Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenFOAMGPT 2.0: end-to-end, trustworthy automation for computational fluid dynamics

Jingsen Feng, Ran Xu, Xu Chu

TL;DR

The paper presents OpenFOAMGPT 2.0, a four-agent, end-to-end framework that converts natural language CFD queries into executable simulations with automated pre-processing, prompt generation, simulation, and post-processing. By tightly coupling specialized agents with deterministic OpenFOAM configurations and a closed-loop error refinement, the approach achieves 100% reproducibility across diverse cases, including single- and multi-phase Poiseuille flow, porous-media transport, and motorbike aerodynamics. The work demonstrates robust, conversation-driven workflows that maintain numerical rigor while reducing manual expertise requirements, validated across more than 450 simulations. This framework has significant practical impact by democratizing access to complex CFD analyses and establishing a foundation for trustworthy, scalable, and extensible AI-assisted scientific computing.

Abstract

We propose the first multi agent framework for computational fluid dynamics that enables fully automated, end to end simulations directly from natural language queries. The approach integrates four specialized agents Pre processing, Prompt Generation, OpenFOAMGPT (simulator), and Post processing decomposing complex computational fluid dynamics workflows into collaborative components powered by large language models. Extensive validation through diverse case studies, including Poiseuille flows, single and multi phase porous media flows, and aerodynamic analyses, demonstrates 100% success and reproducibility rates across over 450 simulations. Rigorous trustworthiness verification confirms that properly designed multi agent systems can achieve the reliability standards necessary for zero tolerance scientific computing applications while significantly lowering entry barriers. The framework establishes a foundation for conversation-driven simulation workflows in computational science, potentially accelerating discovery and innovation through more accessible tools for complex numerical simulations. Results reveal that multi-agent architectures, when properly specialized and orchestrated, can effectively handle the stringent requirements of computational physics while maintaining the intuitive interface of natural language interaction.

OpenFOAMGPT 2.0: end-to-end, trustworthy automation for computational fluid dynamics

TL;DR

The paper presents OpenFOAMGPT 2.0, a four-agent, end-to-end framework that converts natural language CFD queries into executable simulations with automated pre-processing, prompt generation, simulation, and post-processing. By tightly coupling specialized agents with deterministic OpenFOAM configurations and a closed-loop error refinement, the approach achieves 100% reproducibility across diverse cases, including single- and multi-phase Poiseuille flow, porous-media transport, and motorbike aerodynamics. The work demonstrates robust, conversation-driven workflows that maintain numerical rigor while reducing manual expertise requirements, validated across more than 450 simulations. This framework has significant practical impact by democratizing access to complex CFD analyses and establishing a foundation for trustworthy, scalable, and extensible AI-assisted scientific computing.

Abstract

We propose the first multi agent framework for computational fluid dynamics that enables fully automated, end to end simulations directly from natural language queries. The approach integrates four specialized agents Pre processing, Prompt Generation, OpenFOAMGPT (simulator), and Post processing decomposing complex computational fluid dynamics workflows into collaborative components powered by large language models. Extensive validation through diverse case studies, including Poiseuille flows, single and multi phase porous media flows, and aerodynamic analyses, demonstrates 100% success and reproducibility rates across over 450 simulations. Rigorous trustworthiness verification confirms that properly designed multi agent systems can achieve the reliability standards necessary for zero tolerance scientific computing applications while significantly lowering entry barriers. The framework establishes a foundation for conversation-driven simulation workflows in computational science, potentially accelerating discovery and innovation through more accessible tools for complex numerical simulations. Results reveal that multi-agent architectures, when properly specialized and orchestrated, can effectively handle the stringent requirements of computational physics while maintaining the intuitive interface of natural language interaction.
Paper Structure (20 sections, 6 figures, 1 table)

This paper contains 20 sections, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Multi-Agent LLM framework for CFD
  3. Pre-processing Agent:
  4. Prompt Generation Agent:
  5. OpenFOAMGPT:
  6. Post-processing Agent:
  7. Experiments
  8. Single-phase Poiseuille flow
  9. Multi-phase Poiseuille flow
  10. 1. Grid Independence Study:
  11. 2. Saturation Variation Study:
  12. 3. Viscosity Ratio Study:
  13. Single-phase flow in porous media
  14. Multi-phase flow in porous media
  15. 1. Contact Angle Study:
  16. ...and 5 more sections

Figures (6)

  • Figure 1: The design of multi-Agent framework for CFD. Red components represent user interfaces for input queries and output results. Green components indicate LLM-based intelligent agents. Blue components show built-in code modules working alongside agents to implement specific functionalities. Yellow components denote information and states provided during workflow execution. Orange components represent text-based output logs.
  • Figure 2: Validation and parametric studies of Poiseuille flow simulations. (a) Comparison between numerical results and analytical solution for single-phase Poiseuille flow; (b) Grid independence study for multi-phase Poiseuille flow; (c) Effect of viscosity ratio on velocity distribution in multi-phase Poiseuille flow; (d) Influence of wetting phase (water) saturation on velocity profiles in multi-phase Poiseuille flow. All subfigures are generated by the OpenFOAMGPT 2.0
  • Figure 3: Results of single-phase flow in porous media simulation. (a) Grid independence study of permeability; (b) Determination of permeability REV size; (c) Relationship between flux and pressure difference. All three upper figures are generated by OpenFOAMGPT 2.0
  • Figure 4: Displacement efficiency in drainage process of multi-phase flow through porous media. (a) Effect of contact angle variation; (b) Influence of inlet velocity; (c) Impact of viscosity ratio between displaced and invading fluids. All three upper figures are generated by OpenFOAMGPT 2.0
  • Figure 5: Results of aerodynamics of a motorbike simulation. (a) Drag coefficient ($C_d$) in different velocity; (b) The streamline flow characteristics surrounding a motorbike operating at $100\text{m/s}$. The left figure is generated by OpenFOAMGPT 2.0
  • ...and 1 more figures