MooseAgent: A LLM Based Multi-agent Framework for Automating Moose Simulation (2504.08621v1)

Abstract: The Finite Element Method (FEM) is widely used in engineering and scientific computing, but its pre-processing, solver configuration, and post-processing stages are often time-consuming and require specialized knowledge. This paper proposes an automated solution framework, MooseAgent, for the multi-physics simulation framework MOOSE, which combines large-scale pre-trained LLMs with a multi-agent system. The framework uses LLMs to understand user-described simulation requirements in natural language and employs task decomposition and multi-round iterative verification strategies to automatically generate MOOSE input files. To improve accuracy and reduce model hallucinations, the system builds and utilizes a vector database containing annotated MOOSE input cards and function documentation. We conducted experimental evaluations on several typical cases, including heat transfer, mechanics, phase field, and multi-physics coupling. The results show that MooseAgent can automate the MOOSE simulation process to a certain extent, especially demonstrating a high success rate when dealing with relatively simple single-physics problems. The main contribution of this research is the proposal of a multi-agent automated framework for MOOSE, which validates its potential in simplifying finite element simulation processes and lowering the user barrier, providing new ideas for the development of intelligent finite element simulation software. The code for the MooseAgent framework proposed in this paper has been open-sourced and is available at https://github.com/taozhan18/MooseAgent

Analysis of "MooseAgent: A LLM Based Multi-agent Framework for Automating Moose Simulation"

The paper presents "MooseAgent," an innovative framework that combines LLMs with multi-agent systems to automate simulations within the Multiphysics Object Oriented Simulation Environment (MOOSE) framework. MOOSE, widely used in applications like nuclear energy and biomechanics, provides a flexible yet complex platform for multi-physics simulations. However, its inherent complexity, particularly in preprocessing, solver configuration, and post-processing stages, necessitates significant expertise and manual effort from users. This paper addresses these challenges by leveraging advancements in NLP to automate these tasks, thereby reducing the required expertise for users, specifically in finite element methods (FEM).

MooseAgent exploits LLMs' ability to understand user-described simulation requirements in natural language. By employing multi-round iterative verification strategies, MooseAgent automatically generates the necessary MOOSE input files. A key innovation in MooseAgent is the use of a vector database that stores annotated MOOSE input cards and function documentation. This database is instrumental in minimizing model hallucinations—erroneous outputs generated by LLMs—thereby enhancing accuracy.

Key Contributions and Methodology

The paper details several notable contributions:

• Multi-Agent Automation: The development of a multi-agent system that automates MOOSE via LLMs, streamlining preprocessing, solver configuration, and post-processing.
• Task Decomposition and Verification: Incorporates task decomposition with iterative verification, enabling adaptation to diverse simulation tasks and automatic corrections of sub-tasks for improved stability.
• Evaluative Experimentation: Systematically tests the framework on various simulation scenarios, including heat transfer, elasticity, and multi-physics coupling, illustrating the framework's utility in enhancing executability and accuracy.

The experiments conducted encompass steady-state and transient phenomena, elasticity and plasticity, porous media flow, phase change, and coupled thermal-mechanical simulations. The results suggest that MooseAgent shows a high success rate, especially in single-physics problems such as steady-state heat conduction. These findings underscore MooseAgent's potential to simplify complex simulation tasks and lower the barrier to entry for users lacking deep-domain expertise.

Experimental Observations

Success rates varied across different problem domains, with cases like steady-state heat conduction achieving 100% success, while more intricate thermal-mechanical coupling cases demonstrated lower success rates. This discrepancy highlights the ongoing challenges in fully automating complex, multi-physics simulations with current AI technologies. Token usage patterns also revealed insights into the efficiency of model operations. For instance, simpler cases consumed fewer tokens, reflecting efficient computational processes.

Implications and Future Work

The automation offered by MooseAgent implies significant practical and theoretical implications:

• Practical Implications: By lowering the expertise threshold, MooseAgent makes MOOSE more accessible to interdisciplinary researchers and users, enhancing productivity in simulation-driven research fields.
• Theoretical Implications: The integration of LLMs with simulation frameworks may inspire future research on intelligent automation in scientific computing, merging AI capabilities with traditional engineering frameworks.

Future developments proposed by the authors include optimizing knowledge retrieval and iteration strategies, potentially incorporating human feedback to improve handling of complex problems involving strong physics coupling.

In summary, "MooseAgent" navigates the intricate landscape of automating complex simulations with a novel amalgamation of LLMs and multi-agent frameworks, ushering in new potentials for FEM applications. The evaluative data, while promising, also delineates areas requiring further advancement, inviting continued exploration and refinement in the integration of AI with computational simulation environments.