Turbulence.ai: an end-to-end AI scientist for fluid mechanics
• To our best knowledge, we developed the first AI Scientist for fluid mechanics: turbulence.ai. The AI Scientist is able to generate ideas, evaluate ideas, make simulation plans, perform simulations, analyze data and write&review manuscripts. • We also present the first pair of full-length, journal-style papers entirely drafted by an autonomous AI scientist (turbulence.ai), covering two-phase flow in porous media and flow-regime transitions in dense micro-fracture networks. • Both generated manuscripts, specially the second one, can be published on above-averaged journals. Fluid mechanics holds an almost infinite range of unsolved questions whose answers could improve energy efficiency, environmental protection and public health. Yet progress is throttled by scarce resources—human expertise, time cost and research funding. We introduce turbulence.ai , the first fully autonomous AI scientist for fluid mechanics, designed to lift those constraints. A multi-agent architecture unifies hypothesis generation, CFD execution and draft writing. From a single natural-language query the platform (i) formulates testable ideas, (ii) orchestrates a series of numerical experiments, (iii) interprets results and (iv) produces a draft (attached). This advance inaugurates a new chapter in fluid-mechanics research and could expand human’s knowledge base by orders of magnitude.
Authors
- Jingsen Feng (ORCID: https://orcid.org/0009-0007-5431-1954)
- Sandeep Pandey (ORCID: https://orcid.org/0000-0002-1858-1734)
- Xu Chu (ORCID: https://orcid.org/0000-0001-8456-1865)
- Yuanchun Qi (ORCID: https://orcid.org/0000-0003-4590-5279)
- Ran Xu
Institutions
- University of Stuttgart (DE)
- Technische Universität Ilmenau (DE)
- University of Exeter (GB)
Publication Details
- Journal
- Theoretical and Applied Mechanics Letters
- Published
- 2025-09-23
- DOI
- https://doi.org/10.1016/j.taml.2025.100620
- Primary Topic
- Computational Physics and Python Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00