transolver_drivaerml / explainability.md
ktangsali's picture
Add explainability subcard
9ef9adf verified
| Field | Response |
|:-------------------------------------------------------------------------------------------------------|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Intended Task/Domain: | Automotive Aerodynamics |
| Model Type: | Transformer with Physics-Attention (Transolver) |
| Intended Users: | Automotive engineers and researchers accelerating Computational Fluid Dynamics (CFD) predictions using AI. |
| Output: | Tensor (surface pressure and wall shear stress; volumetric velocity and pressure fields). |
| Describe how the model works: | Transolver introduces Physics-Attention, which decomposes the mesh domain into M learnable physical state slices by learning soft point-to-slice assignments. Multi-head attention is applied over the M slice tokens rather than all N mesh points (O(N) complexity), then results are broadcast back to mesh points. Each of 8 Transolver layers applies this Physics-Attention with residual connections and a feed-forward block, effectively approximating a learnable integral operator over the PDE domain. |
| Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of: | Not Applicable |
| Technical Limitations & Mitigation: | The number of slices M requires task-specific tuning; over-large M can degrade performance through over-fragmentation. The model may perform poorly for vehicle geometries significantly different from the training data or for flow conditions outside the training distribution. |
| Verified to have met prescribed NVIDIA quality standards: | Yes |
| Performance Metrics: | Surface and volume prediction relative L1 error; drag and lift coefficient R² |
| Potential Known Risks: | This model may inaccurately predict aerodynamic fields for vehicle designs outside the training distribution. |
| Licensing: | Use of this model is governed by the [NVIDIA Open Model Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/). |