A Multi-AI-agent Framework Enabling End-to-end Finite Element Analysis for Solid Mechanics Problems

Chenxi Wang; Ling Yue; Muhammed Jawaad Zulqernine; Shaowu Pan; Shiyao Lin; Titu Ranjan Sarker

arxiv: 2606.00138 · v1 · pith:EWGS2VACnew · submitted 2026-05-28 · 💻 cs.AI

A Multi-AI-agent Framework Enabling End-to-end Finite Element Analysis for Solid Mechanics Problems

Titu Ranjan Sarker , Muhammed Jawaad Zulqernine , Ling Yue , Shaowu Pan , Chenxi Wang , Shiyao Lin This is my paper

Pith reviewed 2026-06-29 06:36 UTC · model grok-4.3

classification 💻 cs.AI

keywords finite element analysismulti-agent systemslarge language modelssolid mechanicsAbaqusautomationcomputational mechanicssimulation workflow

0 comments

The pith

A six-agent LLM framework called AbaqusAgent converts natural-language instructions into executed Abaqus finite element analyses for solid mechanics problems at 86 percent success on 50 cases.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AbaqusAgent as a multi-agent system that handles the full workflow of finite element analysis in Abaqus, from interpreting user requests to writing input files, running simulations, reviewing outputs, and visualizing results. It aims to reduce the expertise needed for solid mechanics simulations by replacing years of manual setup with coordinated large language model agents. If the approach holds, it would make computational mechanics accessible to more users and allow faster iteration on engineering problems. The validation covers a range of standard problems with an overall success rate of 86 percent. The work also positions the system as a step toward integrating AI with optimization and material characterization tasks.

Core claim

AbaqusAgent, built from six LLM agents (interpreter, architect, input writer, runner, reviewer, and visualizer), successfully generates and executes Abaqus FEA analyses for a wide variety of solid mechanics problems, achieving an overall success rate of 86 percent on 50 validated cases while encompassing all essential pre-processing and post-processing steps.

What carries the argument

AbaqusAgent, a multi-agent framework of six LLM agents that together convert natural-language instructions into complete, executed Abaqus FEA analyses and visualizations.

If this is right

FEA for solid mechanics becomes faster and requires less specialized training.
The system lowers the barrier for computational mechanics education.
Human-simulation interaction shifts toward natural language inputs.
The framework can integrate with AI-driven optimization and material characterization workflows.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same agent structure could be adapted to other commercial FEA packages beyond Abaqus.
Extending validation to dynamic or nonlinear problems would test whether success rates remain high.
The approach may enable automated sensitivity studies by repeatedly generating and running variants of a base model.

Load-bearing premise

LLM-generated Abaqus input files and post-processing steps will produce physically correct results for problems not covered in the 50-case validation set without systematic human review.

What would settle it

Expert review of Abaqus models generated by the system on a new solid mechanics problem outside the original 50 cases reveals incorrect boundary conditions, loads, or solution variables that produce non-physical results.

Figures

Figures reproduced from arXiv: 2606.00138 by Chenxi Wang, Ling Yue, Muhammed Jawaad Zulqernine, Shaowu Pan, Shiyao Lin, Titu Ranjan Sarker.

**Figure 2.** Figure 2: Category distribution of the RAG cases. The complete repository contains 104 [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

**Figure 3.** Figure 3: Demonstration of the geometry building capability of AbaqusAgent: geometries [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

**Figure 4.** Figure 4: Demonstration of AbaqusAgent’s capability to solve complex, dynamic, and [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

**Figure 5.** Figure 5: Illustration of geometry inconsistency caused by LLM hallucination. With the [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗

**Figure 6.** Figure 6: Comparison of the FEA results for an open-hole plate under tension, illustrating [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗

read the original abstract

Finite element analysis (FEA) is the most important numerical approach for solid mechanics. Challenges of FEA include a steep learning curve for entry-level users and potential false simulations due to incorrect definitions of key simulation components, such as boundary conditions, load cases, and solution variables. Years of engineering experience are usually necessary for real-world problem-solving. To address these issues, we present AbaqusAgent, a multi-agent framework grounded in large language models (LLMs) for solid mechanics analyses. AbaqusAgent is developed to facilitate analysis case generation and execution using Abaqus, one of the most widely used FEA packages, by turning users' natural-language instructions into executed FEA analyses and result visualization. AbaqusAgent is composed of six agents, including interpreter, architect, input writer, runner, reviewer, and visualizer agents, encompassing all the essential pre-processing and post-processing steps of standard FEA analyses. A wide variety of 50 solid mechanics problems have been successfully validated, achieving an overall success rate of 86%. Beyond improving the efficiency of FEA for solid mechanics problems and lowering the barrier to computational mechanics education, AbaqusAgent advances the human-simulation interaction paradigm and enables integration with AI-empowered optimization and material characterization workflows. The code is available at https://github.com/LIRAM-LIN/AbaqusAgent

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a concrete six-agent LLM setup for turning text into Abaqus runs, but the 86% success claim rests on an undefined metric and an LLM reviewer that cannot confirm physical accuracy.

read the letter

The core contribution is a six-agent system that splits the Abaqus workflow into interpreter, architect, input writer, runner, reviewer, and visualizer roles. This matches the actual steps users take when setting up solid-mechanics problems, and the authors released the code. That makes the work immediately testable rather than just a description.

