Download file (3.4 MB)
Download as zip (2.75 MB)

Collection

Citation

Description:

This study presents CrystalPlasticitySim, a multi-agent system that leverages large language models (LLMs) to automate complex workflows in crystal plasticity simulations. Traditional simulation processes require extensive expertise in materials science, software operation, and computational techniques, making them time-consuming and inaccessible to non-specialists. To address these challenges, our system integrates three collaborating AI agents—a Supervisor Agent, Simulation Agent, and Computational Assistant Agent—that autonomously handle task decomposition, input file generation, simulation execution, result extraction, and parameter optimization.

Using a case study on the anisotropic deformation behavior of Ni₃Al single crystals during cold rolling, we demonstrate that CrystalPlasticitySim can significantly reduce manual effort and improve efficiency. Tasks that previously required months of human work can be completed within hours through autonomous execution. The system also incorporates self-correction mechanisms, enabling it to detect and resolve common runtime errors without human intervention.

Furthermore, we propose a four-level taxonomy of AI agents in materials simulation and position our system at the transition between task-solving and problem-solving agents. The results highlight the potential of AI-driven multi-agent systems to enhance reproducibility, accessibility, and efficiency in materials research workflows.

Rights:

• Creative Commons BY Attribution 4.0 International
  

Keyword: Multi-agent LLM, Crystal plasticity simulation, Automated simulation

Date published: 2026-12-31

Publisher: Informa UK Limited

Journal:

• Science and Technology of Advanced Materials: Methods (ISSN: 27660400) vol. 6 issue. 1 2630445

Funding:

• Ministry of Education, Culture, Sports, Science and Technology (MEXT) Program: Data Creation and Utilization-Type Material Research and Development Project JPMXP1122684766

Manuscript type: Publisher's version (Version of record)

MDR DOI:

First published URL: https://doi.org/10.1080/27660400.2026.2630445

Related item:

Other identifier(s):

Contact agent:

Updated at: 2026-07-03 10:01:23 +0900

Published on MDR: 2026-07-03 12:30:35 +0900