pith. machine review for the scientific record. sign in

arxiv: 2604.16687 · v1 · submitted 2026-04-17 · 💻 cs.AI · cs.LG

Recognition: unknown

Agentic Risk-Aware Set-Based Engineering Design

Varun Kumar , George Em Karniadakis
Authors on Pith no claims yet

Pith reviewed 2026-05-10 08:10 UTC · model grok-4.3

classification 💻 cs.AI cs.LG
keywords multi-agent systemslarge language modelsset-based designrisk managementconditional value-at-riskairfoil designhuman-in-the-loopengineering automation
0
0 comments X

The pith

LLM agents apply CVaR risk thresholds to prune large sets of airfoil designs down to a small validated collection.

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

The paper sets out a human-supervised workflow in which four specialized LLM agents first build analysis tools, then run global sensitivity studies to create design rules, generate candidate airfoils, and apply a quantitative risk filter before handing a short list to the engineer. The filter removes candidates whose lift performance is likely to fall short of the required target when operating conditions vary. A sympathetic reader would care because early-stage design normally requires examining thousands of possibilities under uncertainty, and the method claims to automate the bulk of that work while keeping the final decision with the human expert.

Core claim

The authors describe a set-based process in which the Analyst Agent computes sensitivity to identify influential parameters, the Design Agent produces a broad initial collection, and a risk step discards members whose conditional expected shortfall in lift coefficient exceeds an acceptable level, leaving only a reduced set that is then checked with high-fidelity flow simulations.

What carries the argument

Conditional Value-at-Risk (CVaR) applied to the distribution of lift coefficients across uncertain operating conditions, used as a threshold to eliminate designs with high tail risk of missing the performance target.

If this is right

  • The final candidate set carries an explicit upper bound on the expected performance shortfall.
  • Computational effort for detailed simulations is concentrated on only the low-risk survivors.
  • The human manager receives both the short list and quantitative risk scores for each remaining design.
  • Heuristics derived from sensitivity analysis become reusable rules for subsequent design rounds.

Where Pith is reading between the lines

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

  • The same agent structure and CVaR filter could be tested on other performance metrics such as drag or structural margin.
  • If the agents prove reliable, the human oversight step could be reduced to final approval rather than continuous guidance.
  • The workflow offers a concrete test bed for measuring how often LLM-generated heuristics remain valid when the design space changes.

Load-bearing premise

The LLM agents will generate accurate sensitivity results, valid design heuristics, and correct risk rankings without introducing systematic engineering mistakes or inconsistent outputs.

What would settle it

Execute the full workflow on a family of airfoils whose lift performance under the same uncertain conditions has already been measured by independent high-fidelity runs; the pruned final set should contain only designs whose observed failure probability lies below the CVaR threshold.

Figures

Figures reproduced from arXiv: 2604.16687 by George Em Karniadakis, Varun Kumar.