The practical side is straightforward: natural-language input gets turned into a runnable model, the solver is called, and results are plotted. On 50 cases they report 86% success. For readers who already use Abaqus for routine problems, the repo could save time on repetitive setup tasks.

The main weakness is the evaluation. The abstract does not define what counts as success—solver exit code, absence of warnings, or agreement with known answers. The reviewer agent is another LLM, so it can only check syntax and obvious inconsistencies; it cannot certify that boundary conditions or material models produce correct physics. No per-problem breakdown or baseline comparison appears in the supplied description, which leaves the 86% figure hard to interpret.

The work is aimed at engineers and students who want to experiment with AI-assisted FEA rather than at researchers developing new numerical methods. The public code lowers the cost of trying it. Because the implementation is reproducible and the task is well-scoped, it is worth sending to referees who can ask for clearer success criteria and more detailed error analysis. I would not cite it yet for my own papers, but the framework itself is worth a look if the validation can be tightened.

Referee Report

2 major / 1 minor

Summary. The manuscript presents AbaqusAgent, a multi-LLM-agent framework with six specialized agents (interpreter, architect, input writer, runner, reviewer, visualizer) that converts natural-language instructions into complete, executed Abaqus finite-element analyses for solid mechanics, including pre- and post-processing steps. It reports an overall success rate of 86% on 50 validated cases and releases the code at the cited GitHub repository.

Significance. If the validation protocol can be shown to confirm physical fidelity rather than merely syntactic executability, the work would lower the barrier to computational mechanics education and enable AI-driven optimization and material-characterization pipelines. The open-source release is a clear strength supporting reproducibility.

major comments (2)

[Abstract] Abstract: the central empirical claim of an 86% success rate on 50 cases supplies no definition of success (solver exit code 0, absence of warnings, or quantitative agreement with analytical/experimental benchmarks). Because the reviewer agent is itself LLM-based, this omission prevents verification that generated models are physically faithful rather than merely runnable.
[Results / Validation] Validation protocol (implicit in the results description): no per-problem-type breakdown, no error analysis of the 14% failures, and no baseline comparison (single-LLM prompting or scripted templates) is reported. These omissions are load-bearing for the claim that the six-agent architecture enables reliable end-to-end FEA.

minor comments (1)

[Abstract] The abstract states that the framework 'advances the human-simulation interaction paradigm' without citing prior work on LLM-assisted simulation interfaces; a brief related-work paragraph would clarify novelty.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments, which help clarify the validation aspects of our work. We respond point by point below and indicate revisions where the manuscript will be updated.

read point-by-point responses

Referee: [Abstract] Abstract: the central empirical claim of an 86% success rate on 50 cases supplies no definition of success (solver exit code 0, absence of warnings, or quantitative agreement with analytical/experimental benchmarks). Because the reviewer agent is itself LLM-based, this omission prevents verification that generated models are physically faithful rather than merely runnable.

Authors: We agree that the definition of success must be stated explicitly. Success in our evaluation is defined as the Abaqus job completing with exit code 0 (no solver errors) as reported by the runner agent, together with the reviewer agent confirming that the model setup matches the natural-language problem description. We acknowledge that LLM-based review does not guarantee quantitative physical fidelity against analytical or experimental benchmarks. We will revise the abstract and methods section to include this definition and note the limitation regarding physical validation. revision: yes
Referee: [Results / Validation] Validation protocol (implicit in the results description): no per-problem-type breakdown, no error analysis of the 14% failures, and no baseline comparison (single-LLM prompting or scripted templates) is reported. These omissions are load-bearing for the claim that the six-agent architecture enables reliable end-to-end FEA.

Authors: We will add a table in the results section providing success rates broken down by problem category (e.g., linear elasticity, nonlinear material behavior, contact). We will also include a brief error analysis of the seven failure cases, classifying issues such as incorrect boundary conditions or convergence failures. Baseline comparisons against single-LLM prompting or template methods were not performed in this study; we will note this as a limitation and identify it as future work, but cannot supply such data without new experiments. revision: partial

standing simulated objections not resolved

Providing quantitative baseline comparisons (single-LLM or scripted templates), which would require new experimental runs not available in the current study.

