Functional assemblies (e.g., gear systems, robotic arms, mechatronic devices) are typically complex due to geometry + kinematics + physics. AI models now aim to:

• Reconstruct or generate 3D geometry
• Infer or simulate functional/kinematic behavior
• Create assembly-aware CAD models

Key AI paradigms used:

• 3D Generative Models (e.g., 3D-GANs, Diffusion Models, VAEs)
• Neural Fields (e.g., NeRF, SDFs)
• Graph Neural Networks (GNNs) for relational reasoning

https://www.bricsys.com/en-eu/blog/how-artificial-intelligence-is-reshaping-generative-3d-modeling?srsltid=AfmBOop6ZawlC-wB-G3fPW5q9v_P5TK6bHnLh6csTyf_ez30FlSRbvGE

Some useful references (and associated bibliographies) as a starting point to help you answer all enumerated points:

[Computational mechanics & CAD]

• Herrmann et al., "Deep learning in computational mechanics: a review", 2024 -

  Article Deep learning in computational mechanics: a review

• Vergez et al., "Multi-part kinematic constraint prediction for automatic generation of CAD model assemblies using graph convolutional networks", 2025 -

  Article Multi-part kinematic constraint prediction for automatic gen...

• Bai et al., "Towards the future of physics- and data-guided AI frameworks in computational mechanics", 2025 -

  Article Towards the future of physics- and data-guided AI frameworks...

• Bai et al., "Regularized Diffusion Modeling for CAD Representation Generation", 2024 - https://openreview.net/forum?id=sCGIbhv4Yv
• Su et al., "ArtFormer: Controllable Generation of Diverse 3D Articulated Objects", 2024 -

  Preprint ArtFormer: Controllable Generation of Diverse 3D Articulated Objects

• Qiu et al., "Articulate AnyMesh: Open-Vocabulary 3D Articulated Objects Modeling", 2025 -

  Preprint Articulate AnyMesh: Open-Vocabulary 3D Articulated Objects Modeling

• Guo et al., "ArticulatedGS: Self-supervised Digital Twin Modeling of Articulated Objects using 3D Gaussian Splatting", 2025 -

  Preprint ArticulatedGS: Self-supervised Digital Twin Modeling of Arti...

• Luo et al., "PhysPart: Physically Plausible Part Completion for Interactable Objects", 2025 -

  Preprint PhysPart: Physically Plausible Part Completion for Interacta...

• Liu et al.,"Survey on Modeling of Human-made Articulated Objects", 2025 - https://onlinelibrary.wiley.com/doi/10.1111/cgf.70092?af=R
• Limoyo et al., "Generative Graphical Inverse Kinematics", 2025 -

  Preprint Generative Graphical Inverse Kinematics

[Mechatronic & articulated assemblies]

• Sihna et al., "Kinematics Support for Design and Simulation of Mechatronic Systems", 2000 -

  Article Kinematics Support for Design and Simulation of Mechatronic Systems

• A. Diaz-Calderon, "A Composable Simulation Environment to Support the Design of Mechatronic Systems", 2000 - https://www.cs.cmu.edu/~compsim/articles/adiaz-thesis.pdf
• Külz et al., "Timor Python: A Toolbox for Industrial Modular Robotics", 2023 -

  Preprint Timor Python: A Toolbox for Industrial Modular Robotics

• Shai et al., "Representing and analysing integrated engineering systems through combinatorial representations", 2004 -

  Article Representing and analysing integrated engineering systems th...

• Bordignon et al., "Model-based Kinematics Generation for Modular Mechatronic Toolkits", 2010 -

  Conference Paper Model-based Kinematics Generation for Modular Mechatronic Toolkits

• Whitman et al., "Modular Robot Design Synthesis with Deep Reinforcement Learning", 2020 - https://ojs.aaai.org/index.php/AAAI/article/view/6611/6465
• Zong et al., "Integrating kinematic and dynamic factors with generative design for high performance additive manufacturing structures", 2025 -

  Article Integrating kinematic and dynamic factors with generative de...

Dear Goutham Sunkara , I am deeply grateful for the information provided. Thank you!

Dear Dhafar Fakhry Hasan , I accessed the blog you sent and there is truly valuable information. Thank you!

Dear Stéphane Breton ,thank you very much for this very extensive list of references. It is helpful to me!

Hi, I hope you’re doing well.

AI generates 3D models of functional assemblies mainly through deep learning models trained on many 2D or 3D CAD examples. These models can reconstruct or create new assemblies from inputs like CAD files, sketches, or point clouds.

Having datasets with 3D assemblies that include functional or kinematic metadata helps improve accuracy by teaching the AI how parts interact and move.

There are also AI tools and frameworks that use generative models (like GANs or diffusion models) to produce these 3D designs.

Here are two articles that explain this well and might help you:

• https://www.mdpi.com/2076-3417/15/12/6681
• https://arxiv.org/html/2402.17695v1

I hope these help you out. Have a great day!