pith. sign in

arxiv: 2606.06405 · v1 · pith:AQN3J7XEnew · submitted 2026-06-04 · 💻 cs.CE · cs.CG

Bridging CAD and Data-Driven Design: Attributed Feature Graphs for Engineering Design

Abhishek Indupally , Ibraheem Alawadhi , Satchit Ramnath , Jami J. Shah This is my paper

Pith reviewed 2026-06-27 22:53 UTC · model grok-4.3

classification 💻 cs.CE cs.CG
keywords Attributed Feature GraphsCAD designGraph Neural Networkssurrogate modelingparametric designfeature-based representationengineering optimization
0
0 comments X

The pith

Attributed Feature Graphs allow graph neural networks to learn from CAD designs while preserving parametric editability and interpretability.

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

The paper proposes Attributed Feature Graphs to represent engineering designs in a form that supports both machine learning and traditional CAD workflows. It encodes individual design features such as extrusions and ribs as nodes, with their relations as edges, keeping the original parametric structure intact. This enables training graph neural networks directly on the feature graphs to predict performance metrics like those in automotive hood designs. The result is a surrogate model that performs competitively with other data-driven methods but allows engineers to see how specific features influence outcomes and to make changes in the CAD environment that feed back into the model.

Core claim

Attributed Feature Graphs (AFGs) represent designs by treating CAD features as nodes and their geometric or dependency relations as directed edges. This structure preserves design intent and parametric information, making it compatible with graph-based learning. When used with a Graph Neural Network on the CarHoods10K dataset, the model predicts performance metrics competitively while permitting mapping of results back to specific features for interpretation and enabling reevaluation after CAD edits.

What carries the argument

Attributed Feature Graphs (AFGs), a representation where CAD design features are nodes and geometric or dependency relations are directed edges, carrying parametric structure into graph learning.

If this is right

  • Engineers can use the trained GNN as a fast evaluation engine during design iteration.
  • Performance predictions can be attributed to individual design features for better understanding.
  • Changes made to the CAD model can be directly reflected in the graph and reevaluated without additional preprocessing.
  • The approach maintains compatibility with standard GNN methods while retaining feature-level semantics.

Where Pith is reading between the lines

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

  • AFGs might extend to optimization loops where feature parameters are adjusted based on model gradients.
  • Similar feature-graph approaches could apply to other simulation-heavy fields like structural analysis beyond automotive parts.
  • By avoiding mesh conversion, this could lower computational overhead in design exploration compared to traditional geometric deep learning.

Load-bearing premise

That the information in the attributed feature graph from native CAD features is sufficient for a GNN to achieve accurate performance predictions without losing key details compared to direct geometric representations.

What would settle it

If a GNN trained on AFGs from the CarHoods10K dataset produces substantially less accurate predictions of hood performance metrics than equivalent models using voxel or mesh inputs, the claim of sufficient information retention would be falsified.

Figures

Figures reproduced from arXiv: 2606.06405 by Abhishek Indupally, Ibraheem Alawadhi, Jami J. Shah, Satchit Ramnath.

Figure 1
Figure 1. Figure 1: Different Types of 3D Representations [1] [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Feature-Based Design Approach [39] 2.1.2 Implicit Representations Beyond explicit geometric encodings such as B-reps, meshes, and point clouds—long-established in CAD for their editability but limited by resolution and topologi￾cal rigidity—implicit representations provide continuous, differentiable alternatives that enhance topology flexibil￾ity and compatibility with ML-driven design workflows. These rep… view at source ↗
Figure 3
Figure 3. Figure 3: Outer Skin and Hood Frame 3 Methodology: Attributed Feature Graphs Feature-based CAD modeling constructs complex geome￾try through parametric sequences of meaningful engineer￾ing operations (e.g., extrusions for ribs, offsets for depres￾sions, cuts for pockets), organized as hierarchical feature trees that preserve shape and design intent. These trees encode manufacturing intent through their tree-like top… view at source ↗
Figure 4
Figure 4. Figure 4: Features Defined on a Hood Frame This process yields a compact, semantically consistent graph representation in which each node corresponds to a mechanically meaningful feature, and each edge reflects a verifiable CAD support or dependency relationship. Be￾cause each node corresponds to a specific CAD feature, learned node and graph embeddings can be mapped back to individual features and their hierarchica… view at source ↗
Figure 5
Figure 5. Figure 5: Feature Classification and Taxonomy • Front Flats: They are secondary features that allocate space for a hood lock. • Hinge Flats: They are secondary features that allocate space for side hinges. • Front Rib: This is a secondary rib located on the front side of the hood frame. • Rear Rib: This is a secondary rib located on the back side of the hood frame. • Middle Depression: This is a first-level node who… view at source ↗
Figure 7
Figure 7. Figure 7: AFG Structure for a Hood Frame; Nodes corre [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the proposed multi-task graph neu [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Data Distributions Training computes per-target MSE against ground truth, then takes a weighted sum before the backward pass: L = ∑︁ 3 𝑖=1 𝑤𝑖 · MSE(𝑦metric𝑖 , 𝑦ˆmetric𝑖 ) (4) where metric ∈ [stress, mass, deflection], 𝑦metric𝑖 is the predicted value, 𝑦ˆmetric𝑖 is the actual value, and MSE is the 11 [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Test-set 𝑅 2 distributions for maximum stress, mass, and directional deflection predictions by the Evaluation Engine [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Engineering Design Workflow using AFGs ML model can be applied. • Gradient-based analyses over AFGs can be misleading under out-of-distribution designs, as learned gradients may not reflect true physical or design sensitivities in unseen regions. • Representation quality is limited by data coverage; richer datasets spanning more design families and manufactur￾ing constraints are needed. • Requires standar… view at source ↗
read the original abstract

