pith. sign in

arxiv: 2604.06747 · v2 · submitted 2026-04-08 · 💻 cs.AI

TurboAgent: An LLM-Driven Autonomous Multi-Agent Framework for Turbomachinery Aerodynamic Design

Juan Du , Yueteng Wu , Pan Zhao , Yuze Liu , Min Zhang , Xiaobin Xu , Xinglong Zhang This is my paper

Pith reviewed 2026-05-10 18:09 UTC · model grok-4.3

classification 💻 cs.AI
keywords autonomous designmulti-agent systemslarge language modelsturbomachineryaerodynamic optimizationCFD validationcompressor designperformance prediction
0
0 comments X

The pith

An LLM can orchestrate specialized agents to autonomously generate, optimize, and validate turbomachinery designs from natural language requirements.

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

The paper introduces TurboAgent to automate the full aerodynamic design process for turbomachinery like compressors. It uses a large language model to plan tasks and coordinate specialized agents that generate geometries, predict performance quickly, optimize for better efficiency, and validate with physics simulations. This turns the traditional slow, manual trial-and-error into a collaborative, data-driven loop that starts from plain language goals. The authors test it on a transonic compressor and find the designs match target performance closely in simulations, with the optimization step boosting efficiency and pressure ratio. The whole process runs in about half an hour.

Core claim

TurboAgent uses a large language model to plan and coordinate tasks among agents responsible for generative design, rapid performance prediction, multi-objective optimization, and physics-based validation. This creates an autonomous closed-loop process from natural language requirements to final design. Validation on a transonic compressor demonstrates strong agreement with CFD simulations, with coefficients of determination exceeding 0.91 for key metrics and normalized RMSE below 8%. The optimization step improves isentropic efficiency by 1.61% and total pressure ratio by 3.02%.

What carries the argument

The multi-agent framework with an LLM core for task planning and coordination, supported by agents for geometry generation, performance prediction, optimization, and high-fidelity validation.

If this is right

  • The design process becomes fully autonomous from input requirements to validated output.
  • High-fidelity simulations are reserved for final verification rather than every iteration.
  • Performance metrics like isentropic efficiency and total pressure ratio can be improved through the optimization agent.
  • The workflow completes in approximately 30 minutes under parallel computing.

Where Pith is reading between the lines

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

  • This could shorten design cycles in industries like aviation and energy by automating what is currently a labor-intensive sequence.
  • Adding more specialized agents might allow the system to tackle multi-stage or unsteady flow problems.
  • Integration with existing simulation software could make the framework a plug-in tool for engineers.

Load-bearing premise

The language model must reliably coordinate the agents and the agents must generate valid designs and accurate predictions without major errors that would require human fixes.

What would settle it

A test where the final CFD validation shows coefficients of determination below 0.8 or where the optimized design performs worse than the initial generated design on the target metrics.

Figures

Figures reproduced from arXiv: 2604.06747 by Juan Du, Min Zhang, Pan Zhao, Xiaobin Xu, Xinglong Zhang, Yueteng Wu, Yuze Liu.

Figure 1
Figure 1. Figure 1: Multi-agent AI framework for turbomachinery design The main contributions of this study are summarized as follows: (1) An LLM-driven autonomous multi-agent framework is proposed for turbomachinery aerodynamic design. This framework enables the unified orchestration of requirement interpretation, task planning, geometry generation, performance prediction, optimization decision-making, and high-fidelity phys… view at source ↗
Figure 7
Figure 7. Figure 7: Generated blade geometries under different design objectives. [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 16
Figure 16. Figure 16: Cost analysis of TurboAgent [PITH_FULL_IMAGE:figures/full_fig_p028_16.png] view at source ↗
read the original abstract