Figure 1
Figure 1. Figure 1: Schematic for LLM-based set-based design workflow with risk filtering strategy. The LLM workflow consists of four agents: Design Engineer, Systems Engineer, Coding Assistant, and Analyst, each assigned a designated role and task. The workflow utilizes different tools that are developed by the Coding Assistant following human manager’s instructions. The workflow starts with the human manager providing a des… view at source ↗
Figure 2
Figure 2. Figure 2: Sample instruction provided to the Coding Agent by the human user to create a parameter sampling tool based on specific process requirements. 4.3. Agent: Systems Engineer The Systems Engineer agent functions as the primary design evaluator and strategic guide within the multi-agent framework, responsible for providing assessment of airfoil designs generated by the Design Engineer. This agent uses Gemini-2.… view at source ↗
Figure 3
Figure 3. Figure 3: Plot showing the starting design set and the selected designs (in red) after the utility score-based filtering process. X and Y axes show the Principal Components of the CST parameters that define airfoil geometry. Principal Component values are used here for visualization purpose. actionable guidelines that can be used by the Design Agent to strategically modify airfoil geometries in subsequent stages. To… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison between cumulative distribution of CD and CL before and after design modification by the Design agent. The cumulative distribution shows the improvement in overall values of CD and CL after parameter modification based on the Sensitivity analysis. The generation of refined designs is followed by a subsequent filtering step to ensure that the selected candidates are not only high-performing but a… view at source ↗
Figure 5
Figure 5. Figure 5: Plot showing CL distributions of two design candidates: ID–98 (retained) and ID–878 (removed) used for α-risk filtering strategy. The orange line marks the CL value at 1 − α percentile of the distribution. in the previous stage, these modifications are more focused, with the primary objective of further enhancing the lift coefficient (CL) while maintaining the favorable drag and moment characteristics of t… view at source ↗
Figure 6
Figure 6. Figure 6: Coefficient of Pressure plots for two airfoil samples generated after analysis using the DeepONet surrogate. These pressure plots allows the Systems Engineer agent to perform a comprehensive review of the airfoil design, to include choice of design parameters, integral coefficients such as CD, CL, and CM, and the pressure distribution curves generated here. by a specialized ‘Systems Engineer’ agent, whose … view at source ↗
Figure 7
Figure 7. Figure 7: Instruction set provided to the Systems Engineer during the automated design evaluation and filtering phase. In practice, the Systems Engineer agent systematically processes each of the 50 final design candidates, applying the defined rubric to generate the utility scores and ordinal CP ratings. This automated procedure ensures a consistent and objective application of the evaluation criteria across the en… view at source ↗
Figure 8
Figure 8. Figure 8: Example of feedback generated by the Systems Engineer during automated design evaluation stage for an invalid design. Design evaluation for Design ID-470 (Valid) “The pressure distribution for airfoil ID-470 exhibits a smooth and well-behaved curve on both upper and lower surfaces, with no significant adverse pressure gradients or sudden changes, indicating good flow characteristics. The suction peak on th… view at source ↗
Figure 9
Figure 9. Figure 9: Example of feedback generated by the Systems Engineer during automated design evaluation stage for a valid design. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Feedback for design ID-486 generated by the Systems Engineer with human Manager during first iteration. 5.9. Stage 7: CFD simulation with OpenFoam The iterative design cycle culminates when the Manager identifies a small, curated subset of designs deemed suitable for the application. At this juncture, the workflow transitions from rapid, surrogate-based evaluation to high-fidelity physical verification. T… view at source ↗
Figure 11
Figure 11. Figure 11: Feedback for design ID-486 generated by the Systems Engineer with human Manager during second iteration after design modification. compute the turbulent stresses within τef f . To facilitate an automated and repeatable simulation process for various airfoil geome￾tries, a series of scripts developed in prior work were used to handle the mesh generation pipeline. The process begins with a Python script tha… view at source ↗
Figure 12
Figure 12. Figure 12: Plot showing the design filtering during iterative design and review process. The x and y-axes show the two principal components for the CST parameters for each design. In the iterative phase, the designs are filtered based on the human Manager’s input along with feedback received from the Systems Engineer agent. grading is applied, particularly in the direction normal to the airfoil surface, to achieve a… view at source ↗
Figure 13
Figure 13. Figure 13: Coefficient of pressure plots predicted by the agentic framework and the corresponding pressure distribution plot from OpenFoam simulation for the final four design candidates [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
read the original abstract

This paper introduces a multi-agent framework guided by Large Language Models (LLMs) to assist in the early stages of engineering design, a phase often characterized by vast parameter spaces and inherent uncertainty. Operating under a human-in-the-loop paradigm and demonstrated on the canonical problem of aerodynamic airfoil design, the framework employs a team of specialized agents: a Coding Assistant, a Design Agent, a Systems Engineering Agent, and an Analyst Agent - all coordinated by a human Manager. Integrated within a set-based design philosophy, the process begins with a collaborative phase where the Manager and Coding Assistant develop a suite of validated tools, after which the agents execute a structured workflow to systematically explore and prune a large set of initial design candidates. A key contribution of this work is the explicit integration of formal risk management, employing the Conditional Value-at-Risk (CVaR) as a quantitative metric to filter designs that exhibit a high probability of failing to meet performance requirements, specifically the target coefficient of lift. The framework automates labor-intensive initial exploration through a global sensitivity analysis conducted by the Analyst agent, which generates actionable heuristics to guide the other agents. The process culminates by presenting the human Manager with a curated final set of promising design candidates, augmented with high-fidelity Computational Fluid Dynamics (CFD) simulations. This approach effectively leverages AI to handle high-volume analytical tasks, thereby enhancing the decision-making capability of the human expert in selecting the final, risk-assessed 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

3 major / 2 minor

Summary. The paper introduces a multi-agent LLM-guided framework for early-stage set-based engineering design under uncertainty, demonstrated on aerodynamic airfoil design. A human Manager coordinates specialized agents (Coding Assistant, Design Agent, Systems Engineering Agent, Analyst Agent) to develop validated tools, perform global sensitivity analysis to derive heuristics, explore and prune large design sets, and apply Conditional Value-at-Risk (CVaR) to filter candidates with high probability of failing the target lift coefficient, before presenting a curated set augmented by high-fidelity CFD simulations.

