Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards an AI Fluid Scientist: LLM-Powered Scientific Discovery in Experimental Fluid Mechanics

Haodong Feng, Lugang Ye, Dixia Fan

TL;DR

The paper introduces an AI Fluid Scientist that autonomously conducts the full experimental research cycle in experimental fluid mechanics, validated on vortex- and wake-induced vibrations in tandem cylinders. It combines a computer-controlled circulating water tunnel with a multi-agent LLM framework (including human-in-the-loop options) to generate hypotheses, design and execute experiments, analyze data, and craft manuscripts. Key findings include faithful reproduction of classical VIV/WIV benchmarks, discovery of novel WIV phenomena and velocity-dependent regimes, and a shift from physics-based to neural-network-based empirical modeling that markedly improves fit quality. The work demonstrates practical feasibility of end-to-end autonomous scientific discovery in a high-cost, safety-constrained experimental domain, while acknowledging limitations of current LLM evaluation and the need for robust discriminators.

Abstract

The integration of artificial intelligence into experimental fluid mechanics promises to accelerate discovery, yet most AI applications remain narrowly focused on numerical studies. This work proposes an AI Fluid Scientist framework that autonomously executes the complete experimental workflow: hypothesis generation, experimental design, robotic execution, data analysis, and manuscript preparation. We validate this through investigation of vortex-induced vibration (VIV) and wake-induced vibration (WIV) in tandem cylinders. Our work has four key contributions: (1) A computer-controlled circulating water tunnel (CWT) with programmatic control of flow velocity, cylinder position, and forcing parameters (vibration frequency and amplitude) with data acquisition (displacement, force, and torque). (2) Automated experiments reproduce literature benchmarks (Khalak and Williamson [1999] and Assi et al. [2013, 2010]) with frequency lock-in within 4% and matching critical spacing trends. (3) The framework with Human-in-the-Loop (HIL) discovers more WIV amplitude response phenomena, and uses a neural network to fit physical laws from data, which is 31% higher than that of polynomial fitting. (4) The framework with multi-agent with virtual-real interaction system executes hundreds of experiments end-to-end, which automatically completes the entire process of scientific research from hypothesis generation, experimental design, experimental execution, data analysis, and manuscript preparation. It greatly liberates human researchers and improves study efficiency, providing new paradigm for the development and research of experimental fluid mechanics.

Towards an AI Fluid Scientist: LLM-Powered Scientific Discovery in Experimental Fluid Mechanics

TL;DR

The paper introduces an AI Fluid Scientist that autonomously conducts the full experimental research cycle in experimental fluid mechanics, validated on vortex- and wake-induced vibrations in tandem cylinders. It combines a computer-controlled circulating water tunnel with a multi-agent LLM framework (including human-in-the-loop options) to generate hypotheses, design and execute experiments, analyze data, and craft manuscripts. Key findings include faithful reproduction of classical VIV/WIV benchmarks, discovery of novel WIV phenomena and velocity-dependent regimes, and a shift from physics-based to neural-network-based empirical modeling that markedly improves fit quality. The work demonstrates practical feasibility of end-to-end autonomous scientific discovery in a high-cost, safety-constrained experimental domain, while acknowledging limitations of current LLM evaluation and the need for robust discriminators.

Abstract

The integration of artificial intelligence into experimental fluid mechanics promises to accelerate discovery, yet most AI applications remain narrowly focused on numerical studies. This work proposes an AI Fluid Scientist framework that autonomously executes the complete experimental workflow: hypothesis generation, experimental design, robotic execution, data analysis, and manuscript preparation. We validate this through investigation of vortex-induced vibration (VIV) and wake-induced vibration (WIV) in tandem cylinders. Our work has four key contributions: (1) A computer-controlled circulating water tunnel (CWT) with programmatic control of flow velocity, cylinder position, and forcing parameters (vibration frequency and amplitude) with data acquisition (displacement, force, and torque). (2) Automated experiments reproduce literature benchmarks (Khalak and Williamson [1999] and Assi et al. [2013, 2010]) with frequency lock-in within 4% and matching critical spacing trends. (3) The framework with Human-in-the-Loop (HIL) discovers more WIV amplitude response phenomena, and uses a neural network to fit physical laws from data, which is 31% higher than that of polynomial fitting. (4) The framework with multi-agent with virtual-real interaction system executes hundreds of experiments end-to-end, which automatically completes the entire process of scientific research from hypothesis generation, experimental design, experimental execution, data analysis, and manuscript preparation. It greatly liberates human researchers and improves study efficiency, providing new paradigm for the development and research of experimental fluid mechanics.

Paper Structure

This paper contains 16 sections, 2 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Automated Experimental Apparatus
  4. AI Scientist Framework
  5. Results
  6. Reliability Validation Against Classical Literatures
  7. VIV Validation
  8. WIV Validation and High-$U_r$ Discovery
  9. Standard WIV Experiments
  10. High-$U_r$ Discovery
  11. Feasibility of Scientific Discovery
  12. Physical Phenomena Discovery
  13. Physical Formula Discovery
  14. Usability: End-to-End Autonomous Research System
  15. Conclusion
  16. ...and 1 more sections

Figures (3)

  • Figure 1: The pipline of AI fluid scientist. The green block represents the LLM agent, the blue block represents the automatic hardware agent, and the specific composition is shown in subfigure (c). The orange block represents the generated content, including hypotheses, experimental parameters, manuscript, etc.
  • Figure 2: Results and errors of neural network formula.
  • Figure 3: Contour plot of displacement amplitude in the $f_f$--$L/D$ plane. A pronounced peak at $f \approx 1.2$ Hz and $L/D \approx 4.5$ indicates a critical interaction zone where energy transfer is maximized.