, enabling true natural-language-driven simulation setup.
The agent's core innovation is a 7-stage evolution loop that iteratively creates, evaluates, and refines simulations while preventing model collapse through three safeguards: RAG-augmented retry context, surgical dictionary-level patching, and prompt-diversity paraphrasing. This approach accelerates rapid prototyping while democratizing access to advanced CFD workflows that traditionally require specialized expertise.
- AutoFOAM is built on Qwen-coder 2.5-14B and fine-tuned on 252 text prompts for 7 OpenFOAM solvers
- The agent uses a 7-stage evolution loop with RAG-augmented retry and anti-collapse safeguards
- Enables natural-language-driven CFD simulation setup and autonomous refinement
Why It Matters
Automates complex CFD workflows, reducing expertise barriers and accelerating engineering simulations.