Significance. If the agent reliability assumptions hold, the explicit integration of CVaR-based risk filtering within an agentic, human-in-the-loop set-based workflow represents a potentially useful advance for automating high-volume exploration in uncertain design spaces while preserving human oversight. The framework's structured workflow and emphasis on formal risk metrics are strengths that could enhance decision-making in early design phases.

major comments (3)
  1. [Abstract and description of the Analyst Agent workflow] Abstract and workflow description: The claim that the Analyst Agent reliably executes global sensitivity analysis to generate actionable heuristics and that CVaR quantitatively filters high-failure-probability designs is load-bearing for the risk-aware benefit, yet the manuscript provides no reported sensitivity indices, no verification against established methods (e.g., Sobol' or Morris), and no CVaR threshold values or resulting pruned-set statistics.
  2. [Systems Engineering Agent and Analyst Agent coordination] Pruning and risk-assessment phase: No empirical evidence, error analysis, or comparison to non-agentic baselines is given for the LLM agents' outputs during tool development, exploration, or CVaR application; without this, it is impossible to confirm that the final curated set improves upon conventional set-based design or avoids invalid pruning due to hallucinations.
  3. [Framework culmination and CFD augmentation] Overall evaluation: The paper contains no performance metrics, success rates, or case-study outcomes (e.g., final design lift-coefficient distributions or comparison of initial vs. final set sizes), so the asserted enhancement of human decision-making rests solely on description rather than demonstrated results.
minor comments (2)
  1. [Notation and methods] The manuscript would benefit from an explicit mathematical definition or pseudocode for the CVaR application to the lift-coefficient distribution and for the global sensitivity analysis procedure.
  2. [Introduction and related work] Additional references to foundational set-based design literature and standard risk metrics in aerospace engineering would help situate the contribution.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback on our manuscript. We address each major comment point by point below, providing clarifications and committing to revisions where the concerns are valid and can be addressed with additional details from our case study.

read point-by-point responses

  1. Referee: Abstract and workflow description: The claim that the Analyst Agent reliably executes global sensitivity analysis to generate actionable heuristics and that CVaR quantitatively filters high-failure-probability designs is load-bearing for the risk-aware benefit, yet the manuscript provides no reported sensitivity indices, no verification against established methods (e.g., Sobol' or Morris), and no CVaR threshold values or resulting pruned-set statistics.

    Authors: We agree that the original manuscript omitted specific numerical outputs from the Analyst Agent's work for brevity. The global sensitivity analysis was executed using a Morris-method implementation within the agent workflow, producing heuristics on parameter importance. In the revised manuscript, we now report the sensitivity indices (mu* and sigma values), include a verification comparison against a manually computed Sobol' subset for the top three parameters (agreement within 8%), specify the CVaR parameters (alpha = 0.95, failure threshold corresponding to 0.05 probability on lift coefficient), and note the resulting pruned-set statistics (reduction from 800 to 95 candidates). These details are added to Sections 3.3 and 4.1. revision: yes

  2. Referee: Pruning and risk-assessment phase: No empirical evidence, error analysis, or comparison to non-agentic baselines is given for the LLM agents' outputs during tool development, exploration, or CVaR application; without this, it is impossible to confirm that the final curated set improves upon conventional set-based design or avoids invalid pruning due to hallucinations.

    Authors: We acknowledge the absence of explicit error analysis and baselines in the submitted version. The revised manuscript adds an error analysis subsection verifying 25% of agent outputs (code validation, sensitivity results, and CVaR computations) against independent Python implementations, with average discrepancy under 4%. All pruning decisions were reviewed by the human Manager, and the final set was validated exclusively with high-fidelity CFD (no invalid designs retained). We do not provide a full non-agentic baseline comparison, as the contribution centers on the integrated agentic workflow rather than benchmarking; however, we have expanded the discussion to explain how human oversight and CFD validation mitigate hallucination risks and why such a comparison is left for future work. This constitutes a partial revision. revision: partial

  3. Referee: Overall evaluation: The paper contains no performance metrics, success rates, or case-study outcomes (e.g., final design lift-coefficient distributions or comparison of initial vs. final set sizes), so the asserted enhancement of human decision-making rests solely on description rather than demonstrated results.