Engineering design is an iterative, simulation-driven process where traditional workflows rely heavily on computationally expensive analyses such as finite element and computational fluid dynamics. Although data-driven methods have accelerated design evaluation and optimization, most existing geometric representations discard parametric and feature-level semantics, limiting their integration with CAD-driven design workflows and reducing model interpretability. To address this gap, this work introduces Attributed Feature Graphs (AFGs), a feature-based representation that encodes design features, such as extrusions, ribs, and pockets, as nodes and their geometric or dependency relations as directed edges. AFGs preserve design intent and parametric structure while remaining compatible with standard graph-based learning methods, enabling end-to-end learning directly on CAD-derived feature graphs. The paper demonstrates the proposed representation through a surrogate-modeling case study on the CarHoods10K automotive hood frame dataset, where a Graph Neural Network (GNN) is trained as an evaluation engine to predict performance metrics from AFG inputs. The learned model achieves competitive surrogate performance compared with traditional data-driven approaches, but with the added benefit that engineers can map predictions back to specific CAD features and interpret how individual design elements influence system behavior. Furthermore, because AFGs are built from native CAD features, engineers can directly edit the underlying geometry in the CAD environment and reevaluate the design through the same learned model.

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 / 1 minor

Summary. The paper proposes Attributed Feature Graphs (AFGs) to represent engineering designs by encoding CAD features (e.g., extrusions, ribs, pockets) as nodes and their geometric or dependency relations as directed edges. It applies Graph Neural Networks to AFGs derived from the CarHoods10K automotive hood frame dataset to train a surrogate model predicting performance metrics, claiming competitive accuracy relative to traditional data-driven approaches together with improved interpretability (mapping predictions back to specific CAD features) and direct compatibility with CAD editing workflows.

Significance. If the central empirical claim holds, the work would provide a practical bridge between native CAD representations and graph-based machine learning, enabling interpretable, end-to-end surrogate modeling that preserves design intent and supports iterative CAD-driven optimization without requiring conversion to mesh or B-rep formats.

major comments (2)
  1. [Abstract] Abstract: the assertion that the GNN 'achieves competitive surrogate performance' supplies no quantitative metrics (MAE, RMSE, R², etc.), no baseline comparisons (mesh-based, point-cloud, or B-rep surrogates), no error bars, and no statistical tests, so it is impossible to evaluate whether AFGs incur material geometric information loss on CarHoods10K as required by the central claim.
  2. [Abstract] Abstract: the description of AFG construction and the subsequent GNN training provides no information on node/edge attribute definitions, GNN architecture, loss function, training procedure, or data splits, all of which are load-bearing for assessing whether the reported performance is reproducible or actually competitive.
minor comments (1)
  1. The abstract would be strengthened by a single sentence stating the concrete performance numbers and the exact baselines used, even if full tables appear later in the manuscript.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these precise observations on the abstract. Both comments correctly identify that the current abstract is insufficiently quantitative and lacks key methodological specifics, which limits evaluation of the central claims. We will revise the abstract to incorporate the requested information while preserving its brevity.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that the GNN 'achieves competitive surrogate performance' supplies no quantitative metrics (MAE, RMSE, R², etc.), no baseline comparisons (mesh-based, point-cloud, or B-rep surrogates), no error bars, and no statistical tests, so it is impossible to evaluate whether AFGs incur material geometric information loss on CarHoods10K as required by the central claim.

    Authors: We agree that the abstract must supply concrete metrics to substantiate the claim. In the revision we will insert the key performance numbers (MAE, RMSE, R²) obtained on the CarHoods10K test set, together with the corresponding baseline results, error bars, and a brief statement on statistical significance. These values are already computed and reported in Section 4; the abstract will now summarize them explicitly. revision: yes

  2. Referee: [Abstract] Abstract: the description of AFG construction and the subsequent GNN training provides no information on node/edge attribute definitions, GNN architecture, loss function, training procedure, or data splits, all of which are load-bearing for assessing whether the reported performance is reproducible or actually competitive.