The aerodynamic design of turbomachinery is a complex and tightly coupled multi-stage process involving geometry generation, performance prediction, optimization, and high-fidelity physical validation. Existing intelligent design approaches typically focus on individual stages or rely on loosely coupled pipelines, making fully autonomous end-to-end design challenging. To address this issue, this study proposes TurboAgent, a large language model (LLM)-driven autonomous multi-agent framework for turbomachinery aerodynamic design and optimization. The LLM serves as the core for task planning and coordination, while specialized agents handle generative design, rapid performance prediction, multi-objective optimization, and physics-based validation. The framework transforms traditional trial-and-error design into a data-driven collaborative workflow, with high-fidelity simulations retained for final verification. A transonic single-rotor compressor is used for validation. The results show strong agreement between target performance, generated designs, and CFD simulations. The coefficients of determination for mass flow rate, total pressure ratio, and isentropic efficiency all exceed 0.91, with normalized RMSE values below 8%. The optimization agent further improves isentropic efficiency by 1.61% and total pressure ratio by 3.02%. The complete workflow can be executed within approximately 30 minutes under parallel computing. These results demonstrate that TurboAgent enables an autonomous closed-loop design process from natural language requirements to final design generation, providing an efficient and scalable paradigm for turbomachinery aerodynamic design.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces TurboAgent, an LLM-driven multi-agent framework for end-to-end turbomachinery aerodynamic design. An LLM coordinates specialized agents for geometry generation, rapid performance prediction, multi-objective optimization, and high-fidelity CFD validation. On a transonic single-rotor compressor case, the framework achieves R² > 0.91 and nRMSE < 8% agreement between target performance, generated designs, and CFD results for mass flow rate, total pressure ratio, and isentropic efficiency; the optimization step yields additional gains of 1.61% in efficiency and 3.02% in pressure ratio, with the full workflow completing in approximately 30 minutes under parallel execution.

Significance. If the reported performance gains are substantiated by CFD validation of the final optimized designs (rather than surrogate predictions alone), the work would demonstrate a practical advance in autonomous, closed-loop engineering design systems. The retention of independent high-fidelity simulations for verification and the reported execution speed are positive features that could support scalable application in turbomachinery, though the single-case validation constrains broader claims of generality.

major comments (2)
  1. [Abstract] Abstract: The optimization improvements of 1.61% in isentropic efficiency and 3.02% in total pressure ratio are stated without explicit indication that these deltas were measured on CFD-validated designs rather than the surrogate model outputs. Given the reported nRMSE < 8%, these gains fall within the range that could be explained by prediction error or surrogate bias, directly affecting whether the closed-loop improvement claim is physically substantiated.
  2. [Validation and Results sections] Validation and Results sections: No details are provided on the training procedure, data splits, or baseline comparisons for the rapid performance prediction agent, nor on how the multi-objective optimization agent avoids exploiting surrogate artifacts. This information is required to evaluate whether the R² > 0.91 agreement and subsequent gains reflect genuine design improvements.
minor comments (2)
  1. [Abstract and Introduction] The abstract and introduction would benefit from a brief statement on the specific LLM model used and any safeguards against agent hallucinations or invalid geometry generation.
  2. [Figures and Tables] Figure captions and table labels should explicitly distinguish surrogate predictions from final CFD results to improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We address each major comment below and commit to revisions that will strengthen the clarity and substantiation of our claims without altering the core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The optimization improvements of 1.61% in isentropic efficiency and 3.02% in total pressure ratio are stated without explicit indication that these deltas were measured on CFD-validated designs rather than the surrogate model outputs. Given the reported nRMSE < 8%, these gains fall within the range that could be explained by prediction error or surrogate bias, directly affecting whether the closed-loop improvement claim is physically substantiated.

    Authors: We agree that the abstract requires explicit clarification on this point to avoid ambiguity. The reported gains originate from the multi-objective optimization step using the surrogate predictions, but the final optimized designs were subsequently evaluated with independent high-fidelity CFD simulations, confirming that the improvements exceed the surrogate error bounds (nRMSE < 8%) and are physically realized. In the revised manuscript, we will update the abstract to state that the deltas were measured on CFD-validated designs and add a brief note in the results section cross-referencing the CFD verification of the optimized geometry to substantiate the closed-loop claim. revision: yes

  2. Referee: [Validation and Results sections] Validation and Results sections: No details are provided on the training procedure, data splits, or baseline comparisons for the rapid performance prediction agent, nor on how the multi-objective optimization agent avoids exploiting surrogate artifacts. This information is required to evaluate whether the R² > 0.91 agreement and subsequent gains reflect genuine design improvements.