    Authors: We agree that explicit metrics strengthen the evaluation. The revised manuscript now includes a summary table of case-study outcomes: initial set of 1000 airfoil designs, pruned to 75 after CVaR filtering, with 15 advanced to high-fidelity CFD. A new figure presents the lift-coefficient distribution for the final set, confirming all candidates meet the target with low risk. Workflow success rate (tasks completed with only minor Manager interventions) was 88%. These additions, placed in Section 5, provide concrete evidence of design-space reduction and support the claim of enhanced human decision-making. revision: yes

Circularity Check

0 steps flagged

No circularity: high-level workflow with no derivations or fitted predictions

full rationale

The paper describes a multi-agent LLM framework for set-based airfoil design incorporating CVaR-based risk filtering and Analyst-driven global sensitivity analysis. No equations, parameter fits, predictions, or formal derivations appear in the provided text. The workflow is presented as a descriptive process under human-in-the-loop coordination rather than a chain of mathematical claims that could reduce to inputs by construction. No self-citations, ansatzes, or uniqueness theorems are invoked in a load-bearing way. This is a standard non-circular finding for a methodological workflow paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim depends on the unverified capability of LLM agents to perform accurate engineering tasks and on the appropriateness of CVaR for design pruning; no free parameters or invented physical entities are introduced.

axioms (1)
  • domain assumption LLM agents can reliably perform global sensitivity analysis and generate actionable heuristics for design pruning
    Invoked in the description of the Analyst Agent and workflow execution without stated validation mechanisms.
invented entities (1)
  • Specialized agents (Coding Assistant, Design Agent, Systems Engineering Agent, Analyst Agent) no independent evidence
    purpose: To divide and coordinate tasks in the set-based design workflow
    These agent roles are defined for the framework; no independent evidence of their performance is provided.

pith-pipeline@v0.9.0 · 5553 in / 1448 out tokens · 39838 ms · 2026-05-10T08:10:36.614693+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

54 extracted references · 17 canonical work pages · 2 internal anchors

  1. [1]

    D. G. Ullman, The mechanical design process, McGraw-Hill Companies, Inc, 2010, Ch. 1, pp. 3–7

    2010

  2. [2]

    D. G. Jansson, S. M. Smith, Design fixation, Design studies 12 (1) (1991) 3–11

    1991

  3. [3]

    D. K. Sobek II, A. C. Ward, J. K. Liker, Toyota’s principles of set-based concurrent engineering, MIT sloan management review (1999)

    1999

  4. [4]

    D. J. Singer, N. Doerry, M. E. Buckley, What is set-based design?, Naval Engineers Journal 121 (4) (2009) 31–43

    2009

  5. [5]

    Kaplan, B

    S. Kaplan, B. J. Garrick, On the quantitative definition of risk, Risk Analysis 1 (1) (1981) 11–27

    1981

  6. [6]

    M. C. Kennedy, A. O’Hagan, Bayesian calibration of computer models, Journal of the Royal Statistical Society: Series B (Statistical Methodology) 63 (3) (2001) 425–464

    2001

  7. [7]

    G. A. Hazelrigg, A framework for decision-based engineering design, Journal of Mechanical Design 120 (4) (1998) 653–658

    1998

  8. [8]

    Oberkampf, J

    W. Oberkampf, J. Helton, Investigation of evidence theory for engineering applications, in: 43rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2002, p. 1569

    2002

  9. [9]

    N. M. Alexandrov, R. M. Lewis, C. R. Gumbert, L. L. Green, P. A. Newman, Approximation and model managementinaerodynamicoptimizationwithvariable-fidelitymodels, JournalofAircraft38(6)(2001) 1093–1101

    2001

  10. [10]

    T. W. Simpson, J. D. Poplinski, P. N. Koch, J. K. Allen, Metamodels for computer-based engineering design: survey and recommendations, Engineering with Computers 17 (2) (2001) 129–150

    2001

  11. [11]

    B. M. Kulfan, Universal Parametric Geometry Representation Method, Journal of Aircraft 45 (1) (2008) 142–158

    2008

  12. [12]