    Authors: We accept that the abstract omits these details. The full specifications appear in Sections 3.2 (AFG node/edge attributes), 4.1 (GNN architecture), 4.2 (loss and optimizer), 4.3 (training procedure), and 4.4 (data splits). For the revised abstract we will add a concise clause summarizing the attribute definitions, the GNN type and depth, the loss function, and the train/validation/test split ratios, thereby making the abstract self-contained for reproducibility assessment. revision: yes

Circularity Check

0 steps flagged

No significant circularity; AFG definition and GNN surrogate are independent constructions

full rationale

The paper defines Attributed Feature Graphs (AFGs) by encoding native CAD features (extrusions, ribs, pockets) as nodes with geometric/dependency edges, then applies standard GNNs for surrogate prediction on CarHoods10K. This is a representational choice followed by empirical evaluation, with no equations, fitted parameters, or self-citations that reduce any claimed prediction or uniqueness result back to the paper's own inputs by construction. The abstract and description contain no load-bearing self-references or ansatzes smuggled via citation; performance is asserted via experiment rather than tautology. This matches the default non-circular outcome.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper's contribution is the introduction of the AFG representation itself. No explicit free parameters are described in the abstract. The main supporting assumption is that GNNs can learn useful mappings from the constructed graphs.

axioms (1)
  • domain assumption Graph neural networks can learn mappings from attributed feature graphs to engineering performance metrics.
    Invoked when the abstract states that a GNN is trained as an evaluation engine on AFG inputs.
invented entities (1)
  • Attributed Feature Graph (AFG) no independent evidence
    purpose: To encode CAD design features (extrusions, ribs, pockets) as nodes and their geometric or dependency relations as directed edges while preserving parametric structure.
    New representation introduced to bridge CAD and data-driven methods; no external falsifiable evidence is supplied in the abstract.

pith-pipeline@v0.9.1-grok · 5785 in / 1447 out tokens · 27320 ms · 2026-06-27T22:53:22.731639+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

70 extracted references · 39 canonical work pages

  1. [1]

    Single-view 3d reconstruction: A sur- vey of deep learning methods

    George Fahim, Khalid Amin, and Sameh Zarif. Single-view 3d reconstruction: A sur- vey of deep learning methods. 94:164–190. ISSN 0097-8493. doi: 10.1016/j.cag.2020.12

    work page doi:10.1016/j.cag.2020.12 2020

  2. [2]

    URLhttps://www.sciencedirect.com/ science/article/pii/S0097849320301849

  3. [3]

    Mapping artificial intelligence- based methods to engineering design stages: a focused literature review

    Pranav Milind Khanolkar, Ademir Vrolijk, and Alison Olechowski. Mapping artificial intelligence- based methods to engineering design stages: a focused literature review. 37:e25. ISSN 0890-0604, 1469-1760. doi: 10.1017/S0890060423000203. URLhttps://www.cambridge.org/core/ product/identifier/S0890060423000203/ type/journal_article

    work page doi:10.1017/s0890060423000203

  4. [4]

    Camba, Manuel Contero, and Pedro Com- pany

    Jorge D. Camba, Manuel Contero, and Pedro Com- pany. Parametric CAD modeling: An analy- sis of strategies for design reusability. 74:18–

  5. [5]

    doi: 10.1016/j.cad.2016

    ISSN 00104485. doi: 10.1016/j.cad.2016. 01.003. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0010448516000051

    work page doi:10.1016/j.cad.2016 2016

  6. [6]

    A study on the extraction of geo- metrical parameters from flexible mechanical com- ponents and assemblies and their impact on per- formance: A machine learning approach

    Prakash Kumar. A study on the extraction of geo- metrical parameters from flexible mechanical com- ponents and assemblies and their impact on per- formance: A machine learning approach. URL https://keep.lib.asu.edu/items/195239

  7. [9]

    De- veloping multi-modal machine learning model for predicting performance of automotive hood frames

    Abhishek Indupally and Satchit Ramnath. De- veloping multi-modal machine learning model for predicting performance of automotive hood frames. American Society of Mechanical En- gineers Digital Collection. doi: 10.1115/ DETC2025-168841. URLhttps://dx.doi.org/ 10.1115/DETC2025-168841

    work page doi:10.1115/detc2025-168841

  8. [10]

    A graph neural network approach for product rela- tionship prediction

    Faez Ahmed, Yaxin Cui, Yan Fu, and Wei Chen. A graph neural network approach for product rela- tionship prediction. American Society of Mechan- ical Engineers Digital Collection. doi: 10.1115/ DETC2021-69462. URLhttps://dx.doi.org/ 10.1115/DETC2021-69462

    work page doi:10.1115/detc2021-69462

  9. [11]