Circularity Check

0 steps flagged

No circularity: empirical software framework with external case validation

full rationale

The paper describes a multi-LLM-agent system for generating and running Abaqus FEA analyses, reporting an 86% success rate on 50 validated cases. No mathematical derivations, equations, fitted parameters, or uniqueness theorems appear in the provided text. The central claim rests on empirical execution outcomes rather than any self-referential reduction of a prediction to its inputs or load-bearing self-citations. The validation protocol, while potentially underspecified for physical accuracy, does not create a definitional loop by construction. This is a standard non-circular empirical software paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 6 invented entities

The central claim rests on the empirical performance of LLM agents rather than new mathematical axioms or fitted constants. No free parameters are introduced. The six named agents are software components rather than new physical entities.

axioms (1)

domain assumption Large language models can reliably generate correct Abaqus input syntax and interpret simulation outputs when given appropriate role prompts.
Invoked implicitly by the construction of the interpreter, input writer, and reviewer agents.

invented entities (6)

Interpreter agent no independent evidence
purpose: Translate natural-language user instructions into FEA task specifications.
One of the six agents introduced by the framework; no independent physical evidence required.
Architect agent no independent evidence
purpose: Design the finite-element model structure.
One of the six agents introduced by the framework.
Input writer agent no independent evidence
purpose: Generate Abaqus input files.
One of the six agents introduced by the framework.
Runner agent no independent evidence
purpose: Execute the Abaqus solver.
One of the six agents introduced by the framework.
Reviewer agent no independent evidence
purpose: Check simulation results for correctness.
One of the six agents introduced by the framework.
Visualizer agent no independent evidence
purpose: Generate result visualizations.
One of the six agents introduced by the framework.

pith-pipeline@v0.9.1-grok · 5799 in / 1563 out tokens · 20696 ms · 2026-06-29T06:36:21.339506+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

58 extracted references · 6 canonical work pages

[1]

L. Yue, N. Somasekharan, T. Zhang, Y. Cao, Z. Chen, S. Di, S. Pan, Foam-agent: Towards automated intelligent cfd workflows, arXiv preprint arXiv:2505.04997 (2025)

work page arXiv 2025
[2]

Y. Chen, X. Zhu, H. Zhou, Z. Ren, Metaopenfoam: an llm-based multi- agent framework for cfd, arXiv preprint arXiv:2407.21320 (2024)

work page arXiv 2024
[3]

L. Yue, P. Xu, T. Zhang, N. Somasekharan, S. Pan, Foamgpt: Fine- tuning large language model for agentic automation of cfd simulations with openfoam, San Diego, CA (2025)

2025
[4]

Z. Dong, Z. Lu, Y. Yang, Fine-tuning a large language model for au- tomating computational fluid dynamics simulations, Theoretical and Applied Mechanics Letters 15 (3) (2025) 100594

2025
[5]

Y. Qu, K. Huang, M. Yin, K. Zhan, D. Liu, D. Yin, H. C. Cousins, W. A. Johnson, X. Wang, M. Shah, et al., Crispr-gpt for agentic automation of gene-editing experiments, Nature Biomedical Engineering 10 (2) (2026) 245–258

2026
[6]

L. Cong, D. Smerkous, X. Wang, D. Yin, Z. Zhang, R. Jin, Y. Wang, M. Gerasimiuk, R. K. Dinesh, A. Smerkous, et al., Labos: The ai-xr co-scientist that sees and works with humans, arXiv preprint arXiv:2510.14861 (2025)

work page arXiv 2025
[7]

Y. Ruan, C. Lu, N. Xu, Y. He, Y. Chen, J. Zhang, J. Xuan, J. Pan, Q. Fang, H. Gao, et al., An automatic end-to-end chemical synthe- sis development platform powered by large language models, Nature communications 15 (1) (2024) 10160

2024
[8]

Y. Zou, A. H. Cheng, A. Aldossary, J. Bai, S. X. Leong, J. A. Campos- Gonzalez-Angulo, C. Choi, C. T. Ser, G. Tom, A. Wang, et al., El agente: An autonomous agent for quantum chemistry, Matter 8 (7) (2025). 19

2025
[9]

Ghosh, A

S. Ghosh, A. Tewari, Llm-based ai agents for automated extraction of material properties and structural features, Computational Materials Science 265 (2026) 114521

2026
[10]

Jalali, Y

M. Jalali, Y. Luo, L. Caulfield, E. Sauter, A. Nefedov, C. Wöll, Large language models in electronic laboratory notebooks: Transforming ma- terials science research workflows, Materials Today Communications 40 (2024) 109801