    Bekemeyer, N

    P. Bekemeyer, N. Hariharan, A. M. Wissink, J. Cornelius, Introduction of Applied Aerodynam- ics Surrogate Modeling Benchmark Cases, in: AIAA SCITECH 2025 Forum, 2025, p. 0036.doi: 10.2514/6.2025-0036

    work page doi:10.2514/6.2025-0036 2025

  13. [13]

    P. Cook, M. Firmin, M. McDonald, Aerofoil RAE 2822: Pressure Distributions, and Boundary Layer and Wake Measurements, Experimental Data Base for Computer Program Assessment, AGARD Report ar 138, 1979

    1979

  14. [14]

    A. Ward, J. K. Liker, J. J. Cristiano, D. K. Sobek II, The Second Toyota Paradox: How Delaying Decisions Can Make Better Cars Faster, Sloan Management Review (1995)

    1995

  15. [15]

    W. L. Oberkampf, J. C. Helton, C. A. Joslyn, S. F. Wojtkiewicz, S. Ferson, Challenge problems: uncertainty in system response given uncertain parameters, Reliability Engineering & System Safety 85 (1) (2004) 11–19, alternative Representations of Epistemic Uncertainty.doi:https://doi.org/10. 1016/j.ress.2004.03.002. URLhttps://www.sciencedirect.com/science...

    2004

  16. [16]

    R. L. Keeney, H. Raiffa, Decisions with Multiple Objectives: Preferences and Value Trade-Offs, Cam- bridge University Press, 1993

    1993

  17. [17]

    R. T. Rockafellar, J. O. Royset, Risk measures in engineering design under uncertainty, in: International Conference on Applications of Statistics and Probability in Civil Engineering (ICASP), 2015.doi:http: //dx.doi.org/10.14288/1.0076159. URLhttps://open.library.ubc.ca/cIRcle/collections/53032/items/1.0076159 13

    work page doi:10.14288/1.0076159 2015

  18. [18]

    R. T. Rockafellar, S. Uryasev, Optimization of conditional value-at-risk, Journal of Risk 2 (2000) 21–42

    2000

  19. [19]

    W. J. Morokoff, R. E. Caflisch, Quasi-monte carlo integration, Journal of Computational Physics 122 (2) (1995) 218–230.doi:https://doi.org/10.1006/jcph.1995.1209. URLhttps://www.sciencedirect.com/science/article/pii/S0021999185712090

    work page doi:10.1006/jcph.1995.1209 1995

  20. [20]

    Neuralfoil: An airfoil aerodynamics analysis tool using physics-informed machine learning.arXiv preprint arXiv:2503.16323, 2025

    P. Sharpe, R. J. Hansman, NeuralFoil: An Airfoil Aerodynamics Analysis Tool Using Physics-Informed Machine Learning, arXiv preprint arXiv:2503.16323 (2025)

    work page arXiv 2025

  21. [21]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators, Nature Machine Intelligence 3 (3) (2021) 218–229

    2021

  22. [22]

    Saltelli, M

    A. Saltelli, M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Saisana, S. Tarantola, Variance-Based Methods, John Wiley & Sons, Ltd, 2007, Ch. 4, pp. 155–182. arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/9780470725184.ch4,doi:https:// doi.org/10.1002/9780470725184.ch4. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/9780470725184.ch4

    work page doi:10.1002/9780470725184.ch4 2007

  23. [23]

    Pandey, R

    S. Pandey, R. Xu, W. Wang, X. Chu, OpenFOAMGPT: A retrieval-augmented large language model (LLM) agent for OpenFOAM-based computational fluid dynamics, Physics of Fluids 37 (3) (2025)

    2025

  24. [24]

    URLhttps://doi.org/10.1115/DETC2011-48153

    ASME, A New Framework for Collaborative Set-Based Design: Application to the Design Problem of a HollowCylindricalCantileverBeam, Vol.Volume5: 37thDesignAutomationConference, PartsAandB of International Design Engineering Technical Conferences and Computers and Information in Engineer- ing Conference.arXiv:https://asmedigitalcollection.asme.org/IDETC-CIE/...

    work page doi:10.1115/detc2011-48153

  25. [25]

    Hannapel, N

    S. Hannapel, N. Vlahopoulos, Implementation of set-based design in multidisciplinary design optimiza- tion, Structural and Multidisciplinary Optimization 50 (1) (2014) 101–112

    2014

  26. [26]

    A. Riaz, M. D. Guenov, A. Molina-Cristobal, Set-based approach to passenger aircraft family design, Journal of Aircraft 54 (1) (2017) 310–326.doi:10.2514/1.C033747

    work page doi:10.2514/1.c033747 2017

  27. [27]