    OrthoCAD-322k: A cross-modal approach for retrieving 3d CAD mod- els from orthographic views using a graph-based framework on a developed large-scale dataset

    SwapnilNagnathMahajan,KarthikKrishnaM.,and Ramanathan Muthuganapathy. OrthoCAD-322k: A cross-modal approach for retrieving 3d CAD mod- els from orthographic views using a graph-based framework on a developed large-scale dataset. 132: 104357. ISSN 00978493. doi: 10.1016/j.cag.2025. 104357. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0097849325001980

    work page doi:10.1016/j.cag.2025 2025

  10. [12]

    GraphBRep: Explicit graph diffusion of b–rep topology for efficient CAD generation

    Weilin Lai, Tie Xu, Ben Niu, and Hu Wang. GraphBRep: Explicit graph diffusion of b–rep topology for efficient CAD generation. 13(1): 259–274. ISSN 2288-5048. doi: 10.1093/jcde/ qwaf136. URLhttps://academic.oup.com/ jcde/article/13/1/259/8381229

    work page doi:10.1093/jcde/

  11. [13]

    BrepGPT: Au- toregressive b-rep generation with voronoi half- patch

    Pu Li, Wenhao Zhang, Weize Quan, Biao Zhang, Peter Wonka, and Dongming Yan. BrepGPT: Au- toregressive b-rep generation with voronoi half- patch. 44(6):226:1–226:18, . ISSN 0730-0301. doi: 10.1145/3763323. URLhttps://dl.acm. org/doi/10.1145/3763323

    work page doi:10.1145/3763323

  12. [14]

    Nor- malNet: A voxel-based CNN for 3d ob- ject classification and retrieval

    Cheng Wang, Ming Cheng, Ferdous Sohel, Mo- hammed Bennamoun, and Jonathan Li. Nor- malNet: A voxel-based CNN for 3d ob- ject classification and retrieval. 323:139–147. ISSN 09252312. doi: 10.1016/j.neucom.2018. 09.075. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0925231218311561

    work page doi:10.1016/j.neucom.2018 2018

  13. [15]

    VoxNet: A 3d convolutional neural network for real-time object recognition

    Daniel Maturana and Sebastian Scherer. VoxNet: A 3d convolutional neural network for real-time object recognition. In2015 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS), pages 922–928. IEEE. ISBN 978-1-4799-9994-1. doi: 10.1109/IROS.2015.7353481. URLhttp: //ieeexplore.ieee.org/document/7353481/

    work page doi:10.1109/iros.2015.7353481 2015

  14. [16]

    Shah, Duane Detwiler, and Stefan Menzel

    Patricia Wollstadt, Mariusz Bujny, Satchit Ram- nath, Jami J. Shah, Duane Detwiler, and Stefan Menzel. CarHoods10k: An industry-grade data 15 set for representation learning and design opti- mization in engineering applications. 26(6):1221–

  15. [17]

    doi: 10.1109/TEVC.2022.3147013

    ISSN 1089-778X, 1089-778X, 1941-0026. doi: 10.1109/TEVC.2022.3147013. URLhttps: //ieeexplore.ieee.org/document/9696004/

    work page doi:10.1109/tevc.2022.3147013 1941

  16. [18]

    Dryad | data: OSU-honda automobile hood dataset (CarHoods10k),.URLhttps://datadryad.org/ dataset/doi:10.5061/dryad.2fqz612pt

    work page doi:10.5061/dryad.2fqz612pt

  17. [19]

    Lambourne, Karl D

    PradeepKumarJayaraman,AdityaSanghi,JosephG. Lambourne, Karl D. D. Willis, Thomas Davies, Hooman Shayani, and Nigel Morris. UV-net: Learning from boundary representations. In2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11698–11707. IEEE, . ISBN 978-1-6654-4509-2. doi: 10. 1109/CVPR46437.2021.01153. URLhttps:// ieeexplor...

    arXiv 2021

  18. [20]

    Pradeep Kumar Jayaraman, Joseph George Lam- bourne, Nishkrit Desai, Karl Willis, Aditya Sanghi, and Nigel J. W. Morris. SolidGen: An autoregres- sive model for direct b-rep synthesis. . ISSN 2835-

  19. [21]

    URLhttps://openreview.net/forum? id=ZR2CDgADRo

  20. [22]

    Lambourne, Pradeep Kumar Jayaraman,ZhengqingWang,KarlD.D.Willis,and Yasutaka Furukawa

    Xiang Xu, Joseph G. Lambourne, Pradeep Kumar Jayaraman,ZhengqingWang,KarlD.D.Willis,and Yasutaka Furukawa. BrepGen: A b-rep generative diffusion model with structured latent geometry, . URLhttp://arxiv.org/abs/2401.15563

    arXiv

  21. [23]

    MeshPointNet: 3d surface classification using graph neural net- worksandconformalpredictionsonmesh-basedrep- resentations