    Authors: We acknowledge that these methodological details are essential for reproducibility and to demonstrate that the results are not artifacts of the surrogate. In the revised manuscript, we will expand the relevant sections to include: the training procedure for the rapid performance prediction agent (including dataset composition, training hyperparameters, and any regularization techniques); explicit data splits (e.g., 70/15/15 train/validation/test with k-fold cross-validation); baseline comparisons against standard surrogates such as Kriging or feedforward neural networks; and the mechanisms used by the optimization agent to avoid surrogate exploitation, such as ensemble-based uncertainty quantification, enforcement of physical feasibility constraints, and mandatory final CFD validation of candidate designs rather than sole reliance on surrogate outputs. These additions will directly address whether the R² > 0.91 and gains reflect genuine improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity; validation uses independent CFD

full rationale

The paper presents a multi-agent LLM framework for turbomachinery design with specialized agents for generation, surrogate prediction, optimization, and final physics-based validation. Reported metrics (R²>0.91, nRMSE<8% agreement with CFD) and optimization deltas are outcomes of this workflow, not inputs. No equations, self-citations, or fitted parameters are shown reducing the claimed performance gains or agreements to the surrogate predictions or LLM planning by construction. High-fidelity CFD serves as external verification, keeping the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented physical entities; the contribution is the framework architecture itself rather than new physical postulates or fitted constants.

pith-pipeline@v0.9.0 · 5573 in / 1449 out tokens · 53013 ms · 2026-05-10T18:09:21.212446+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages · 1 internal anchor

  1. [1]

    Review of design optimization methods for turbomachinery aerodynamics[J]

    Li Z, Zheng X. Review of design optimization methods for turbomachinery aerodynamics[J]. Progress in Aerospace Sciences, 2017, 93: 1-23

    work page 2017

  2. [2]

    Computational fluid dynamics for turbomachinery design[J]

    Denton J D, Dawes W N. Computational fluid dynamics for turbomachinery design[J]. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 1998, 213(2): 107-124

    work page 1998

  3. [3]

    A review on aerodynamic optimization of turbomachinery using adjoint method[J]

    Lavimi R, Benchikh Le Hocine A E, Poncet S, et al. A review on aerodynamic optimization of turbomachinery using adjoint method[J]. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 2024, 238(13): 6405-6441

    work page 2024

  4. [4]

    Recent progress of efficient global aerodynamic shape optimization using surrogate-based approach[J]

    Han Z H, Xu C Z, Qiao J L, et al. Recent progress of efficient global aerodynamic shape optimization using surrogate-based approach[J]. Acta Aeronautica et Astronautica Sinica, 2020, 41(5): 623344

    work page 2020

  5. [5]

    Design optimization of axial slot casing treatment in a highly- loaded mixed-flow compressor[J]

    Ba D, Zhang Q, Du J, et al. Design optimization of axial slot casing treatment in a highly- loaded mixed-flow compressor[J]. Aerospace Science and Technology, 2020, 107: 106262

    work page 2020

  6. [6]

    Aerodynamics optimization of multi -blade centrifugal fan based on extreme learning machine surrogate model and particle swarm optimization algorithm[J]

    Meng F, Wang L, Ming W, et al. Aerodynamics optimization of multi -blade centrifugal fan based on extreme learning machine surrogate model and particle swarm optimization algorithm[J]. Metals, 2023, 13(7): 1222

    work page 2023

  7. [7]

    Application of artificial intelligence in turbomachinery aerodynamics: Progresses and challenges[J]

    Zou Z, Xu P, Chen Y , et al. Application of artificial intelligence in turbomachinery aerodynamics: Progresses and challenges[J]. Artificial Intelligence Review, 2024, 57(8): 222

    work page 2024

  8. [8]

    Generative adversarial nets[J]

    Goodfellow I J, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[J]. Advances in neural information processing systems, 2014, 27

    work page 2014

  9. [9]

    Auto-Encoding Variational Bayes

    Kingma D P, Welling M. Auto- encoding variational bayes[J]. arXiv preprint arXiv:1312.6114, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  10. [10]