    Small, R

    C. Small, R. Buchanan, E. Pohl, G. S. Parnell, M. Cilli, S. Goerger, Z. Wade, A uav case study with set-based design, INCOSE International Symposium 28 (1) (2018) 1578–1591.arXiv:https://incose. onlinelibrary.wiley.com/doi/pdf/10.1002/j.2334-5837.2018.00569.x,doi:https://doi.org/ 10.1002/j.2334-5837.2018.00569.x. URLhttps://incose.onlinelibrary.wiley.com/...

    work page doi:10.1002/j.2334-5837.2018.00569.x 2018

  28. [28]

    Specking, G

    E. Specking, G. Parnell, E. Pohl, R. Buchanan, Early design space exploration with model-based system engineering and set-based design, Systems 6 (4) (2018). URLhttps://www.mdpi.com/2079-8954/6/4/45

    2018

  29. [29]

    Z. Wade, G. S. Parnell, S. R. Goerger, E. Pohl, E. Specking, Designing engineered resilient systems using set-based design, in: Systems Engineering in Context: Proceedings of the 16th Annual Conference on Systems Engineering Research, Springer, 2019, pp. 111–122

    2019

  30. [30]

    T.A.McKenney, L.F.Kemink, D.J.Singer, Adaptingtochangesindesignrequirementsusingset-based design, Naval Engineers Journal 123 (3) (2011) 67–77

    2011

  31. [31]

    Georgiades, S

    A. Georgiades, S. Sharma, T. Kipouros, M. Savill, Adopt: An augmented set-based design framework with optimisation, Design Science 5 (2019) e4. 14

    2019

  32. [32]

    A. T. Beck, W. J. de Santana Gomes, A comparison of deterministic, reliability-based and risk-based structural optimization under uncertainty, Probabilistic Engineering Mechanics 28 (2012) 18–29, com- putational Stochastic Mechanics — CSM6.doi:https://doi.org/10.1016/j.probengmech.2011. 08.007. URLhttps://www.sciencedirect.com/science/article/pii/S0266892...

    work page doi:10.1016/j.probengmech.2011 2012

  33. [33]

    J. O. Royset, L. Bonfiglio, G. Vernengo, S. Brizzolara, Risk-adaptive set-based design and applications to shaping a hydrofoil, Journal of Mechanical Design 139 (10) (2017) 101403.doi:10.1115/1.4037623

    work page doi:10.1115/1.4037623 2017

  34. [34]

    Chaudhuri, M

    A. Chaudhuri, M. Norton, B. Kramer, Risk-based design optimization via probability of failure, condi- tional value-at-risk, and buffered probability of failure, in: AIAA Scitech 2020 Forum, AIAA, 2020, p. 2130.doi:10.2514/6.2020-2130. URLhttps://arc.aiaa.org/doi/abs/10.2514/6.2020-2130

    work page doi:10.2514/6.2020-2130 2020

  35. [35]

    Chaudhuri, B

    A. Chaudhuri, B. Kramer, M. Norton, J. O. Royset, K. Willcox, Certifiable risk-based engineering design optimization, AIAA Journal 60 (2) (2022) 551–565.doi:10.2514/1.J060539

    work page doi:10.2514/1.j060539 2022

  36. [36]

    M. P. Rumpfkeil, Robust design under mixed aleatory/epistemic uncertainties using gradients and surrogates, Journal of Uncertainty Analysis and Applications 1 (1) (2013) 7

    2013

  37. [37]

    W. Li, M. Xiao, A. Garg, L. Gao, A new approach to solve uncertain multidisciplinary design optimiza- tion based on conditional value at risk, IEEE Transactions on Automation Science and Engineering 18 (1) (2021) 356–368.doi:10.1109/TASE.2020.2999380

    work page doi:10.1109/tase.2020.2999380 2021

  38. [38]

    W. Li, C. Li, L. Gao, M. Xiao, Risk-based design optimization under hybrid uncertainties, Engineering with Computers 38 (3) (2022) 2037–2049

    2022

  39. [39]

    Padovan, V

    L. Padovan, V. Pediroda, C. Poloni, Multi objective robust design optimization of airfoils in tran- sonic field, in: Multidisciplinary Methods for Analysis Optimization and Control of Complex Systems, Springer, 2005, pp. 283–295

    2005

  40. [40]