2024
[11]

K. M. Jablonka, Q. Ai, A. Al-Feghali, S. Badhwar, J. D. Bocarsly, A. M. Bran, S. Bringuier, L. C. Brinson, K. Choudhary, D. Circi, et al., 14 examples of how llms can transform materials science and chemistry: a reflection on a large language model hackathon, Digital discovery 2 (5) (2023) 1233–1250

2023
[12]

S. Gao, A. Fang, Y. Huang, V. Giunchiglia, A. Noori, J. R. Schwarz, Y. Ektefaie, J. Kondic, M. Zitnik, Empowering biomedical discovery with ai agents, Cell 187 (22) (2024) 6125–6151

2024
[13]

C. Lu, C. Lu, R. T. Lange, Y. Yamada, S. Hu, J. Foerster, D. Ha, J. Clune, Towards end-to-end automation of ai research, Nature 651 (8107) (2026) 914–919

2026
[14]

D. A. Boiko, R. MacKnight, B. Kline, G. Gomes, Autonomous chemical research with large language models, Nature 624 (7992) (2023) 570–578

2023
[15]

J. Ji, R. Lei, X. Pan, Z. Wei, H. Sun, Y. Lin, X. Chen, Y. Yang, Y. Li, B. Ding, et al., Leveraging llm-based agents for social science research: insights from citation network simulations, Humanities and Social Sciences Communications (2026)

2026
[16]

M. J. Turner, R. W. Clough, H. C. Martin, L. Topp, Stiffness and deflection analysis of complex structures, Journal of the Aeronautical Sciences 23 (9) (1956) 805–823

1956
[17]

Plevris, G

V. Plevris, G. Markeset, Educational challenges in computer-based finite element analysis and design of structures (2018)

2018
[18]

W. R. Johnson, X.-K. Zhu, R. Sindelar, B. Wiersma, A parametric finite element study for determining burst strength of thin and thick-walled 20 pressure vessels, International Journal of Pressure Vessels and Piping 204 (2023) 104968

2023
[19]

Z. H. Zuo, Y. M. Xie, A simple and compact python code for complex 3d topology optimization, Advances in Engineering Software 85 (2015) 1–11

2015
[20]

Y. Lim, S. Ha, Rufgen: A plug-in for rough surface generation in abaqus/cae, SoftwareX 22 (2023) 101380

2023
[21]

Riaño, Y

L. Riaño, Y. Joliff, An abaqus™plug-in for the geometry generation of representative volume elements with randomly distributed fibers and interphases, Composite Structures 209 (2019) 644–651

2019
[22]

S. L. Omairey, P. D. Dunning, S. Sriramula, Development of an abaqus plugin tool for periodic rve homogenisation, Engineering with Computers 35 (2) (2019) 567–577

2019
[23]

S. Lin, A. M. Waas, Accelerating computational analyses of low velocity impact and compression after impact of laminated composite materials, Composite Structures 260 (2021) 113456

2021
[24]

S. Lin, A. M. Waas, The effect of stacking sequence on the lvi damage of laminated composites; experiments and analysis, Composites Part A: Applied Science and Manufacturing 145 (2021) 106377

2021
[25]

Lin, Himex: A hybrid implicit-explicit analysis platform for progressive damage modeling of composite materials, in: AIAA SCITECH 2025 Forum, 2025, p

S. Lin, Himex: A hybrid implicit-explicit analysis platform for progressive damage modeling of composite materials, in: AIAA SCITECH 2025 Forum, 2025, p. 1598

2025
[26]

Dhondt, Calculix crunchix user’s manual version 2.12, Munich, Ger- many, accessed Sept 21 (2017) 2017

G. Dhondt, Calculix crunchix user’s manual version 2.12, Munich, Ger- many, accessed Sept 21 (2017) 2017

2017
[27]

S. Hou, R. Johnson, R. Makhija, L. Chen, Y. Ye, Autofea: Enhancing ai copilot by integrating finite element analysis using large language models with graph neural networks, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39, 2025, pp. 24078–24085

2025
[28]

M. Yu, Y. Sheng, Z. Wan, S. Yang, M. Sun, H. Cai, An intelligent interactive system based on llm to combine cad api development with fea, in: 2024 8th International Conference on Electrical, Mechanical and Computer Engineering (ICEMCE), IEEE, 2024, pp. 1749–1755. 21

2024
[29]

Z. Li, H. Ge, G. Tang, Y. Liu, Modsolagent: Automated finite element code generation for abaqus via llm-based agent, IEEE Transactions on Industrial Informatics (2026)

2026
[30]

Zhang, Z

T. Zhang, Z. Liu, Y. Xin, Y. Jiao, Mooseagent: A llm based multi- agent framework for automating moose simulation, arXiv preprint arXiv:2504.08621 (2025)

work page arXiv 2025
[31]