    Denoising diffusion probabilistic models[J]

    Ho J, Jain A, Abbeel P. Denoising diffusion probabilistic models[J]. Advances in neural information processing systems, 2020, 33: 6840-6851

    work page 2020

  11. [11]

    Research progress and prospects of aircraft aerodynamic 31 design based on generative models[J]

    LIN J, TANG Z, QIAN W, et al. Research progress and prospects of aircraft aerodynamic 31 design based on generative models[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(10)

    work page 2025

  12. [12]

    A generative design method of airfoil based on conditional variational autoencoder[J]

    Wang X, Qian W, Zhao T, et al. A generative design method of airfoil based on conditional variational autoencoder[J]. Engineering Applications of Artificial Intelligence, 2025, 139: 109461

    work page 2025

  13. [13]

    Deep generative design: Integration of topology optimization and generative models[J]

    Oh S, Jung Y , Kim S, et al. Deep generative design: Integration of topology optimization and generative models[J]. Journal of Mechanical Design, 2019, 141(11): 111405

    work page 2019

  14. [14]

    A survey on evaluation of large language models[J]

    Chang Y , Wang X, Wang J, et al. A survey on evaluation of large language models[J]. ACM transactions on intelligent systems and technology, 2024, 15(3): 1-45

    work page 2024

  15. [15]

    Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation[J]

    Liu J, Xia C S, Wang Y , et al. Is your code generated by chatgpt really correct? rigorous evaluation of large language models for code generation[J]. Advances in neural information processing systems, 2023, 36: 21558-21572

    work page 2023

  16. [16]

    Large language models for computer-aided design fine tuned: Dataset and experiments[J]

    Sun Y , Li X, Sha Z. Large language models for computer-aided design fine tuned: Dataset and experiments[J]. Journal of Mechanical Design, 2025, 147(4): 041710

    work page 2025

  17. [17]

    Large language models for computer -aided design: A survey[J]

    Zhang L, Le B, Akhtar N, et al. Large language models for computer -aided design: A survey[J]. ACM Computing Surveys, 2026, 58(9): 1-39

    work page 2026

  18. [18]

    Metaopenfoam: an llm-based multi-agent framework for cfd

    Chen Y , Zhu X, Zhou H, et al. MetaOpenFOAM: an LLM-based multi-agent framework for CFD[J]. arXiv preprint arXiv:2407.21320, 2024

    work page arXiv 2024

  19. [19]

    ChatMOF: an artificial intelligence system for predicting and generating metal-organic frameworks using large language models[J]

    Kang Y , Kim J. ChatMOF: an artificial intelligence system for predicting and generating metal-organic frameworks using large language models[J]. Nature communications, 2024, 15(1): 4705

    work page 2024

  20. [20]

    Autonomous microscopy experiments through large language model agents

    Mandal I, Soni J, Zaki M, et al. Autonomous microscopy experiments through large language model agents[J]. arXiv preprint arXiv:2501.10385, 2024

    work page arXiv 2024

  21. [21]

    Debugbench: Evaluating debugging capability of large language models[C]//Findings of the Association for Computational Linguistics: ACL 2024

    Tian R, Ye Y , Qin Y , et al. Debugbench: Evaluating debugging capability of large language models[C]//Findings of the Association for Computational Linguistics: ACL 2024. 2024: 4173-4198

    work page 2024

  22. [22]

    Towards scientific intelligence: A survey of llm-based scientific agents

    Ren S, Xie C, Jian P, et al. Towards scientific intelligence: A survey of llm-based scientific agents[J]. arXiv preprint arXiv:2503.24047, 2025

    work page arXiv 2025

  23. [23]

    Airfoil GAN: Encoding and synthesizing airfoils for aerodynamic shape optimization[J]

    Wang Y , Shimada K, Barati Farimani A. Airfoil GAN: Encoding and synthesizing airfoils for aerodynamic shape optimization[J]. Journal of Computational Design and Engineering, 2023, 10(4): 1350-1362

    work page 2023

  24. [24]

    Development of a conditional generative adversarial network for airfoil shape optimization[C]//AIAA Scitech 2020 Forum