    Amin Heyrani Nobari, Justin Rey, Suhas Kodali, Matthew Jones, and Faez Ahmed. MeshPointNet: 3d surface classification using graph neural net- worksandconformalpredictionsonmesh-basedrep- resentations. 146(51712). ISSN 1050-0472. doi: 10.1115/1.4064673. URLhttps://doi.org/10. 1115/1.4064673

    work page doi:10.1115/1.4064673

  22. [24]

    A hybrid GNN approach for predicting node data for 3d meshes

    Shwetha Salimath, Francesca Bugiotti, and Frederic Magoules. A hybrid GNN approach for predicting node data for 3d meshes. URLhttp://arxiv. org/abs/2310.14707

    arXiv

  23. [25]

    MeshCNN: A network with an edge

    Rana Hanocka, Amir Hertz, Noa Fish, Raja Giryes, Shachar Fleishman, and Daniel Cohen- Or. MeshCNN: A network with an edge. 38 (4):1–12. ISSN 0730-0301, 1557-7368. doi: 10.1145/3306346.3322959. URLhttp://arxiv. org/abs/1809.05910

    work page doi:10.1145/3306346.3322959

  24. [26]

    2024 , url =

    Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, TatianaTommasi,DanieleSirigatti,VladislavRosov, AngelaDai, andMatthiasNießner. MeshGPT:Gen- erating triangle meshes with decoder-only trans- formers. In2024 IEEE/CVF Conference on Com- puterVisionandPatternRecognition(CVPR),pages 19615–19625.IEEE.ISBN979-8-3503-5300-6.doi: 10.1109/CVPR52733.2024.01855. URL...

    work page doi:10.1109/cvpr52733.2024.01855 2024

  25. [27]

    R.QiCharles,HaoSu,MoKaichun,andLeonidasJ. Guibas. PointNet: Deeplearningonpointsetsfor3d classification and segmentation. In2017 IEEE Con- ferenceonComputerVisionandPatternRecognition (CVPR), pages 77–85. IEEE. ISBN 978-1-5386- 0457-1. doi: 10.1109/CVPR.2017.16. URLhttp: //ieeexplore.ieee.org/document/8099499/

    work page doi:10.1109/cvpr.2017.16 2017

  26. [28]

    Mesh-drivenresamplingandregularizationfor robust point cloud-based flow analysis directly on scanned objects

    Monu Jaiswal, Ashton M Corpuz, and Ming-Chen Hsu. Mesh-drivenresamplingandregularizationfor robust point cloud-based flow analysis directly on scanned objects

  27. [29]

    Melkote, and David W

    Xiaofang Liu, Zhichao Wang, Shreyes N. Melkote, and David W. Rosen. Manufacturing pro- cess identification from 3d point cloud mod- els using semantic segmentation. 82:858–873, . ISSN 02786125. doi: 10.1016/j.jmsy.2025. 07.023. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0278612525001980

    work page doi:10.1016/j.jmsy.2025 2025

  28. [30]

    Chapman, Dongpu Cao, and Jonathan Li

    Ying Li, Lingfei Ma, Zilong Zhong, Fei Liu, Michael A. Chapman, Dongpu Cao, and Jonathan Li. Deep learning for LiDAR point clouds in au- tonomous driving: A review. 32(8):3412–3432, . ISSN 2162-2388. doi: 10.1109/TNNLS.2020. 3015992.URLhttps://ieeexplore.ieee.org/ document/9173706/

    work page doi:10.1109/tnnls.2020 2020

  29. [31]

    2024 , url =

    Xuanchi Ren, Jiahui Huang, Xiaohui Zeng, Ken Museth,SanjaFidler,andFrancisWilliams. XCube: Large-scale 3d generative modeling using sparse voxel hierarchies. In2024 IEEE/CVF Confer- ence on Computer Vision and Pattern Recogni- tion (CVPR), pages 4209–4219. IEEE. ISBN 979- 8-3503-5300-6. doi: 10.1109/CVPR52733.2024. 00403. URLhttps://ieeexplore.ieee.org/ d...

    work page doi:10.1109/cvpr52733.2024 2024

  30. [32]

    A history-based parametric CAD sketch dataset with advanced engineering commands

    Rubin Fan, Fazhi He, Yuxin Liu, and Jing Lin. A history-based parametric CAD sketch dataset with advanced engineering commands. 182:103848. ISSN 0010-4485. doi: 10.1016/j.cad.2025.103848. URLhttps://www.sciencedirect.com/ science/article/pii/S0010448525000107

    work page doi:10.1016/j.cad.2025.103848 2025

  31. [33]

    AI-based knowl- edge extraction for automatic design proposals using design-related patterns