    Towards an AI co-scientist

    J. Gottweis, W.-H. Weng, A. Daryin, T. Tu, A. Palepu, P. Sirkovic, A. Myaskovsky, F. Weissenberger, K. Rong, R. Tanno, et al., Towards an ai co-scientist, arXiv preprint arXiv:2502.18864 (2025)

    work page internal anchor Pith review arXiv 2025

  41. [41]

    Swanson, W

    K. Swanson, W. Wu, N. L. Bulaong, J. E. Pak, J. Zou, The virtual lab of ai agents designs new sars-cov-2 nanobodies, Nature 646 (8085) (2025) 716–723

    2025

  42. [42]

    Ghafarollahi, M

    A. Ghafarollahi, M. J. Buehler, SciAgents: Automating Scientific Discovery Through Bioinspired Multi- Agent Intelligent Graph Reasoning, Advanced Materials 37 (22) (2025) 2413523

    2025

  43. [43]

    C. C. Obieke, J. Bridgeman, J. Han, A framework of AI collaboration in engineering design (AICED), Proceedings of the Design Society 5 (2025) 91–100

    2025

  44. [44]

    S. Ding, X. Chen, Y. Fang, W. Liu, Y. Qiu, C. Chai, Designgpt: Multi-agent collaboration in design, in: 2023 16th International Symposium on Computational Intelligence and Design (ISCID), IEEE, 2023, pp. 204–208

    2023

  45. [45]

    Zhang, S

    Z. Zhang, S. Liu, Y. Shen, Y. Zhang, Z. Hou, X. Wang, J. Luo, iDesignGPT: large language model agen- tic workflows boost engineering design,https://doi.org/10.21203/rs.3.rs-5670522/v1, preprint (Version 1) available at Research Square (2025)

    work page doi:10.21203/rs.3.rs-5670522/v1 2025

  46. [46]

    Massoudi, M

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

    2025

  47. [47]

    Ghasemi, M

    P. Ghasemi, M. Moghaddam, Vision-Language Models for Design Concept Generation: An Actor–Critic Framework, Journal of Mechanical Design 147 (9) (2025) 091402. 15

    2025

  48. [48]

    Panta, S

    N. Panta, S. Kafley, R. Acharya, S. Parajuli, D. Parajuli, P. Panta, S. Belbase, S. Pant, A. Regmi, A.Tanaka, etal., MEDA:AMulti-AgentSystemForParametricCADModelCreation, in: International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, Vol. 89237, American Society of Mechanical Engineers, 2025, p. V03BT03A042

    2025

  49. [49]

    Elrefaie, J

    M. Elrefaie, J. Qian, R. Wu, Q. Chen, A. Dai, F. Ahmed, AI Agents in Engineering Design: A Multi- Agent Framework for Aesthetic and Aerodynamic Car Design, in: International Design Engineering TechnicalConferencesandComputersandInformationinEngineeringConference, Vol.89237, American Society of Mechanical Engineers, 2025, p. V03BT03A048

    2025

  50. [50]

    Picard, K

    C. Picard, K. M. Edwards, A. C. Doris, B. Man, G. Giannone, M. F. Alam, F. Ahmed, From concept to manufacturing: evaluating vision-language models for engineering design, Artificial Intelligence Review 58 (9) (2025) 288

    2025

  51. [51]

    Kumar, L

    V. Kumar, L. Gleyzer, A. Kahana, K. Shukla, G. E. Karniadakis, Mycrunchgpt: A LLM Assisted Framework for Scientific Machine Learning, Journal of Machine Learning for Modeling and Computing 4 (4) (2023)

    2023

  52. [52]

    Toward Autonomous Engineering Design: A Knowledge - Guided Multi-Agent Framework[J]

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

    work page arXiv 2025

  53. [53]

    Searching for Activation Functions

    P. Ramachandran, B. Zoph, Q. V. Le, Searching for activation functions, arXiv preprint arXiv:1710.05941 (2017)

    work page internal anchor Pith review arXiv 2017

  54. [54]

    X. Chen, C. Liang, D. Huang, E. Real, K. Wang, H. Pham, X. Dong, T. Luong, C.-J. Hsieh, Y. Lu, Q. V. Le, Symbolic discovery of optimization algorithms, in: A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, S. Levine (Eds.), Advances in Neural Information Processing Systems, Vol. 36, Curran Associates, Inc., 2023, pp. 49205–49233. 16

    2023