C. J. Permann, D. R. Gaston, D. Andrš, R. W. Carlsen, F. Kong, A. D. Lindsay, J. M. Miller, J. W. Peterson, A. E. Slaughter, R. H. Stogner, et al., Moose: Enabling massively parallel multiphysics simulation, Soft- wareX 11 (2020) 100430

2020
[32]

Y. Qi, R. Xu, X. Chu, Feagpt: an end-to-end agentic-ai for finite element analysis, arXiv preprint arXiv:2510.21993 (2025)

work page arXiv 2025
[33]

J. Guo, C. Park, D. Qian, T. J. Hughes, W. K. Liu, Large language model-empowered next-generation computer-aided engineering, Com- puter Methods in Applied Mechanics and Engineering 450 (2026) 118591

2026
[34]

B. Ni, M. J. Buehler, Mechagents: Large language model multi-agent collaborations can solve mechanics problems, generate new data, and integrate knowledge, Extreme Mechanics Letters 67 (2024) 102131

2024
[35]

C. Tian, Y. Zhang, Collaboration dynamics and reliability challenges of multi-agent llm systems in finite element analysis, arXiv preprint arXiv:2408.13406 (2024)

work page arXiv 2024
[36]

Q. Wu, G. Bansal, J. Zhang, Y. Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, et al., Autogen: Enabling next-gen llm applications via multi-agent conversations, in: First conference on language modeling, 2024

2024
[37]

Deotale, A

R. Deotale, A. Srinivasan, Y. Tian, T. Zhang, P. Vlachos, H. Gomez, All-fem: Agentic large language models fine-tuned for finite element methods, Available at SSRN 6103826 (2026)

2026
[38]

Douze, A

M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.-E. Mazaré, M. Lomeli, L. Hosseini, H. Jégou, The faiss library, IEEE Transactions on Big Data (2025). 22

2025
[39]

Alnæs, J

M. Alnæs, J. Blechta, J. Hake, A. Johansson, B. Kehlet, A. Logg, C. Richardson, J. Ring, M. E. Rognes, G. N. Wells, The fenics project version 1.5, Archive of numerical software 3 (100) (2015)

2015
[40]

A. S. U. Manual, Abaqus 6.11, http://130.149 89 (2080) (2012) v6

2080
[41]

Iannitti, N

G. Iannitti, N. Bonora, A. Ruggiero, G. Testa, Ductile damage in taylor- anvil and rod-on-rod impact experiment, in: Journal of Physics: Confer- ence Series, Vol. 500, 2014, p. 112035. 23 Appendix A. ValidationandExtendedCapabilitiesofAbaqusAgent Appendix A.1. Validation Against Analytical Benchmark The physical correctness of AbaqusAgent-generated models...

2014
[42]

Rewrite the user's raw request into a clear, professional Abaqus-style,→ engineering prompt
[43]

Check whether these 5 required items are present: - geometry - material properties - boundary conditions - loading conditions - requested output Rules:
[44]

Preserve the user's original technical intent
[45]

Rewrite the request using proper Abaqus and finite element analysis,→ terminology
[46]

Improve clarity, grammar, and structure
[47]

rewritten_prompt

Do not invent numerical values or technical details that the user,→ did not provide. ... Return output strictly in this JSON format: { "rewritten_prompt": "string", "missing_items": [ "list of missing items from: geometry, material properties, boundary conditions, loading conditions, requested output" ] } """ 28 User Prompt — Example Raw Input Perform a s...
[48]

Read and understand the user requirement
[49]

</input_file>.,→

Read and analyze the reference input_file inside <input_file> ... </input_file>.,→
[50]

Identify from the reference file: - geometry structure - part definition - node and element organization - sets (nsets and elsets) - section definitions - material structure - assembly and instance structure - step structure - output structure
[51]

Identify from the user requirement: - intended physical behavior (bending, tension, compression, buckling, etc.),→ - correct loading type and direction - correct boundary conditions - required geometry and dimensions - correct element formulation if implied
[52]

Separate clearly: - Physics→MUST come from user requirement 31 - Syntax and structure --> may come from reference file,→
[53]

Modify ONLY what is required to satisfy the user requirement.,→
[54]

Generate a complete and runnable Abaqus input_file file.,→ ================================== PHYSICS PRIORITY RULE (CRITICAL) ================================== The user requirement ALWAYS overrides the retrieved reference case.,→ If there is any conflict between the user requirement and the reference case:,→ - Use the user requirement for: - load type -...
[55]

*PART - *NODE - *ELEMENT - *NSET / *ELSET - *SECTION
[56]