    Achour G, Sung W J, Pinon- Fischer O J, et al. Development of a conditional generative adversarial network for airfoil shape optimization[C]//AIAA Scitech 2020 Forum. 2020: 2261

    work page 2020

  25. [25]

    Diffairfoil: An efficient novel airfoil sampler based on latent space diffusion model for aerodynamic shape optimization[C]//AIAA A VIATION FORUM AND ASCEND 2024

    Wei Z, Dufour E R, Pelletier C, et al. Diffairfoil: An efficient novel airfoil sampler based on latent space diffusion model for aerodynamic shape optimization[C]//AIAA A VIATION FORUM AND ASCEND 2024. 2024: 3755. 32

    work page 2024

  26. [26]

    Airfoil-DDPM: A flexible airfoil generative design method using a multi -objective sampling based diffusion model[J]

    Wen Z, Lian Y , Zeng L, et al. Airfoil-DDPM: A flexible airfoil generative design method using a multi -objective sampling based diffusion model[J]. Aerospace Science and Technology, 2025: 111070

    work page 2025

  27. [27]

    Establishment of a generalizable model on a small-scale dataset to predict the surface pressure distribution of gas turbine blades[J]

    Wang Q, Yang L, Rao Y . Establishment of a generalizable model on a small-scale dataset to predict the surface pressure distribution of gas turbine blades[J]. Energy, 2021, 214: 118878

    work page 2021

  28. [28]

    Xu R, Gao S, Wang Z, et al. Reconstruction of Two- Dimensional to Three-Dimensional Flow Transition Fields Using Neural Network- Based Generative Adversarial Networks[C]//Turbo Expo: Power for Land, Sea, and Air. American Society of Mechanical Engineers, 2024, 88087: V12DT36A022

    work page 2024

  29. [29]

    Application of deep -learning method in the conjugate heat transfer optimization of full -coverage film cooling on turbine vanes[J]

    He Q, Zhao W, Chi Z, et al. Application of deep -learning method in the conjugate heat transfer optimization of full -coverage film cooling on turbine vanes[J]. International Journal of Heat and Mass Transfer, 2022, 195: 123148

    work page 2022

  30. [30]

    Reduced order modeling of the unsteady pressure on turbine rotor blades using deep learning[J]

    Joachim D, Lionel S, Jean -François T, et al. Reduced order modeling of the unsteady pressure on turbine rotor blades using deep learning[J]. arXiv preprint arXiv:2503.00013, 2025

    work page arXiv 2025

  31. [31]

    Inverse aerodynamic design of gas turbine blades using probabilistic machine learning[J]

    Ghosh S, Anantha Padmanabha G, Peng C, et al. Inverse aerodynamic design of gas turbine blades using probabilistic machine learning[J]. Journal of Mechanical Design, 2022, 144(2): 021706

    work page 2022

  32. [32]

    Inverse Design of Compressor/Fan Blade Profiles Based on Conditional Invertible Neural Networks[J]

    Jia X, Huang X, Jiang S, et al. Inverse Design of Compressor/Fan Blade Profiles Based on Conditional Invertible Neural Networks[J]. Journal of Turbomachinery, 2025, 147(10): 101014

    work page 2025

  33. [33]

    An inverse aerodynamic design framework for compressor blades based on generative model[J]

    Wu Y , Xu X, Zhang M, et al. An inverse aerodynamic design framework for compressor blades based on generative model[J]. Physics of Fluids, 2025, 37(7)

    work page 2025

  34. [34]

    Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model[J]

    Wu Y , Zhang M, Xu X, et al. Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model[J]. Journal of Turbomachinery, 2025: 1-32

    work page 2025

  35. [35]

    Knowledge transfer accelerated turbine blade optimization via an sample-weighted variational autoencoder[J]

    Guo Z, Li C, Chen Y , et al. Knowledge transfer accelerated turbine blade optimization via an sample-weighted variational autoencoder[J]. Aerospace Science and Technology, 2024, 147: 108998

    work page 2024

  36. [36]

    MechRAG: a multimodal large language model for mechanical engineering[J]