    Carmen Krahe, Maksym Kalaidov, Markus Doel- lken, Thomas Gwosch, Andreas Kuhnle, Gisela 16 Lanza, and Sven Matthiesen. AI-based knowl- edge extraction for automatic design proposals using design-related patterns. 100:397–402, . ISSN 22128271. doi: 10.1016/j.procir.2021.05

    work page doi:10.1016/j.procir.2021.05 2021

  32. [34]

    com/retrieve/pii/S2212827121005618

    URLhttps://linkinghub.elsevier. com/retrieve/pii/S2212827121005618

  33. [35]

    URL https://www.ntop.com/resources/blog/ implicit-modeling-for-mechanical-design/

    Implicit modeling for engineering design, . URL https://www.ntop.com/resources/blog/ implicit-modeling-for-mechanical-design/

  34. [36]

    URL https://altair.com/blog/articles/ let-s-talk-about-implicit-modeling

    Let’s talk about implicit modeling, . URL https://altair.com/blog/articles/ let-s-talk-about-implicit-modeling

  35. [37]

    DeepSDF: Learning continuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. DeepSDF: Learning continuous signed distance functions for shape representation. URLhttp: //arxiv.org/abs/1901.05103

    Pith/arXiv arXiv 1901

  36. [38]

    Mechanics simulation with implicit neural repre- sentations of complex geometries

    Samundra Karki, Ming-Chen Hsu, Adarsh Kr- ishnamurthy, and Baskar Ganapathysubramanian. Mechanics simulation with implicit neural repre- sentations of complex geometries. 190:103978, . ISSN 00104485. doi: 10.1016/j.cad.2025. 103978. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0010448525001393

    work page doi:10.1016/j.cad.2025 2025

  37. [39]

    Point2cad: Reverse engi- neering CAD models from 3d point clouds

    Yujia Liu, Anton Obukhov, Jan Dirk Wegner, and Konrad Schindler. Point2cad: Reverse engi- neering CAD models from 3d point clouds. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3763–

    2024

  38. [40]

    2024 , url =

    IEEE, . ISBN 979-8-3503-5300-6. doi: 10.1109/CVPR52733.2024.00361. URLhttps:// ieeexplore.ieee.org/document/10657440/

    work page doi:10.1109/cvpr52733.2024.00361 2024

  39. [41]

    B-rep distance func- tions (BR-DF): How to represent a b-rep model by volumetric distance functions?,

    Fuyang Zhang, Pradeep Kumar Jayaraman, Xiang Xu, and Yasutaka Furukawa. B-rep distance func- tions (BR-DF): How to represent a b-rep model by volumetric distance functions?, . URLhttp: //arxiv.org/abs/2511.14870

    arXiv

  40. [42]

    HybridSDF: Combining deep implicit shapes and geometric primitives for 3d shape representation and manipulation

    Subeesh Vasu, Nicolas Talabot, Artem Lukoianov, Pierre Baqué, Jonathan Donier, and Pascal Fua. HybridSDF: Combining deep implicit shapes and geometric primitives for 3d shape representation and manipulation. In2022 In- ternational Conference on 3D Vision (3DV), pages 617–626. doi: 10.1109/3DV57658.2022. 00072. URLhttps://ieeexplore.ieee.org/ document/1004...

    work page doi:10.1109/3dv57658.2022 2022

  41. [43]

    Reviewofresearchinfeature-based design

    OWSalomons. Reviewofresearchinfeature-based design. 12(2)

  42. [44]

    Shah and Martti Mantyla.Parametric and Feature Based CAD/Cam: Concepts, Techniques, and Applications

    Jami J. Shah and Martti Mantyla.Parametric and Feature Based CAD/Cam: Concepts, Techniques, and Applications. John Wiley & Sons, Inc., 1st edi- tion. ISBN 978-0-471-00214-7

  43. [45]

    A scheme to represent features, dimensions, and tolerances in ge- ometric modeling

    Nanxin Wang and Tulga M Ozsoy. A scheme to represent features, dimensions, and tolerances in ge- ometric modeling. 10(3)

  44. [46]

    Feature- based simplification of boundary representa- tion models using sequential iterative volume decomposition

    Byung Chul Kim and Duhwan Mun. Feature- based simplification of boundary representa- tion models using sequential iterative volume decomposition. 38:97–107. ISSN 0097-

  45. [47]

    URL https://www.sciencedirect.com/science/ article/pii/S0097849313001726

    doi: 10.1016/j.cag.2013.10.031. URL https://www.sciencedirect.com/science/ article/pii/S0097849313001726

    work page doi:10.1016/j.cag.2013.10.031 2013

  46. [48]

    An implicit surface poly- gonizer

    Jules Bloomenthal. An implicit surface poly- gonizer. InGraphics Gems, pages 324–

  47. [49]

    ISBN 978-0-12-336156-1

    Elsevier. ISBN 978-0-12-336156-1. doi: 10.1016/B978-0-12-336156-1.50040-9. URLhttps://linkinghub.elsevier.com/ retrieve/pii/B9780123361561500409

    work page doi:10.1016/b978-0-12-336156-1.50040-9

  48. [50]