*ASSEMBLY - *INSTANCE - *END INSTANCE
[57]

*STEP - procedure (*STATIC, *DYNAMIC, etc.) - *BOUNDARY - loads (*CLOAD, *DLOAD, *DSLOAD) - *OUTPUT - *NODE OUTPUT - *ELEMENT OUTPUT
[58]

"" User Prompt — Abaqus Input File Generator user_prompt = ( f

*END STEP 34 ======================== BLOCK VALIDATION RULES ======================== - Every opened block MUST be closed: *PART→*END PART *INSTANCE→*END INSTANCE *ASSEMBLY→*END ASSEMBLY *STEP→*END STEP ================= PLACEMENT RULES ================= - Sections MUST be inside *PART - *BOUNDARY MUST be inside *STEP - All loads MUST be inside *STEP - Ou...

[1] [1]

L. Yue, N. Somasekharan, T. Zhang, Y. Cao, Z. Chen, S. Di, S. Pan, Foam-agent: Towards automated intelligent cfd workflows, arXiv preprint arXiv:2505.04997 (2025)

work page arXiv 2025

[2] [2]

Y. Chen, X. Zhu, H. Zhou, Z. Ren, Metaopenfoam: an llm-based multi- agent framework for cfd, arXiv preprint arXiv:2407.21320 (2024)

work page arXiv 2024

[3] [3]

L. Yue, P. Xu, T. Zhang, N. Somasekharan, S. Pan, Foamgpt: Fine- tuning large language model for agentic automation of cfd simulations with openfoam, San Diego, CA (2025)

2025

[4] [4]

Z. Dong, Z. Lu, Y. Yang, Fine-tuning a large language model for au- tomating computational fluid dynamics simulations, Theoretical and Applied Mechanics Letters 15 (3) (2025) 100594

2025

[5] [5]

Y. Qu, K. Huang, M. Yin, K. Zhan, D. Liu, D. Yin, H. C. Cousins, W. A. Johnson, X. Wang, M. Shah, et al., Crispr-gpt for agentic automation of gene-editing experiments, Nature Biomedical Engineering 10 (2) (2026) 245–258

2026

[6] [6]

L. Cong, D. Smerkous, X. Wang, D. Yin, Z. Zhang, R. Jin, Y. Wang, M. Gerasimiuk, R. K. Dinesh, A. Smerkous, et al., Labos: The ai-xr co-scientist that sees and works with humans, arXiv preprint arXiv:2510.14861 (2025)

work page arXiv 2025

[7] [7]

Y. Ruan, C. Lu, N. Xu, Y. He, Y. Chen, J. Zhang, J. Xuan, J. Pan, Q. Fang, H. Gao, et al., An automatic end-to-end chemical synthe- sis development platform powered by large language models, Nature communications 15 (1) (2024) 10160

2024

[8] [8]

Y. Zou, A. H. Cheng, A. Aldossary, J. Bai, S. X. Leong, J. A. Campos- Gonzalez-Angulo, C. Choi, C. T. Ser, G. Tom, A. Wang, et al., El agente: An autonomous agent for quantum chemistry, Matter 8 (7) (2025). 19

2025

[9] [9]

Ghosh, A

S. Ghosh, A. Tewari, Llm-based ai agents for automated extraction of material properties and structural features, Computational Materials Science 265 (2026) 114521

2026

[10] [10]

Jalali, Y

M. Jalali, Y. Luo, L. Caulfield, E. Sauter, A. Nefedov, C. Wöll, Large language models in electronic laboratory notebooks: Transforming ma- terials science research workflows, Materials Today Communications 40 (2024) 109801

2024

[11] [11]

K. M. Jablonka, Q. Ai, A. Al-Feghali, S. Badhwar, J. D. Bocarsly, A. M. Bran, S. Bringuier, L. C. Brinson, K. Choudhary, D. Circi, et al., 14 examples of how llms can transform materials science and chemistry: a reflection on a large language model hackathon, Digital discovery 2 (5) (2023) 1233–1250

2023

[12] [12]

S. Gao, A. Fang, Y. Huang, V. Giunchiglia, A. Noori, J. R. Schwarz, Y. Ektefaie, J. Kondic, M. Zitnik, Empowering biomedical discovery with ai agents, Cell 187 (22) (2024) 6125–6151

2024

[13] [13]

C. Lu, C. Lu, R. T. Lange, Y. Yamada, S. Hu, J. Foerster, D. Ha, J. Clune, Towards end-to-end automation of ai research, Nature 651 (8107) (2026) 914–919

2026

[14] [14]

D. A. Boiko, R. MacKnight, B. Kline, G. Gomes, Autonomous chemical research with large language models, Nature 624 (7992) (2023) 570–578

2023

[15] [15]