    Li S, Corney J. MechRAG: a multimodal large language model for mechanical engineering[J]. Communications Engineering, 2025, 4(1): 187

    work page 2025

  37. [37]

    A question answering system for assembly process of wind turbines based on multi-modal knowledge graph and large language model[J]

    Hu Z, Li X, Pan X, et al. A question answering system for assembly process of wind turbines based on multi-modal knowledge graph and large language model[J]. Journal of engineering design, 2025, 36(7-9): 1093-1117

    work page 2025

  38. [38]

    AutoTRIZ: Automating engineering innovation with TRIZ and large language models[J]

    Jiang S, Li W, Qian Y , et al. AutoTRIZ: Automating engineering innovation with TRIZ and large language models[J]. Advanced Engineering Informatics, 2025, 65: 103312. 33

    work page 2025

  39. [39]

    A rapidly structured aircraft concept design method based on generative artificial intelligence[J]

    Tong Y, Luo M, Ren S, et al. A rapidly structured aircraft concept design method based on generative artificial intelligence[J]. Chinese Journal of Aeronautics, 2025: 103629

    work page 2025

  40. [40]

    Prompt a robot to walk with large language models[C]//2024 IEEE 63rd conference on decision and control (CDC)

    Wang Y J, Zhang B, Chen J, et al. Prompt a robot to walk with large language models[C]//2024 IEEE 63rd conference on decision and control (CDC). IEEE, 2024: 1531-1538

    work page 2024

  41. [41]

    Materials science in the era of large language models: a perspective[J]

    Lei G, Docherty R, Cooper S J. Materials science in the era of large language models: a perspective[J]. Digital Discovery, 2024, 3(7): 1257-1272

    work page 2024

  42. [42]

    Towards optimizing with large language models.arXiv preprint arXiv:2310.05204, 2023

    Guo P F, Chen Y H, Tsai Y D, et al. Towards optimizing with large language models[J]. arXiv preprint arXiv:2310.05204, 2023

    work page arXiv 2023

  43. [43]

    Large language model agent as a mechanical designer[J]

    Jadhav Y , Barati Farimani A. Large language model agent as a mechanical designer[J]. Journal of Engineering Design, 2026: 1-37

    work page 2026

  44. [44]

    Using large language models for parametric shape optimization[J]

    Zhang X, Xu Z, Zhu G, et al. Using large language models for parametric shape optimization[J]. Physics of Fluids, 2025, 37(8)

    work page 2025

  45. [45]

    LLMs for Engineering: Teaching Models to Design High Powered Rockets[J]

    Simonds T. LLMs for Engineering: Teaching Models to Design High Powered Rockets[J]. arXiv preprint arXiv:2504.19394, 2025

    work page arXiv 2025

  46. [46]

    Massoudi S, Fuge M. Agentic large language models for conceptual systems engineering and design[C]//International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. American Society of Mechanical Engineers, 2025, 89237: V03BT03A045

    work page 2025

  47. [47]

    Kumar, G

    Kumar V , Karniadakis G E. Toward Autonomous Engineering Design: A Knowledge - Guided Multi-Agent Framework[J]. arXiv preprint arXiv:2511.03179, 2025

    work page arXiv 2025

  48. [48]

    ProtAgents: protein discovery via large language model multi-agent collaborations combining physics and machine learning[J]

    Ghafarollahi A, Buehler M J. ProtAgents: protein discovery via large language model multi-agent collaborations combining physics and machine learning[J]. Digital Discovery, 2024, 3(7): 1389-1409

    work page 2024

  49. [49]

    CFDAgent: A language -guided, zero- shot multi- agent system for complex flow simulation[J]

    Xu Z, Wang L, Wang C, et al. CFDAgent: A language -guided, zero- shot multi- agent system for complex flow simulation[J]. Physics of Fluids, 2025, 37(11)

    work page 2025

  50. [50]

    Elrefaie M, Qian J, Wu R, et al. AI agents in engineering design: a multi-agent framework for aesthetic and aerodynamic car design[C]//International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. American Society of Mechanical Engineers, 2025, 89237: V03BT03A048. 34 Appendix A: Core Prompts and Project Ma...

    work page 2025