    Bloomenthal and B

    J. Bloomenthal and B. Wyvill. In- troduction to implicit surfaces. URL https://www.semanticscholar.org/paper/ Introduction-to-Implicit-Surfaces-Bloomenthal-Wyvill/ 985c64b1b2dad262ba05ad88c6769bd68cf3fcbc

  49. [51]

    Neu- ral implicit representations for multi-view surface reconstruction: A survey

    Xinyun Zhang, Ruiqi Yu, and Shuang Ren. Neu- ral implicit representations for multi-view surface reconstruction: A survey. 31(10):9444–9463, . ISSN 1941-0506. doi: 10.1109/TVCG.2025. 3582627.URLhttps://ieeexplore.ieee.org/ document/11051009/

    work page doi:10.1109/tvcg.2025 1941

  50. [52]

    C.M. Grimm. Simple manifolds for surface mod- eling and parameterization. InProceedings SMI. ShapeModelingInternational2002,pages237–277. doi: 10.1109/SMI.2002.1003551. URLhttps: //ieeexplore.ieee.org/document/1003551/

    work page doi:10.1109/smi.2002.1003551 2002

  51. [53]

    Neural implicit representations for engineering design - ProQuest

    Jaya Manideep Rebbagondla. Neural implicit representations for engineering design - ProQuest. URLhttps://www.proquest.com/docview/ 3132852644?%20Theses&fromopenview= true&pq-origsite=gscholar&sourcetype= Dissertations%20

  52. [54]

    Direct flow simulations with implicit neural rep- resentation of complex geometry

    Samundra Karki, Mehdi Shadkhah, Cheng-Hau Yang, Aditya Balu, Guglielmo Scovazzi, Adarsh Krishnamurthy, andBaskarGanapathysubramanian. Direct flow simulations with implicit neural rep- resentation of complex geometry. 446:118248, . ISSN 00457825. doi: 10.1016/j.cma.2025. 118248. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0045782525005201. 17

    work page doi:10.1016/j.cma.2025 2025

  53. [55]

    Implicit geometry neural net- work for mesh generation

    Ran Xu, Hongqiang Lyu, Jian Yu, Chenyu Bao, Hongfei Wang, Yufei Liu, and Xue- jun Liu. Implicit geometry neural net- work for mesh generation. 38(4):103315, . ISSN 1000-9361. doi: 10.1016/j.cja.2024.11

    work page doi:10.1016/j.cja.2024.11 2024

  54. [56]

    URLhttps://www.sciencedirect.com/ science/article/pii/S1000936124004692

  55. [57]

    Local deep im- plicit functions for 3d shape

    Kyle Genova, Forrester Cole, Avneesh Sud, Aaron Sarna, and Thomas Funkhouser. Local deep im- plicit functions for 3d shape. In2020 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR), pages 4856–4865. IEEE. ISBN 978- 1-7281-7168-5. doi: 10.1109/CVPR42600.2020. 00491. URLhttps://ieeexplore.ieee.org/ document/9157823/

    work page doi:10.1109/cvpr42600.2020 2020

  56. [58]

    Neural sparse voxel fields.InAdvancesinNeuralInformationProcessing Systems, volume 33, pages 15651–15663

    Lingjie Liu, Jiatao Gu, Kyaw Zaw Lin, Tat-Seng Chua, and Christian Theobalt. Neural sparse voxel fields.InAdvancesinNeuralInformationProcessing Systems, volume 33, pages 15651–15663. Curran Associates, Inc., . URLhttps://proceedings. neurips.cc/paper_files/paper/2020/hash/ b4b758962f17808746e9bb832a6fa4b8-Abstract. html

    2020

  57. [59]

    144(2):020301

    Special issue: Artificial intelligence and engi- neering design. 144(2):020301. ISSN 1050- 0472, 1528-9001. doi: 10.1115/1.4053111. URLhttps://asmedigitalcollection. asme.org/mechanicaldesign/ article/144/2/020301/1128870/ Special-Issue-Artificial-Intelligence-and

    work page doi:10.1115/1.4053111

  58. [60]

    Review of artificial intelligence applications in engineering design perspective

    Nurullah Yüksel, Hüseyin Rıza Börklü, Hüseyin Kürşad Sezer, and Olcay Ersel Canyurt. Review of artificial intelligence applications in engineering design perspective. 118:105697. ISSN 09521976. doi: 10.1016/j.engappai.2022.105697. URLhttps://linkinghub.elsevier.com/ retrieve/pii/S095219762200687X

    work page doi:10.1016/j.engappai.2022.105697 2022

  59. [61]

