AI CFD Scientist: Toward Open-Ended Computational Fluid Dynamics Discovery with Physics-Aware AI Agents

arXiv 2026 80.8 method

TLDR

AI CFD Scientist is an open-source AI agent that autonomously conducts computational fluid dynamics discovery with physics-aware verification, achieving a 7.89% improvement in wall Cf RMSE.

Reasoning

Strengths: novel integration of vision-based physics verification and full workflow automation for CFD; demonstrated improvement on a benchmark. Weaknesses: limited to a single backbone (GPT-5.5) and only five tasks; generalizability to other CFD problems not shown.

Read-first score

Read-first score 80.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 90.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=artifact,code,github

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=ablation,baseline,result

Topical relevance 42%

Uses existing LLM keyword relevance scores normalized to 0-100. AI scientist,automated scientific discovery,autonomous research agent,automated research,literature review agent,survey generation,automated experimentation,experiment design agent,AI for scientific research,paper writing agent,research automation,scientific discovery agent

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: stable; rank range: 1.

Keyword Scores

AI scientist

automated scientific discovery

autonomous research agent

AI for scientific research

scientific discovery agent

automated research

automated experimentation

research automation

experiment design agent

paper writing agent

literature review agent

survey generation

Deep Analysis

Innovations

First AI scientist for CFD that integrates literature-grounded ideation, validated execution, vision-based physics verification, source-code modification, and figure-grounded writing in a single inspectable workflow
Vision-language physics-verification gate that inspects rendered flow fields to detect silent failures missed by solver-level checks
Three coupled pathways for parameter sweeps, case-local C++ library compilation for new physical models, and open-ended hypothesis search against a reference comparator

Methodology

The AI CFD Scientist uses a GPT-5.5 backbone and operates on OpenFOAM via Foam-Agent, with three coupled pathways: parameter sweeps, compilation of new physical models as C++ libraries, and open-ended hypothesis search. A central vision-language gate inspects rendered flow fields to verify physical validity before accepting, rerunning, or writing results.

Key Results

The system autonomously discovered a Spalart-Allmaras runtime correction reducing lower-wall Cf RMSE against DNS by 7.89% on a periodic hill at Reh=5600; general AI-scientist baselines (ARIS, DeepScientist) could not produce defensible claims due to missing validity gates; and the vision-language gate detected 14 of 16 silent failures in a planted-failure ablation.

Limitations

The vision-language verification gate missed 2 out of 16 planted silent failures in the ablation study.

Resources

Paper Code