J. Ji, R. Lei, X. Pan, Z. Wei, H. Sun, Y. Lin, X. Chen, Y. Yang, Y. Li, B. Ding, et al., Leveraging llm-based agents for social science research: insights from citation network simulations, Humanities and Social Sciences Communications (2026)

2026

[16] [16]

M. J. Turner, R. W. Clough, H. C. Martin, L. Topp, Stiffness and deflection analysis of complex structures, Journal of the Aeronautical Sciences 23 (9) (1956) 805–823

1956

[17] [17]

Plevris, G

V. Plevris, G. Markeset, Educational challenges in computer-based finite element analysis and design of structures (2018)

2018

[18] [18]

W. R. Johnson, X.-K. Zhu, R. Sindelar, B. Wiersma, A parametric finite element study for determining burst strength of thin and thick-walled 20 pressure vessels, International Journal of Pressure Vessels and Piping 204 (2023) 104968

2023

[19] [19]

Z. H. Zuo, Y. M. Xie, A simple and compact python code for complex 3d topology optimization, Advances in Engineering Software 85 (2015) 1–11

2015

[20] [20]

Y. Lim, S. Ha, Rufgen: A plug-in for rough surface generation in abaqus/cae, SoftwareX 22 (2023) 101380

2023

[21] [21]

Riaño, Y

L. Riaño, Y. Joliff, An abaqus™plug-in for the geometry generation of representative volume elements with randomly distributed fibers and interphases, Composite Structures 209 (2019) 644–651

2019

[22] [22]

S. L. Omairey, P. D. Dunning, S. Sriramula, Development of an abaqus plugin tool for periodic rve homogenisation, Engineering with Computers 35 (2) (2019) 567–577

2019

[23] [23]

S. Lin, A. M. Waas, Accelerating computational analyses of low velocity impact and compression after impact of laminated composite materials, Composite Structures 260 (2021) 113456

2021

[24] [24]

S. Lin, A. M. Waas, The effect of stacking sequence on the lvi damage of laminated composites; experiments and analysis, Composites Part A: Applied Science and Manufacturing 145 (2021) 106377

2021

[25] [25]

Lin, Himex: A hybrid implicit-explicit analysis platform for progressive damage modeling of composite materials, in: AIAA SCITECH 2025 Forum, 2025, p

S. Lin, Himex: A hybrid implicit-explicit analysis platform for progressive damage modeling of composite materials, in: AIAA SCITECH 2025 Forum, 2025, p. 1598

2025

[26] [26]

Dhondt, Calculix crunchix user’s manual version 2.12, Munich, Ger- many, accessed Sept 21 (2017) 2017

G. Dhondt, Calculix crunchix user’s manual version 2.12, Munich, Ger- many, accessed Sept 21 (2017) 2017

2017

[27] [27]

S. Hou, R. Johnson, R. Makhija, L. Chen, Y. Ye, Autofea: Enhancing ai copilot by integrating finite element analysis using large language models with graph neural networks, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39, 2025, pp. 24078–24085

2025

[28] [28]

M. Yu, Y. Sheng, Z. Wan, S. Yang, M. Sun, H. Cai, An intelligent interactive system based on llm to combine cad api development with fea, in: 2024 8th International Conference on Electrical, Mechanical and Computer Engineering (ICEMCE), IEEE, 2024, pp. 1749–1755. 21

2024

[29] [29]

Z. Li, H. Ge, G. Tang, Y. Liu, Modsolagent: Automated finite element code generation for abaqus via llm-based agent, IEEE Transactions on Industrial Informatics (2026)

2026

[30] [30]

Zhang, Z

T. Zhang, Z. Liu, Y. Xin, Y. Jiao, Mooseagent: A llm based multi- agent framework for automating moose simulation, arXiv preprint arXiv:2504.08621 (2025)

work page arXiv 2025

[31] [31]

C. J. Permann, D. R. Gaston, D. Andrš, R. W. Carlsen, F. Kong, A. D. Lindsay, J. M. Miller, J. W. Peterson, A. E. Slaughter, R. H. Stogner, et al., Moose: Enabling massively parallel multiphysics simulation, Soft- wareX 11 (2020) 100430

2020

[32] [32]

Y. Qi, R. Xu, X. Chu, Feagpt: an end-to-end agentic-ai for finite element analysis, arXiv preprint arXiv:2510.21993 (2025)

work page arXiv 2025

[33] [33]

J. Guo, C. Park, D. Qian, T. J. Hughes, W. K. Liu, Large language model-empowered next-generation computer-aided engineering, Com- puter Methods in Applied Mechanics and Engineering 450 (2026) 118591

2026

[34] [34]

