Tags: #AIAD #CAD #3DModeling #SysML #BPMN #DesignAssistance
AIAD – Use Case #001: Cross-Locking Mechanical Assembly (Japanese Kigumi Joinery)
Category:
Research > Use Case Feedback > AI for Engineering / Design Tools
Hashtags:
#AIAD #CAD #MechanicalDesign #3DModeling #PromptKit #SysML #BPMN #DesignAssistance #OpenAIEngineering
Label:
AI Quality & Reasoning – Engineering Domain
Context:
This use case is part of an ongoing evaluation of GPT’s ability to assist in real-world engineering design tasks. We refer to this domain as AIAD – Artificial Intelligence Aided Design, an extension of CAD with GenAI capabilities.
Use Case #001 – Cross-locking Kigumi-style Wood Joint
This test challenges the model to:
- Understand and describe a complex wooden joint (traditional Japanese kigumi)
- Reason about mechanical locking mechanisms and spatial interlocks
- Generate accurate 2D/3D representations from a single image or description
- Avoid hallucinations in geometry, such as confusing a true 3D X-shaped insert with a flat or T-shaped one
Test Results (ChatGPT 4.0 – April 2025):
Correct identification of the cross layout
Hallucinated the locking insert as flat or in double-T shape- 2D/3D visualizations generated were not mechanically accurate
Some corrections possible with iterative user feedback
Why This Matters:
This case highlights current limitations in:
- Spatial reasoning
- Interlocking geometry
- Physical constraint logic (friction fit, locking angles, insertion paths)
- Generalization of mechanical assemblies beyond visual appearance
It also bridges into broader ambitions: modeling complex engineered systems, not just storytelling or logic flows.
Suggestions:
- Improve support for 3D interlocking logic and mechanical constraints
- Add core CAD primitives and 2D/3D projection capabilities
- Enable symbolic or parametric representations (e.g., Fusion 360 / FreeCAD style)
- Extend logic support to SysML, BPMN, and system-level modeling
Tracking Tag:
AIAD_001_Kigumi_CrossLock
More test cases to follow: dovetail joints, mortise & tenon, parametric sketches, SysML/BPMN modeling…
@openai : This use case may be relevant for internal research / model improvement around 3D reasoning, CAD, and system modeling (AIAD initiative).
Attached images include prompt results, hallucinated outputs, user corrections, and final isometric drawing. All were generated during an iterative ChatGPT-4 session (April 2025). No manual reordering — let the AI reassemble the logic.
This post is part of a live AIAD evaluation – full trace available if needed.
Update with materials ( attached images used with my prompts and your outputs *.png