    URLhttp://arxiv.org/abs/2511.20730

    Nehal Afifi, Christoph Wittig, Lukas Paehler, An- dreas Lindenmann, Kai Wolter, Felix Leitenberger, Melih Dogru, Patric Grauberger, Tobias Düser, Al- bertAlbers,andSvenMatthiesen.Data-drivenmeth- odsandAIinengineeringdesign: Asystematicliter- ature review focusing on challenges and opportuni- ties. URLhttp://arxiv.org/abs/2511.20730

    arXiv

  60. [62]

    Deep learning in computational design synthesis: A comprehensive review

    Shubhendu Kumar Singh, Rahul Rai, Raj Pradip Khawale, Darshil Patel, Dustin Bielecki, Ryan Nguyen, Jun Wang, and Zhibo Zhang. Deep learning in computational design synthesis: A comprehensive review. 24(4):040801. ISSN 1530-9827, 1944-7078. doi: 10.1115/1.4064215. URLhttps://asmedigitalcollection. asme.org/computingengineering/ article/24/4/040801/1179741/...

    work page doi:10.1115/1.4064215 1944

  61. [63]

    AI-based com- puter aided engineering for automated prod- uct design - a first approach with a multi- view based classification

    Carmen Krahe, Maximilian Iberl, Alexander Jacob, and Gisela Lanza. AI-based com- puter aided engineering for automated prod- uct design - a first approach with a multi- view based classification. 86:104–109, . ISSN 22128271. doi: 10.1016/j.procir.2020.01

    work page doi:10.1016/j.procir.2020.01 2020

  62. [64]

    com/retrieve/pii/S2212827120300524

    URLhttps://linkinghub.elsevier. com/retrieve/pii/S2212827120300524

  63. [65]

    Deep genera- tive model for efficient 3d airfoil parameterization and generation

    Wei Chen and Arun Ramamurthy. Deep genera- tive model for efficient 3d airfoil parameterization and generation. InAIAA Scitech 2021 Forum. doi: 10.2514/6.2021-1690. URLhttp://arxiv.org/ abs/2101.02744

    work page doi:10.2514/6.2021-1690 2021

  64. [66]

    Data- driven methods for stress field predictions in random heterogeneous materials

    Enjamamul Hoq, Osama Aljarrah, Jun Li, Jing Bi, Alfa Heryudono, and Wenzhen Huang. Data- driven methods for stress field predictions in random heterogeneous materials. 123:106267. ISSN 09521976. doi: 10.1016/j.engappai.2023. 106267. URLhttps://linkinghub.elsevier. com/retrieve/pii/S0952197623004517

    work page doi:10.1016/j.engappai.2023 2023

  65. [67]

    Deep generative tread pattern design framework for efficient conceptual design

    Mingyu Lee, Youngseo Park, Hwisang Jo, Kibum Kim, Seungkyu Lee, and Ikjin Lee. Deep generative tread pattern design framework for efficient conceptual design. 144(7):071703. ISSN 1050-0472, 1528-9001. doi: 10.1115/1.4053469. URLhttps://asmedigitalcollection. asme.org/mechanicaldesign/ article/144/7/071703/1131073/ Deep-Generative-Tread-Pattern-Design-Fram...

    work page doi:10.1115/1.4053469

  66. [68]

    ConceptNet5.5: Anopenmultilingualgraphofgen- eral knowledge

    Robyn Speer, Joshua Chin, and Catherine Havasi. ConceptNet5.5: Anopenmultilingualgraphofgen- eral knowledge. URLhttp://arxiv.org/abs/ 1612.03975

    Pith/arXiv arXiv

  67. [69]

    Design representation for performance evaluation of 3d shapes in structure-aware generative design

    XingangLi,CharlesXie,andZhenghuiSha. Design representation for performance evaluation of 3d shapes in structure-aware generative design. 9: e27, . ISSN 2053-4701. doi: 10.1017/dsj.2023.25. URLhttps://www.cambridge.org/core/ journals/design-science/article/ design-representation-for-performance-evaluation-of-3d-shapes-in-structureaware-generative-design/ 0...

    work page doi:10.1017/dsj.2023.25 2053

  68. [70]

    Battaglia

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez- Gonzalez, and Peter W. Battaglia. Learning mesh- based simulation with graph networks. URLhttp: //arxiv.org/abs/2010.03409. 18

    arXiv 2010

  69. [71]

    Ibraheem Alawadhi, Prakash Kumar, Satchit Ram- nath,AbhishekIndupally,AllenSheldon,andJamiJ. Shah. A comparative study of novel design genera- tionforthinwalledstructuralcomponents: Topology optimization vs ga vs gen-ai. InProceedings of the ASME2026InternationalDesignEngineeringTech- nical Conferences and Computers and Information in Engineering Conferen...

    2026

  70. [72]

    Semi-supervised classification with graph convolutional networks

    ThomasN.KipfandMaxWelling. Semi-supervised classification with graph convolutional networks. URLhttp://arxiv.org/abs/1609.02907. 19

    Pith/arXiv arXiv