B. Ni, M. J. Buehler, Mechagents: Large language model multi-agent collaborations can solve mechanics problems, generate new data, and integrate knowledge, Extreme Mechanics Letters 67 (2024) 102131

2024

[35] [35]

C. Tian, Y. Zhang, Collaboration dynamics and reliability challenges of multi-agent llm systems in finite element analysis, arXiv preprint arXiv:2408.13406 (2024)

work page arXiv 2024

[36] [36]

Q. Wu, G. Bansal, J. Zhang, Y. Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, et al., Autogen: Enabling next-gen llm applications via multi-agent conversations, in: First conference on language modeling, 2024

2024

[37] [37]

Deotale, A

R. Deotale, A. Srinivasan, Y. Tian, T. Zhang, P. Vlachos, H. Gomez, All-fem: Agentic large language models fine-tuned for finite element methods, Available at SSRN 6103826 (2026)

2026

[38] [38]

Douze, A

M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.-E. Mazaré, M. Lomeli, L. Hosseini, H. Jégou, The faiss library, IEEE Transactions on Big Data (2025). 22

2025

[39] [39]

Alnæs, J

M. Alnæs, J. Blechta, J. Hake, A. Johansson, B. Kehlet, A. Logg, C. Richardson, J. Ring, M. E. Rognes, G. N. Wells, The fenics project version 1.5, Archive of numerical software 3 (100) (2015)

2015

[40] [40]

A. S. U. Manual, Abaqus 6.11, http://130.149 89 (2080) (2012) v6

2080

[41] [41]

Iannitti, N

G. Iannitti, N. Bonora, A. Ruggiero, G. Testa, Ductile damage in taylor- anvil and rod-on-rod impact experiment, in: Journal of Physics: Confer- ence Series, Vol. 500, 2014, p. 112035. 23 Appendix A. ValidationandExtendedCapabilitiesofAbaqusAgent Appendix A.1. Validation Against Analytical Benchmark The physical correctness of AbaqusAgent-generated models...

2014

[42] [42]

Rewrite the user's raw request into a clear, professional Abaqus-style,→ engineering prompt

[43] [43]

Check whether these 5 required items are present: - geometry - material properties - boundary conditions - loading conditions - requested output Rules:

[44] [44]

Preserve the user's original technical intent

[45] [45]

Rewrite the request using proper Abaqus and finite element analysis,→ terminology

[46] [46]

Improve clarity, grammar, and structure

[47] [47]

rewritten_prompt

Do not invent numerical values or technical details that the user,→ did not provide. ... Return output strictly in this JSON format: { "rewritten_prompt": "string", "missing_items": [ "list of missing items from: geometry, material properties, boundary conditions, loading conditions, requested output" ] } """ 28 User Prompt — Example Raw Input Perform a s...

[48] [48]

Read and understand the user requirement

[49] [49]

</input_file>.,→

Read and analyze the reference input_file inside <input_file> ... </input_file>.,→

[50] [50]

Identify from the reference file: - geometry structure - part definition - node and element organization - sets (nsets and elsets) - section definitions - material structure - assembly and instance structure - step structure - output structure

[51] [51]

Identify from the user requirement: - intended physical behavior (bending, tension, compression, buckling, etc.),→ - correct loading type and direction - correct boundary conditions - required geometry and dimensions - correct element formulation if implied

[52] [52]

Separate clearly: - Physics→MUST come from user requirement 31 - Syntax and structure --> may come from reference file,→

[53] [53]

Modify ONLY what is required to satisfy the user requirement.,→

[54] [54]

Generate a complete and runnable Abaqus input_file file.,→ ================================== PHYSICS PRIORITY RULE (CRITICAL) ================================== The user requirement ALWAYS overrides the retrieved reference case.,→ If there is any conflict between the user requirement and the reference case:,→ - Use the user requirement for: - load type -...

[55] [55]

*PART - *NODE - *ELEMENT - *NSET / *ELSET - *SECTION

[56] [56]

*ASSEMBLY - *INSTANCE - *END INSTANCE

[57] [57]

*STEP - procedure (*STATIC, *DYNAMIC, etc.) - *BOUNDARY - loads (*CLOAD, *DLOAD, *DSLOAD) - *OUTPUT - *NODE OUTPUT - *ELEMENT OUTPUT

[58] [58]

"" User Prompt — Abaqus Input File Generator user_prompt = ( f

*END STEP 34 ======================== BLOCK VALIDATION RULES ======================== - Every opened block MUST be closed: *PART→*END PART *INSTANCE→*END INSTANCE *ASSEMBLY→*END ASSEMBLY *STEP→*END STEP ================= PLACEMENT RULES ================= - Sections MUST be inside *PART - *BOUNDARY MUST be inside *STEP - All loads MUST be inside *STEP - Ou...