Inspiration
AI can generate impressive 3D designs in seconds, but a visually convincing model does not prove that a mechanical product will actually work. Engineers still need to understand how structures deform, where stresses concentrate, and how fluids behave around moving components.
We built ForceForge X to bridge this gap: transforming natural-language product ideas into real CAD geometry and connecting them to physics-based engineering simulations.
Our vision is to make computational engineering more accessible without sacrificing numerical accuracy, physical traceability, or transparency.
What it does
ForceForge X is an AI-assisted mechanical engineering platform that connects natural language, CAD generation, finite element analysis (FEA), and computational fluid dynamics (CFD).
Users can describe supported mechanical products in ordinary English or Chinese. The system extracts engineering parameters, generates actual 3D STEP geometry, and determines which simulation workflows can be executed.
Our key capabilities include:
1. Natural-Language Mechanical Modeling
Users can describe mechanical products without manually constructing every CAD feature.
Our current product catalog supports:
- L-shaped structural brackets
- Wall-mounted axial fan assemblies
- Hollow shafts and sleeves
- Rectangular mounting plates with optional through-holes
These models are generated through real parametric CAD operations rather than AI-generated pictures.
2. Physics-Based Structural Analysis
ForceForge X integrates a real three-dimensional finite element workflow for supported structural components.
The system evaluates:
- Structural displacement
- Von Mises stress
- Reaction forces and equilibrium
- Component mass and design constraints
Users can compare alternative geometries and inspect numerical evidence supporting engineering decisions.
3. Real Rotating-Flow CFD
Our most significant technical milestone was integrating a CAD-generated fan rotor into a genuine OpenFOAM transient simulation.
The workflow extracts the actual rotor geometry from a STEP model, creates the rotating-mesh configuration, and uses OpenFOAM's pimpleFoam solver with a sliding Arbitrary Mesh Interface (AMI).
The resulting velocity and pressure fields come directly from numerical simulation.
4. Continuous 3D Engineering Visualization
ForceForge X visualizes the actual source rotor geometry rotating continuously while displaying verified CFD flow data.
We distinguish between smooth geometric blade animation and the discrete physical time steps computed by OpenFOAM. This prevents visual interpolation from being mistaken for additional simulation results.
How we built it
Our technology stack includes:
- Python and Streamlit for the interactive engineering interface
- Pydantic for structured engineering requirements and parameter validation
- Gmsh and OpenCASCADE for CAD geometry, STEP export, and mesh generation
- NumPy and SciPy for numerical computations and structural analysis
- OpenFOAM for transient computational fluid dynamics
- Docker for isolated CFD execution
- VTK and Plotly for engineering visualization
- GitHub Actions for automated regression and integration testing
- ChatGPT/OpenAI for development assistance and optional AI-guided design exploration
Our architecture separates design interpretation, geometry generation, solver execution, and visualization so that engineering calculations remain independent of AI-generated descriptions.
Challenges we ran into
One of our biggest challenges was reliably executing real CFD simulations.
We encountered problems involving multiple simultaneous Docker jobs, stalled postprocessing, and compatibility issues when reading OpenFOAM's polyhedral VTK cells.
On an ARM64 Mac, a VTK decomposition operation caused a segmentation fault even after the transient CFD solver had completed its calculations.
We developed a recovery mechanism using native Python VTK support to read actual computed velocity and pressure fields without rerunning the expensive simulation.
We also improved solver job management, state persistence, and continuous three-dimensional browser animation.
Accomplishments that we're proud of
- Connected natural-language mechanical descriptions to actual parametric STEP generation.
- Implemented a genuine structural FEA workflow with numerical verification.
- Successfully simulated a STEP-derived custom fan rotor using OpenFOAM's transient rotating-mesh solver.
- Extracted authentic velocity and pressure results from completed CFD simulations.
- Implemented continuous 3D rotor animation alongside verified flow data.
- Validated core CAD and simulation functionality through automated Linux, macOS, and browser tests.
These results demonstrate an integrated engineering workflow rather than a purely visual CAD prototype.
What we learned
We learned that engineering AI requires more than generating geometry. A reliable system must preserve the connection between the original design, physical assumptions, mesh, numerical solver, and final results.
We also learned how to integrate different numerical tools while maintaining transparency about what has actually been computed.
What's next for ForceForge X
Our next development goals are to:
- Expand natural-language modeling to additional mechanical product families.
- Support more advanced CAD features and assemblies.
- Model realistic fan enclosures, wall mounts, and structural connections.
- Connect CFD pressure loads to structural FEA workflows.
- Improve meshing, convergence verification, and simulation performance.
- Develop a more accessible, unified engineering workspace.
Our long-term goal is an engineering assistant that can help users move from a product description to a traceable, simulation-supported design.
Technical limitations and development disclosure
ForceForge X is currently an engineering research prototype. The custom fan rotor CFD uses a reference OpenFOAM surrounding domain; realistic wall installation loads, guards, fasteners, and fully coupled CFD-FEA remain future work. Unsupported product types or analyses are explicitly identified.
The project builds on experimental engineering software developed before the hackathon. During the event, we expanded the natural-language modeling pipeline, integrated source-derived rotating CFD, improved simulation recovery, and added visualization and automated verification. ChatGPT/OpenAI assisted with implementation, debugging, and documentation; physical results originate from numerical solvers.